DRAFT - DO NOT CITE OR QUOTE
EPA/600/R-10/03 8C
January 2010
Agency/Interagency Review Draft
EPA's Response to "Health Risks from
Dioxin and Related Compounds Evaluation
of the EPA Reassessment" Published by the
National Research Council of the National
Academies
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
NOTICE
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
DISCLAIMER
This document is distributed solely for the purpose of pre-dissemination peer review
under applicable information quality guidelines. It has not been formally disseminated by EPA.
It does not represent and should not be construed to represent Agency determination or policy.
Mention of trade names or commercial products does not constitute endorsement or
recommendation for use.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 ii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS
LIST OF TABLES xviii
LIST OF FIGURES xxi
LIST OF ABBREVIATIONS AND ACRONYMS xxv
PREFACE xxix
AUTHORS, CONTRIBUTORS, AND REVIEWERS xxx
EXECUTIVE SUMMARY xxxiii
1. INTRODUCTION 1-1
1.1. SUMMARY OF KEY NAS (2006a) COMMENTS ON DOSE-RESPONSE
MODELING IN THE 2003 REASSESSMENT 1-2
1.2. EPA'S SCIENCE PLAN 1-4
1.3. OVERVIEW OF EPA'S RESPONSE TO NAS (2006a) "HEALTH RISKS
FROM DIOXIN AND RELATED COMPOUNDS: EVALUATION OF EPA's
2003 REASSESSMENT" 1-5
1.3.1. TCDD Literature Update 1-7
1.3.2. EPA's 2009 Workshop on TCDD Dose Response 1-7
1.3.3. Overall Organization of EPA's Response to NAS Recommendations 1-9
2. TRANSPARENCY AND CLARITY IN THE SELECTION OF KEY DATA SETS
I OR DOSE-RESPONSE ANALYSIS 2-1
2.1. SUMMARY OF NAS COMMENTS ON TRANSPARENCY AND CLARITY
IN THE SELECTION OF KEY DATA SETS FOR DOSE-RESPONSE
ANALYSIS 2-1
2.2. EPA's RESPONSE TO NAS COMMENTS ON TRANSPARENCY AND
CLARITY IN THE SELECTION OF KEY DATA SETS FOR DOSE-
RESPONSE ANALYSIS 2-2
2.3. STUDY INCLUSION CRITERIA FOR TCDD DOSE-RESPONSE ANALYSIS 2-4
2.3.1. Study Inclusion Criteria for TCDD Epidemiologic Studies 2-6
2.3.2. Study Inclusion Criteria for TCDD In Vivo Mammalian Bioassays 2-8
2.4. EVALUATION OF KEY STUDIES FOR TCDD DOSE RESPONSE 2-10
2.4.1. Evaluation of Epidemiological Cohorts for Dose-Response Assessment 2-10
2.4.1.1. Cancer 2-10
2.4.1.1.1. Cancer cohorts 2-11
2.4.1.1.1.1. The NIOSH cohort 2-11
2.4.1.1.1.1.1. Fingerhut et al., 1991 2-12
2.4.1.1.1.1.1.1. Study summary 2-12
2.4.1.1.1.1.1.2. Study evaluation 2-13
2.4.1.1.1.1.1.3. Suitability of data for TCDD dose-response
modeling 2-16
2.4.1.1.1.1.2. Steenland et al., 1999 2-17
2.4.1.1.1.1.2.1. Study summary 2-17
2.4.1.1.1.1.2.2. Study evaluation 2-20
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 iii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.1.1.1.1.2.3. Suitability of data for TCDD dose-response
modeling 2-21
2.4.1.1.1.1.3. Steenland et aL 2001 2-22
2.4.1.1.1.1.3.1. Study summary 2-22
2.4.1.1.1.1.3.2. Study evaluation 2-24
2.4.1.1.1.1.3.3. Suitability of data for TCDD dose-response
modeling 2-25
2.4.1.1.1.1.4. Cheng et aL 2006 2-26
2.4.1.1.1.1.4.1. Study summary 2-26
2.4.1.1.1.1.4.2. Study evaluation 2-29
2.4.1.1.1.1.4.3. Suitability of data for TCDD dose-response
modeling 2-29
2.4.1.1.1.1.5. Collins et aL 2009 2-30
2.4.1.1.1.1.5.1. Study summary 2-30
2.4.1.1.1.1.5.2. Study evaluation 2-31
2.4.1.1.1.1.5.3. Suitability of data for dose-response modeling 2-32
2.4.1.1.1.2. The BASF cohort 2-33
2.4.1.1.1.2.1. Thiess and Frentzel-Beyme, 1977 and Thiess etal.,
1982 2-33
2.4.1.1.1.2.1.1. Study summary 2-33
2.4.1.1.1.2.1.2. Study evaluation 2-34
2.4.1.1.1.2.1.3. Suitability of data for TCDD dose-response
modeling 2-34
2.4.1.1.1.2.2. Zoberetal., 1990 2-34
2.4.1.1.1.2.2.1. Study summary 2-34
2.4.1.1.1.2.2.2. Study evaluation 2-36
2.4.1.1.1.2.2.3. Suitability of data for TCDD dose-response
modeling 2-36
2.4.1.1.1.2.3. OttandZober, 1996 2-36
2.4.1.1.1.2.3.1. Study summary 2-36
2.4.1.1.1.2.3.2. Study evaluation 2-39
2.4.1.1.1.2.3.3. Suitability of data for TCDD dose-response
modeling 2-40
2.4.1.1.1.3. The Hamburg cohort 2-40
2.4.1.1.1.3.1. Manzetal., 1991 2-40
2.4.1.1.1.3.1.1. Study summary 2-40
2.4.1.1.1.3.1.2. Study evaluation 2-42
2.4.1.1.1.3.1.3. Suitability of data for TCDD dose-response
modeling 2-43
2.4.1.1.1.3.2. I'lesch-Janys et aL 1995 2-44
2.4.1.1.1.3.2.1. Study summary 2-44
2.4.1.1.1.3.2.2. Study evaluation 2-45
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 iv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.1.1.1.3.2.3. Suitability of data for TCDD dose-response
modeling 2-46
2.4.1.1.1.3.3. I'lesch-Janys et aL 1998 2-46
2.4.1.1.1.3.3.1. Study summary 2-46
2.4.1.1.1.3.3.2. Study evaluation 2-48
2.4.1.1.1.3.3.3. Suitability of data for TCDD dose-response
modeling 2-48
2.4.1.1.1.3.4. Becheretal., 1998 2-48
2.4.1.1.1.3.4.1. Study summary 2-48
2.4.1.1.1.3.4.2. Study evaluation 2-50
2.4.1.1.1.3.4.3. Suitability of data for TCDD dose-response
modeling 2-50
2.4.1.1.1.4. The Seveso cohort 2-51
2.4.1.1.1.4.1. Bertazzi et aL 2001 2-52
2.4.1.1.1.4.1.1. Study summary 2-52
2.4.1.1.1.4.1.2. Study evaluation 2-53
2.4.1.1.1.4.1.3. Suitability of data for TCDD dose-response
modeling 2-54
2.4.1.1.1.4.2. Pesatori et aL 2003 2-54
2.4.1.1.1.4.2.1. Study summary 2-54
2.4.1.1.1.4.2.2. Study evaluation 2-55
2.4.1.1.1.4.2.3. Suitability of data for TCDD dose-response
modeling 2-55
2.4.1.1.1.4.3. Consonni et al., 2008 2-55
2.4.1.1.1.4.3.1. Study summary 2-55
2.4.1.1.1.4.3.2. Study evaluation 2-56
2.4.1.1.1.4.3.3. Suitability of data for TCDD dose-response
modeling 2-56
2.4.1.1.1.4.4. Baccarelli et al., 2006 2-56
2.4.1.1.1.4.4.1. Study summary 2-56
2.4.1.1.1.4.4.2. Study evaluation 2-57
2.4.1.1.1.4.4.3. Suitability of data for TCDD dose-response
modeling 2-57
2.4.1.1.1.4.5. Warner et al.. 2002 2-57
2.4.1.1.1.4.5.1. Study summary 2-57
2.4.1.1.1.4.5.2. Study evaluation 2-58
2.4.1.1.1.4.5.3. Suitability of data for TCDD dose-response
modeling 2-59
2.4.1.1.1.5. Chapaevsk study 2-60
2.4.1.1.1.5.1. Revich et al., 2001 2-60
2.4.1.1.1.5.1.1. Study summary 2-60
2.4.1.1.1.5.1.2. Study evaluation 2-61
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 v DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.1.1.1.5.1.3. Suitability of data for TCDD dose-response
modeling 2-61
2.4.1.1.1.6. The Air Force Health ("Ranch Hands" cohort) study 2-61
2.4.1.1.1.6.1. Akhtaretal., 2004 2-62
2.4.1.1.1.6.1.1. Study summary 2-62
2.4.1.1.1.6.1.2. Study evaluation 2-64
2.4.1.1.1.6.1.3. Suitability of data for TCDD dose-response
modeling 2-65
2.4.1.1.1.6.2. Michalek and Pavuk, 2008 2-65
2.4.1.1.1.6.2.1. Study summary 2-65
2.4.1.1.1.6.2.2. Study evaluation 2-66
2.4.1.1.1.6.2.3. Suitability of data for TCDD dose-response
modeling 2-67
2.4.1.1.1.7. Other studies of potential relevance to dose-response
modeling 2-67
2.4.1.1.1.7.1. t' Mannetje et al., 2005—New Zealand herbicide
sprayers 2-67
2.4.1.1.1.7.1.1. Study summary 2-67
2.4.1.1.1.7.1.2. Study evaluation 2-69
2.4.1.1.1.7.1.3. Suitability of data for TCDD dose-response
modeling 2-69
2.4.1.1.1.7.2. McBride et al., 2009b—New Zealand herbicide
sprayers 2-70
2.4.1.1.1.7.2.1. Study summary 2-70
2.4.1.1.1.7.2.2. Study evaluation 2-71
2.4.1.1.1.7.2.3. Suitability of data for TCDD dose-response
modeling 2-71
2.4.1.1.1.7.3. McBride et al., 2009a—New Zealand herbicide
sprayers 2-71
2.4.1.1.1.7.3.1. Study summary 2-71
2.4.1.1.1.7.3.2. Study evaluation 2-74
2.4.1.1.1.7.3.3. Suitability of data for TCDD dose-response
modeling 2-74
2.4.1.1.1.7.4. Hooiveld et al., 1998—Netherlands workers 2-74
2.4.1.1.1.7.4.1. Study summary 2-74
2.4.1.1.1.7.4.2. Study evaluation 2-76
2.4.1.1.1.7.4.3. Suitability of data for TCDD dose-response
modeling 2-76
2.4.1.1.2. Key characteristics of epidemiologic cancer studies 2-77
2.4.1.1.3. Feasibility of TCDD cancer dose-response modeling—summary
discussion by cohort 2-77
2.4.1.1.3.1. Using the NIOSH cohort in dose-response modeling 2-77
2.4.1.1.3.2. Using the BASF cohort in dose-response modeling 2-79
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 vi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.1.1.3.3. Using the Hamburg cohort in dose-response modeling 2-80
2.4.1.1.3.4. Using the Seveso cohort in dose-response modeling 2-81
2.4.1.1.3.5. Using the Chapaevsk related data in dose-response
modeling 2-82
2.4.1.1.3.6. Using the Ranch Hands cohort in dose-response modeling 2-82
2.4.1.1.4. Discussion of general issues related to dose-response modeling 2-82
2.4.1.1.4.1. Ascertainment of exposures 2-82
2.4.1.1.4.2. Latency intervals 2-83
2.4.1.1.4.3. Use of the SMR metric 2-83
2.4.1.1.4.4. All cancers versus site-specific 2-85
2.4.1.1.4.5. Summary of epidemiologic cancer study evaluations for
dose-response modeling 2-86
2.4.1.2. Noncancer 2-86
2.4.1.2.1. Noncancer cohorts 2-87
2.4.1.2.1.1. The NIOSH cohort 2-87
2.4.1.2.1.1.1. Steenland et aL 1999 2-87
2.4.1.2.1.1.1.1. Study summary 2-87
2.4.1.2.1.1.1.2. Study evaluation 2-87
2.4.1.2.1.1.1.3. Suitability of data for TCDD dose-response
modeling 2-88
2.4.1.2.1.1.2. Collins et aL 2009 2-88
2.4.1.2.1.1.2.1. Study summary 2-88
2.4.1.2.1.1.2.2. Study evaluation 2-89
2.4.1.2.1.1.2.3. Suitability of data for TCDD dose-response
modeling 2-89
2.4.1.2.1.2. The BASF cohort 2-89
2.4.1.2.1.2.1. Ott and Zober. 1996 2-89
2.4.1.2.1.2.1.1. Study summary 2-89
2.4.1.2.1.2.1.2. Study evaluation 2-90
2.4.1.2.1.2.1.3. Suitability of data for TCDD dose-response
modeling 2-90
2.4.1.2.1.3. The Hamburg cohort 2-91
2.4.1.2.1.3.1. I'lesch-Janys et aL 1995 2-91
2.4.1.2.1.3.1.1. Study summary 2-91
2.4.1.2.1.3.1.2. Study evaluation 2-91
2.4.1.2.1.3.1.3. Suitability of data for TCDD dose-response
modeling 2-92
2.4.1.2.1.4. The Seveso Women's Health Study (SWHS) 2-92
2.4.1.2.1.4.1. Eskenazi et al., 2002a—Menstrual cycle characteristics....2-93
2.4.1.2.1.4.1.1. Study summary 2-93
2.4.1.2.1.4.1.2. Study evaluation 2-94
2.4.1.2.1.4.1.3. Suitability of data for TCDD dose-response
modeling 2-95
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 vii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.1.2.1.4.2. Eskenazi et al., 2002b—Endometriosis 2-95
2.4.1.2.1.4.2.1. Study summary 2-95
2.4.1.2.1.4.2.2. Study evaluation 2-96
2.4.1.2.1.4.2.3. Suitability of data for TCDD dose-response
modeling 2-96
2.4.1.2.1.4.3. Eskenazi et al., 2003—Adverse birth outcomes 2-96
2.4.1.2.1.4.3.1. Study summary 2-96
2.4.1.2.1.4.3.2. Study evaluation 2-97
2.4.1.2.1.4.3.3. Suitability of data for TCDD dose-response
modeling 2-98
2.4.1.2.1.4.4. Warner et al., 2004—Age at menarche 2-98
2.4.1.2.1.4.4.1. Study summary 2-98
2.4.1.2.1.4.4.2. Study evaluation 2-99
2.4.1.2.1.4.4.3. Suitability of data for TCDD dose-response
modeling 2-99
2.4.1.2.1.4.5. Eskenazi etal., 2005—Age at menopause 2-99
2.4.1.2.1.4.5.1. Study summary 2-99
2.4.1.2.1.4.5.2. Study evaluation 2-101
2.4.1.2.1.4.5.3. Suitability of data for TCDD dose-response
modeling 2-101
2.4.1.2.1.4.6. Warner etal., 2007—Ovarian function 2-101
2.4.1.2.1.4.6.1. Study summary 2-101
2.4.1.2.1.4.6.2. Study evaluation 2-102
2.4.1.2.1.4.6.3. Suitability of data for TCDD dose-response
modeling 2-102
2.4.1.2.1.4.7. Eskenazi et al., 2007—Uterine leiomyoma 2-103
2.4.1.2.1.4.7.1. Study summary 2-103
2.4.1.2.1.4.7.2. Study evaluation 2-103
2.4.1.2.1.4.7.3. Suitability of data for TCDD dose-response
modeling 2-103
2.4.1.2.1.5. Other Seveso noncancer studies 2-104
2.4.1.2.1.5.1. Mocarelli etal., 2008—Semen quality 2-104
2.4.1.2.1.5.1.1. Study summary 2-104
2.4.1.2.1.5.1.2. Study evaluation 2-105
2.4.1.2.1.5.1.3. Suitability of data for TCDD dose-response
modeling 2-105
2.4.1.2.1.5.2. Mocarelli etal., 1996, 2000—Sex ratio 2-105
2.4.1.2.1.5.2.1. Study summary 2-105
2.4.1.2.1.5.2.2. Study evaluation 2-107
2.4.1.2.1.5.2.3. Suitability of data for TCDD dose-response
modeling 2-108
2.4.1.2.1.5.3. Baccarelli et al., 2008—Neonatal thyroid function 2-108
2.4.1.2.1.5.3.1. Study summary 2-108
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 viii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.1.2.1.5.3.2. Study evaluation 2-110
2.4.1.2.1.5.3.3. Suitability of data for TCDD dose-response
modeling 2-110
2.4.1.2.1.5.4. Alaluusua et al., 2004—Oral hygiene 2-111
2.4.1.2.1.5.4.1. Study summary 2-111
2.4.1.2.1.5.4.2. Study evaluation 2-112
2.4.1.2.1.5.4.3. Suitability of data for TCDD dose-response
modeling 2-112
2.4.1.2.1.5.5. Bertazzi et al., 1989; Consonni et al., 2008—Mortality
outcomes 2-113
2.4.1.2.1.5.5.1. Study summary 2-113
2.4.1.2.1.5.5.2. Study evaluation 2-114
2.4.1.2.1.5.5.3. Suitability of data for TCDD dose-response
modeling 2-115
2.4.1.2.1.5.6. Baccarelli et al., 2005—Chloracne 2-115
2.4.1.2.1.5.6.1. Study summary 2-115
2.4.1.2.1.5.6.2. Study evaluation 2-115
2.4.1.2.1.5.6.3. Suitability of data for TCDD dose-response
modeling 2-116
2.4.1.2.1.5.7. Baccarelli et al., 2002, 2004—Immunologic effects 2-116
2.4.1.2.1.5.7.1. Study summary 2-116
2.4.1.2.1.5.7.2. Study evaluation 2-117
2.4.1.2.1.5.7.3. Suitability of data for TCDD dose-response
modeling 2-117
2.4.1.2.1.6. The Chapaevsk study 2-118
2.4.1.2.1.6.1. Revich et al. (2001)—Mortality and reproductive
health 2-118
2.4.1.2.1.6.1.1. Study summary 2-118
2.4.1.2.1.6.1.2. Study evaluation 2-119
2.4.1.2.1.6.1.3. Suitability of data for TCDD dose-response
modeling 2-119
2.4.1.2.1.7. The Air Force Health ("Ranch Hands" cohort) study 2-119
2.4.1.2.1.7.1. Michalek and Pavuk (2008)—Diabetes 2-119
2.4.1.2.1.7.1.1. Study summary 2-119
2.4.1.2.1.7.1.2. Study evaluation 2-120
2.4.1.2.1.7.1.3. Suitability of data for TCDD dose-response
modeling 2-120
2.4.1.2.1.8. Other noncancer studies of TCDD 2-120
2.4.1.2.1.8.1. McBride et al., 2009a—Noncancer mortality 2-120
2.4.1.2.1.8.1.1. Study summary 2-120
2.4.1.2.1.8.1.2. Study evaluation 2-121
2.4.1.2.1.8.1.3. Suitability of data for dose-response analysis 2-122
2.4.1.2.1.8.2. McBride etal., 2009b—Noncancer mortality 2-122
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 ix DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.1.2.1.8.2.1. Study summary 2-122
2.4.1.2.1.8.2.2. Study evaluation 2-123
2.4.1.2.1.8.2.3. Suitability of data for TCDD dose-response
modeling 2-123
2.4.1.2.1.8.3. Ryan et al., 2002—Sex ratio 2-123
2.4.1.2.1.8.3.1. Study summary 2-123
2.4.1.2.1.8.3.2. Study evaluation 2-124
2.4.1.2.1.8.3.3. Suitability of data for TCDD dose-response
modeling 2-125
2.4.1.2.2. Feasibility of dose-response modeling for noncancer 2-125
2.4.1.2.3. Summary of epidemiologic noncancer study evaluations for
dose-response modeling 2-126
2.4.2. Summary of Animal Bioassay Studies 2-126
2.4.2.1. Reproductive Studies 2-127
2.4.2.1.1. Bowman et al., 1989a, b (and related Schantz and Bowman,
1989; Schantz et al., 1986) 2-127
2.4.2.1.2. Hochstein et al., 2001 2-128
2.4.2.1.3. Ikeda et al., 2005 2-130
2.4.2.1.4. Ishihara et al., 2007 2-131
2.4.2.1.5. Latchoumycandane and Mathur, 2002 (and related:
Latchoumycandane et al., 2002a, b, 2003) 2-132
2.4.2.1.6. Murray et al, 1979 2-133
2.4.2.1.7. Rier et al., 1993, 1995 2-134
2.4.2.1.8. Shi et al., 2007 2-136
2.4.2.1.9. Yang et al., 2000 2-136
2.4.2.2. Developmental Studies 2-138
2.4.2.2.1. Amin et al., 2000 2-138
2.4.2.2.2. Bell et al., 2007a 2-139
2.4.2.2.3. Franczak et al., 2006 2-141
2.4.2.2.4. Hojo et al., 2002 (and related: Zareba et al., 2002) 2-142
2.4.2.2.5. Kattainen et al., 2001 2-143
2.4.2.2.6. Keller et al., 2007, 2008a, b 2-144
2.4.2.2.7. Kuchiiwa et al., 2002 2-148
2.4.2.2.8. I.i et al.. 2006 2-149
2.4.2.2.9. Markowski et al., 2001 2-150
2.4.2.2.10. Miettinen et al., 2006 2-150
2.4.2.2.11. Nohara et al., 2000 2-151
2.4.2.2.12. Ohsako et al., 2001 2-152
2.4.2.2.13. Schantz et al., 1996 2-153
2.4.2.2.14. Seo et al., 1995 2-154
2.4.2.2.15. Simanainen et al, 2004 2-155
2.4.2.2.16. Sugita-Konishi et al., 2003 2-156
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 x DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.2.3. Acute Studies 2-157
2.4.2.3.1. Burleson et al., 1996 2-157
2.4.2.3.2. Crofton et al., 2005 2-158
2.4.2.3.3. Kitchin and Woods, 1979 2-159
2.4.2.3.4. I.i et al.. 1997 2-160
2.4.2.3.5. Lucier et al., 1986 2-160
2.4.2.3.6. Nohara et al., 2002 2-161
2.4.2.3.7. Simanainen et al., 2003 2-161
2.4.2.3.8. Simanainen et al., 2002 2-162
2.4.2.3.9. Smialowicz et al., 2004 2-163
2.4.2.3.10. Yanden I Ieuvel et al., 1994 2-163
2.4.2.4. Subchronic Studies 2-165
2.4.2.4.1. Chu et al.. 2001 2-165
2.4.2.4.2. Chu et al., 2007 2-165
2.4.2.4.3. DeCaprio et al., 1986 2-167
2.4.2.4.4. Devito et al., 1994 2-168
2.4.2.4.5. Fattore et al., 2000 2-168
2.4.2.4.6. Fox et al.. 1993 2-170
2.4.2.4.7. Hassoun et al., 1998 2-170
2.4.2.4.8. Hassoun et al., 2000 2-171
2.4.2.4.9. Hassoun et al., 2003 2-172
2.4.2.4.10. Kociba et al., 1976 2-173
2.4.2.4.11. VIally and Chipman, 2002 2-174
2.4.2.4.12. Slezak et al., 2000 2-175
2.4.2.4.13. Smialowicz et al., 2008 2-176
2.4.2.4.14. Van Birgelen et al. 1995a, b 2-177
2.4.2.4.15. Vos et al., 1973 2-178
2.4.2.4.16. White et al.. 1986 2-179
2.4.2.5. Chronic Studies (Noncancer Endpoints) 2-180
2.4.2.5.1. Cantoni et al., 1981 2-180
2.4.2.5.2. Croutch et al., 2005 2-180
2.4.2.5.3. Hassoun et al., 2002 2-182
2.4.2.5.4. Kociba et al.. 1978 2-183
2.4.2.5.5. Maronpot et al., 1993 2-184
2.4.2.5.6. National Toxicology Program, 1982 2-185
2.4.2.5.7. National Toxicology Program, 2006 2-185
2.4.2.5.8. Rier et al., 2001a, b 2-187
2.4.2.5.9. Sewall et al., 1993 2-189
2.4.2.5.10. Sewall et al.. 1995 2-190
2.4.2.5.11. Toth et al.. 1979 2-192
2.4.2.6. Chronic Studies (Cancer Endpoints) 2-193
2.4.2.6.1. Kociba et al.. 1978 2-193
2.4.2.6.2. l oth et al.. 1979 2-195
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
2.4.2.6.3. NTP, 1982 2-195
2.4.2.6.4. NTP, 2006 2-196
2.4.3. Summary of Key Data Set Selection for TCDD Dose-Response Modeling ... 2-198
3. THE USE OF TOXICOKINETICS IN THE DOSE-RESPONSE MODELING FOR
CANCER AND NONCANCER ENDPOINTS 3-1
3.1. SUMMARY OF NAS COMMENTS ON THE USE OF TOXICOKINETICS IN
DOSE-RESPONSE MODELING APPROACHES FOR TCDD 3-1
3.2. OVERVIEW OF EPA'S RESPONSE TO THE NAS COMMENTS ON THE USE
OF TOXICOKENTICS IN DOSE-RESPONSE MODELING APPROACHES
FOR TCDD 3-3
3.3. PHARMACOKINETICS (PK) AND PK MODELING 3-4
3.3.1. PK Data and Models in TCDD Dose-Response Modeling: Overview and
Scope 3-4
3.3.2. PK of TCDD in Animals and Humans 3-6
3.3.2.1. Absorption and Bioavailability 3-6
3.3.2.2. Distribution 3-6
3.3.2.3. Metabolism and Protein Binding 3-9
3.3.2.4. Elimination 3-11
3.3.2.5. Interspecies Differences and Similarities 3-11
3.3.3. PK of TCDD in Humans: Interindividual Variability 3-12
3.3.3.1. Life Stage and Gender 3-12
3.3.3.1.1. Prenatal period 3-14
3.3.3.1.2. Infancy and childhood 3-15
3.3.3.1.3. Adulthood and old age 3-16
3.3.3.2. Physiological States: Pregnancy and Lactation 3-16
3.3.3.3. Lifestyle and Habits 3-17
3.3.3.4. Genetic Traits and Polymorphism 3-17
3.3.4. Dose Metrics and Pharmacokinetic Models for TCDD 3-18
3.3.4.1. Alternative Dose Metrics for Dose-Response Modeling 3-18
3.3.4.2. First-Order Kinetic Modeling 3-21
3.3.4.3. Biologically-Based Kinetic Models 3-25
3.3.4.3.1. CADM model 3-26
3.3.4.3.1.1. Model structure 3-26
3.3.4.3.1.2. Mathematical representation 3-27
3.3.4.3.1.3. Parameter estimation 3-28
3.3.4.3.1.4. Model performance and degree of evaluation 3-29
3.3.4.3.1.5. Confidence in CADM model predictions of dose metrics 3-30
3.3.4.3.2. PBPK model 3-31
3.3.4.3.2.1. Model structure 3-31
3.3.4.3.2.2. Mathematical representation 3-32
3.3.4.3.2.3. Parameter estimation 3-36
3.3.4.3.2.4. Model performance and degree of evaluation 3-37
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
3.3.4.3.2.5. Confidence in PBPK model predictions of dose metrics 3-40
3.3.4.4. Applicability of PK Models to Derive Dose Metrics for Dose-Response
Modeling of TCDD: Confidence and Limitations 3-42
3.3.4.5. Recommended Dose Metrics for Key Studies 3-44
3.3.5. Uncertainty in Dose Estimates 3-46
3.3.5.1. Sources of Uncertainty in Dose Metric Predictions 3-46
3.3.5.1.1. Limitations of available PK data 3-46
3.3.5.1.1.1. Animal data 3-46
3.3.5.1.1.2. Human data 3-46
3.3.5.1.2. Uncertainties associated with model specification 3-46
3.3.5.1.3. Impact of human interindividual variability 3-48
3.3.5.2. Potential Magnitude and Sources of Uncertainty in Dose Metrics 3-48
3.3.5.2.1. Magnitude of uncertainty 3-48
3.3.6. Use of the Emond PBPK Models for Dose Extrapolation from Rodents to
Humans 3-50
4. CHRONIC ORAL REFERENCE DOSE 4-1
4.1. NAS COMMENTS AND EPA'S RESPONSE ON IDENTIFYING
NONCANCER EFFECTS OBSERVED AT LOWEST DOSES 4-1
4.2. NONCANCER DOSE-RESPONSE ASSESSMENT OF TCDD 4-5
4.2.1.1. Determination of Toxicologically Relevant Endpoints 4-6
4.2.2. Use of Toxicokinetic Modeling for TCDD Dose-Response Assessment 4-7
4.2.3. Noncancer Dose-Response Assessment of Epidemiological Data 4-8
4.2.3.1. Baccarelli et al. (2008) 4-9
4.2.3.2. Mocarelli et al. (2008) 4-9
4.2.3.3. Alaluusua et al. (2004) 4-10
4.2.3.4. Eskenazi et al. (2002) 4-11
4.2.4. Noncancer Dose-Response Assessment of Animal Bioassay Data 4-12
4.2.4.1. Use of Kinetic Modeling for Animal Bioassay Data 4-13
4.2.4.2. Benchmark Dose Modeling of the Animal Bioassay Data 4-14
4.2.4.3. POD Candidates from Animal Bioassays Based on HED 4-15
4.3. RfD DERIVATION 4-16
4.3.1. Toxicological Endpoints 4-18
4.3.2. Exposure Protocols of Candidate PODs 4-19
4.3.3. Uncertainty Factors (UFs) 4-20
4.3.4. Human Studies 4-20
4.3.5. Derivation of the RfD 4-23
4.4. UNCERTAINTY IN THE RfD 4-23
5. CANCER ASSESSMENT 5-1
5.1. QUALITATIVE WEIGHT-OF-EVIDENCE CARCINOGEN
CLASSIFICATION FOR 2,3,7,8-TETRACHLORODIBENZO-p-DIOXIN
(TCDD) 5-1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xiii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
5.1.1. Summary of National Academy of Sciences (NAS) Comments on the
Qualitative Weight-of-Evidence Carcinogen Classification for
2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) 5-1
5.1.2. EPA's Response to the NAS Comments on the Qualitative Weight-of-
Evidence Carcinogen Classification for TCDD 5-2
5.1.2.1. Summary Evaluation of Epidemiologic Evidence of TCDD and Cancer... 5-3
5.1.2.1.1. Evidence for causality 5-5
5.1.2.2. Summary of Evidence for TCDD Carcinogenicity in Experimental
Animals 5-9
5.1.2.3. TCDD Mode of Action 5-10
5.1.2.3.1. The aryl hydrocarbon receptor (AhR) 5-10
5.1.2.3.1.1. Other AhR considerations 5-12
5.1.2.3.2. TCDD as a tumor promoter 5-14
5.1.2.3.3. Hypothesized modes of action of TCDD in rodents 5-15
5.1.2.3.3.1. Liver tumors 5-15
5.1.2.3.3.2. Lung tumors 5-16
5.1.2.3.3.3. Thyroid tumors 5-17
5.1.2.3.4. Summary of TCDD mode of action in rodents 5-18
5.1.3. Summary of the Qualitative Weight of Evidence Classification for TCDD 5-19
5.2. QUANTITATIVE CANCER ASSESSMENT 5-19
5.2.1. Summary of NAS Comments on Cancer Dose-Response Modeling 5-19
5.2.1.1. Choice of Response Level and Characterization of the Statistical
Confidence Around Low Dose Model Predictions 5-19
5.2.1.2. Model Forms for Predicting Cancer Risks Below the Point of
Departure (POD) 5-20
5.2.2. Overview of EPA Response to NAS Comments on Cancer Dose-
Response Modeling 5-22
5.2.3. Updated Cancer Dose-Response Modeling for Derivation of Oral Slope
Factor 5-22
5.2.3.1. Dose-Response Modeling Based on Epidemiologic Cohort Data 5-23
5.2.3.1.1. Evaluation of Epidemiologic Studies in the 2003 Reassessment
for OSF Derivation 5-23
5.2.3.1.1.1. Steenland et al. (2001) 5-24
5.2.3.1.1.2. Becheretal. (1998) 5-25
5.2.3.1.1.3. Ott and Zober (1996) 5-26
5.2.3.1.2. Evaluation of Epidemiologic Studies Published Since the 2003
Reassessment for OSF Derivation 5-27
5.2.3.1.2.1. Cheng et al. (2006) 5-27
5.2.3.1.2.2. Warner et al. (2002) 5-30
5.2.3.2. Dose-Response Modeling Based on Animal Bioassay Data 5-31
5.2.3.2.1. Selection of key data sets 5-32
5.2.3.2.2. Dose adjustment and extrapolation methods for selected data sets... 5-33
5.2.3.2.2.1. Dose metric estimation for dose-response modeling 5-33
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xiv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
5.2.3.2.2.2. Calculation of human equivalent doses (HEDs) 5-34
5.2.3.2.3. Dose-response modeling approaches for rodent bioassays 5-35
5.2.3.2.3.1. Modeling of individual tumors 5-35
5.2.3.2.3.2. Multiple tumor (Bayesian) models 5-35
5.2.3.2.4. Results of dose-response modeling for rodent bioassays 5-38
5.2.3.2.4.1. Individual tumor models 5-38
5.2.3.2.4.2. Multiple tumor (Bayesian) models 5-39
5.2.3.2.5. Summary evaluation of slope factor estimates from rodent
bioassays 5-39
5.2.3.2.6. Qualitative uncertainties in slope factor estimates from rodent
bioassays 5-40
5.2.3.2.6.1. Quality of studies relied upon for determining POD 5-40
5.2.3.2.6.2. Interpretation of results from studies relied upon for
determining POD 5-41
5.2.3.2.6.3. Consistency of results across chronic rodent bioassays 5-41
5.2.3.2.6.4. Human relevance of rodent tumor data 5-42
5.2.3.2.6.5. Relevance of rodent exposure scenario 5-42
5.2.3.2.6.6. Impact of background TCDD exposures 5-42
5.2.3.2.6.7. Choice of endpoint for POD derivation 5-43
5.2.3.2.6.8. Choice of animal-to-human extrapolation method 5-43
5.2.3.2.6.9. Choice of model for POD and model uncertainty for POD
derivation 5-43
5.2.3.2.6.10. Statistical uncertainty in model fits 5-44
5.2.3.2.6.11. Choice of risk level for POD derivation 5-44
5.2.3.3. EPA's Response to the NAS Comments on Choice of Response Level
and Characterization of the Statistical Confidence Around Low Dose
Model Predictions 5-45
5.2.3.4. EPA's Response to the NAS Comments on Model Forms for Predicting
Cancer Risks Below the POD 5-46
5.2.3.4.1. Choice of extrapolation approach 5-46
5.2.3.4.1.1. TCDD and receptor theory 5-46
5.2.3.4.1.2. Low-dose extrapolation: threshold or no threshold? 5-48
5.2.3.4.1.3. Extrapolation method 5-55
5.2.3.4.1.4. Consideration of nonlinear methods 5-57
5.2.3.4.1.4.1. Illustrative RfDs based on tumorigenesis in experimental
animals 5-58
5.2.3.4.1.4.2. Illustrative RfDs based on hypothesized key events in
TCDD's MO As for liver and lung tumors 5-59
5.2.3.4.1.4.2.1. Liver tumors 5-60
5.2.3.4.1.4.2.2. Lung tumors 5-61
5.2.3.4.1.4.2.3. Limitations of illustrative RfDs based on
hypothesized key events in TCDD's MOAs for
liver and lung tumors 5-62
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
5.2.3.4.1.4.3. Effect of dose metric on linearity of response 5-64
5.3. DERIVATION OF THE TCDD ORAL SLOPE FACTOR AND CANCER RISK
ESTIMATES 5-66
5.3.1. Uncertainty in Estimation of Oral Slope Factors From Human Studies 5-68
5.3.1.1. Uncertainty in Exposure Estimation 5-68
5.3.1.2. Uncertainty in Shape of the Dose-Response Curve 5-72
5.3.1.3. Uncertainty in Defining the Reference Population 5-73
5.3.1.4. Uncertainty in Cancer Risk Estimates 5-73
5.3.2. Other Sources of Uncertainty in Risk Estimates From the Epidemiological
Studies 5-74
5.3.3. Approaches to Combining Estimates From Different Epidemiologic Studies.. 5-76
5.3.3.1. The Crump et al. (2003) Meta-analysis 5-76
5.3.3.2. EPA's Decision Not to Conduct a Meta-analysis 5-78
6. FEASIBILITY OF QUANTITATIVE UNCERTAINTY ANALYSIS FROM NAS
EVALUATION OF THE 2003 REASSESSMENT 6-1
6.1. INTRODUCTION 6-1
6.1.1. Historical Context for Quantitative Uncertainty Analysis 6-1
6.1.2. Definition of Terms 6-3
6.1.3. Basic Requirements of a Quantitative Uncertainty Analysis 6-5
6.1.3.1. Quantitative Model 6-6
6.1.3.2. Marginal Distributions over Model Parameter 6-6
6.1.3.3. Dependence Between Parameter Uncertainties: Aleatoric and Epistemic
(Uncertainty and Variability) 6-6
6.1.3.4. Model Uncertainty 6-7
6.1.3.5. Sampling Method 6-8
6.1.3.6. Method for Extracting and Communicating Results 6-8
6.2. EPA APPROACHES FOR ORAL CANCER AND NONCANCER
ASSESSMENT 6-9
6.3. HIGHLIGHTS OF NAS REVIEW COMMENTS ON UNCERTAINTY
QUANTIFICATION FOR THE 2003 REASSESSMENT 6-11
6.4. FEASIBILITY OF CONDUCTING A QUANTITATIVE UNCERTAINTY
ANALYSIS FOR TCDD 6-14
6.4.1. Feasibility of Conducting a Quantitative Uncertainty Analysis under the
RfD Methodology 6-14
6.4.1.1. Feasibility of Conducting a Quantitative Uncertainty Analysis for the
Point of Departure 6-15
6.4.1.2. Feasibility of Conducting a Quantitative Uncertainty Analysis with
Uncertainty Factors 6-18
6.4.1.3. Conclusion on Feasibility of Quantitative Uncertainty Analysis with
the RfD Approach 6-20
6.4.2. Feasibility of Conducting a Quantitative Uncertainty Analysis for TCDD
Under the Dose-Response Methodology 6-20
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xvi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
6.4.2.1. Feasibility of Quantitatively Characterizing the Uncertainties
Encountered when Determining Appropriate Types of Studies
(Epidemiological, Animal, Both, and Other) 6-22
6.4.2.2. Uncertainty in TCDD Exposure/Dose in Epidemiological Studies 6-23
6.4.2.3. Uncertainty in Toxicity Equivalence (TEQ) Exposures in
Epidemiological Studies 6-26
6.4.2.4. Uncertainty in Background Feed Exposures in Bioassays 6-27
6.4.2.5. Feasibility of Quantifying the Uncertainties Encountered When
Choosing Specific Studies and Subsets of Data 6-28
6.4.2.6. Feasibility of Quantifying the Uncertainties Encountered when
Choosing Specific Endpoints for Dose-Response Modeling 6-29
6.4.2.7. Feasibility of Quantifying the Uncertainties Encountered when
Choosing a Specific Dose Metric (Trade-Off Between Confidence in
Estimated Dose and Relevance of MO A) 6-30
6.4.2.8. Feasibility of Quantifying the Uncertainties Encountered When
Choosing Model Type and Form 6-31
6.4.2.9. Threshold MOA for Cancer 6-33
6.4.2.10. Feasibility of Quantifying the Uncertainties Encountered when
Selecting the BMR 6-34
6.5. CONCLUSIONS 6-34
6.5.1. Summary of NAS Suggestions and Responses 6-34
6.5.2. How Forward? Beyond RfDs and Cancer Slope Factors to Development
of Predictive Human Dose-Response Functions 6-38
REFERENCES R-l
APPENDIX A: DIOXIN WORKSHOP A-l
APPENDIX B: EVALUATION OF CANCER AND NONCANCER
EPIDEMIOLOGICAL STUDIES FOR INCLUSION IN
TCDD DOSE-RESPONSE ASSESSMENT B-l
APPENDIX C: KINETIC MODELING C-l
APPENDIX D: EPIDEMIOLOGICAL KINETIC MODELING D-l
APPENDIX E: NONCANCER BENCHMARK DOSE MODELING E-l
APPENDIX F: CANCER BENCHMARK DOSE MODELING F-l
APPENDIX G: ENDPOINTS EXCLUDED FROM REFERENCE DOSE
DERIVATION BASED ON TOXICOLOGICAL RELEVANCE G-l
APPENDIX H: CANCER PRECURSOR BENCHMARK DOSE MODELING H-l
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xvii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES
2-1. Summary of epidemiologic cancer studies (key characteristics) 2-199
2-2. Epidemiology cancer study selection considerations and criteria 2-201
2-3. Epidemiology noncancer study selection considerations and criteria 2-205
2-4. Epidemiology studies selected for TCDD cancer dose-response modeling 2-209
2-5. Epidemiology studies selected for TCDD noncancer dose-response modeling 2-213
2-6. Animal bioassays selected for cancer dose-response modeling 2-217
2-7. Animal bioassay studies selected for noncancer dose-response modeling 2-218
3-1. Partition coefficients, tissue volumes, and volume of distribution for TCDD in
humans 3-55
3-2. Blood flows, permeability factors and resulting half lives (VA) for perfusion losses
for humans as represented by the TCDD PBPK model of Emond et al. (2005,
2006) 3-55
3-3. Toxicokinetic conversion factors for calculating human equivalent doses from
rodent bioassays 3-56
3-4. Equations used in the CADM PBPK model* 3-57
3-5. Parameters of the CADM model 3-58
3-6. Confidence in the CADM model simulations of TCDD dose metrics 3-59
3-7. Equations used in the TCDD PBPK model of Emond et al. (2006) 3-60
3-8. Parameters of the PBPK model for TCDD 3-62
3-9. Regression analysis results for the relationship between logio serum TCDD at the
midpoint of observations and the logio of the rate constant for decline of TCDD
levels using Ranch Hand data 3-65
3-10. Confidence in the PBPK model simulations of TCDD dose metrics 3-65
3-11. Overall confidence associated with alternative dose metrics for cancer and
noncancer dose-response modeling for TCDD using rat PBPK model 3-66
3-12. Overall confidence associated with alternative dose metrics for cancer and
noncancer dose-response modeling for TCDD using mouse PBPK model 3-66
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xviii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
3-13. Contributors to the overall uncertainty in the selection and use of dose metrics in
the dose-response modeling of TCDD based on rat and human PBPK models 3-66
3-14. Contributors to the overall uncertainty in the selection and use of dose metrics in
the dose-response modeling of TCDD based on mouse and human PBPK models 3-67
3-15. Impact of toxicokinetic modeling on the extrapolation of administered dose to
HED, comparing the Emond PBPK and first-order body burden models 3-67
4-1. POD candidates for epidemiologic studies of TCDD 4-28
4-2. Models run for each study/endpoint combination in the animal bioassay
benchmark dose modeling 4-28
4-3. Summary of key animal study NOAELs, LOAELs, and BMDLs for different
dose metrics (ng/kg-day) 4-29
4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as
LASC) 4-32
4-5. Candidate points of departure for the TCDD RfD using blood-concentration-based
human equivalent doses 4-40
4-6. Qualitative analysis of the strengths and limitations/uncertainties associated with
animal bioassays possessing candidate points-of-departure for the TCDD RfD 4-43
4-7. Basis and derivation of the TCDD reference dose 4-47
5-1. Cancer slope factors calculated from Becher et al. (1998), Steenland et al. (2001)
and Ott and Zober (1996) from 2003 Reassessment Table 5-4 5-79
5-2. Cox regression coefficient estimate and incremental risk for NIOSH cohort data
as presented in Cheng et al. (2006) compared with Steenland et al. (2001) 5-80
5-3. Comparison of lipid-adjusted serum concentrations, fat concentrations, risk
specific dose estimates and equivalent oral slope factors based on upper 95th
percentile estimate of regression coefficient of all fatal cancers reported by
Cheng et al. (2006) for risk levels of 1 x 10~3, 1 x 10~4, 1 x 10~5, 1 x 10~6,
and 1 x 10 ~ 5-81
5-4. Comparison of lipid-adjusted serum concentrations, fat concentrations, risk
specific dose estimates and equivalent oral slope factors based on best estimate
of regression coefficient3 of all fatal cancers reported by Cheng et al. (2006) for
risk levels of 1 x 10~3, 1 x 10~4, 1 x 10~5, 1 x 10~6, and 1 x 10~7 5-81
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xix DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
5-5. Kociba et al. (1978) male rat tumor incidence dataa and blood concentrations for
dose-response modeling 5-82
5-6. Kociba et al. (1978) female rat tumor incidence dataa and blood concentrations
for dose-response modeling 5-82
5-7. NTP (1982) female rat tumor incidence dataa and blood concentrations for dose-
response modeling 5-83
5-8. NTP (1982) male rat tumor incidence dataa and blood concentrations for dose-
response modeling 5-83
5-9. NTP (1982) female mouse tumor incidence dataa and blood concentrations for
dose-response modeling 5-84
5-10. NTP (1982) male mouse tumor incidence dataa and blood concentrations for
dose-response modeling 5-84
5-11. NTP (2006) female rat tumor incidence dataa and blood concentrations for dose-
response modeling13 5-85
5-12. Toth et al. (1979) male mouse tumor incidence dataa and blood concentrations for
dose-response modeling 5-85
5-13. Comparison of multi-stage modeling results across cancer bioassays using blood
concentrations 5-86
5-14. Individual tumor points of departure and slope factors using blood concentrations 5-88
5-15. Multiple tumor points of departure and slope factors using blood concentrations 5-89
5-16. Illustrative RfDs based on tumorigenesis in experimental animals 5-90
5-17. Illustrative RfDs based on hypothesized key events in TCDD's MO As for liver
and lung tumors 5-91
5-18. Dichotomous Hill model fits to combined adenoma and carcinoma data from
Kociba et al. (1987) as re-evaluated by Goodman and Sauer (1992)a 5-92
5-19. Comparison of principal epidemiological studies 5-93
6-1. Key sources of uncertainty 6-40
6-2. PODs and amenability for uncertainty quantification 6-41
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xx DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES
2-1. EPA's process to select and identify in vivo mammalian and epidemiologic studies
for use in the dose-response analysis of TCDD 2-232
2-2. EPA's process to evaluate available epidemiologic studies using study inclusion
criteria for use in the dose-response analysis of TCDD 2-233
2-3. EPA's process to evaluate available animal bioassay studies using study inclusion
criteria for use in the dose-response analysis of TCDD 2-234
3-1. Liver/fat concentration ratios in relation to TCDD dose at various times after oral
administration of TCDD to mice 3-68
3-2. First-order elimination rate fits to 36 sets of serial TCDD sampling data from
Seveso patients as function of initial serum lipid TCDD 3-69
3-3. Observed relationship of fecal 2,3,7,8-TCDD clearance and estimated percent
body fat 3-70
3-4. Unweighted empirical relationship between percent body fat estimated from body
mass index and TCDD elimination half-life—combined Ranch Hand and Seveso
observation 3-71
3-5. Relevance of candidate dose metrics for dose-response modeling, based on mode
of action and target organ toxicity of TCDD 3-72
3-6. Process of estimating a human-equivalent TCDD lifetime average daily oral
exposure (dH) from an experimental animal average daily oral exposure (dA)
based on the body-burden dose metric 3-73
3-7. Human body burden time profiles for achieving a target body burden for different
exposure duration scenarios 3-74
3-8. Schematic of the CADM structure 3-75
3-9. Comparison of observed and simulated fractions of the body burden contained in
the liver and adipose tissues in rats 3-76
3-10. Conceptual representation of PBPK model for rat exposed to TCDD 3-77
3-11. Conceptual representation of PBPK model for rat developmental exposure to
TCDD 3-78
3-12. TCDD distribution in the liver tissue 3-79
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxi DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES (continued)
3-13. Growth rates for physiological changes occurring during gestation, (a) Placental
growth during gestation 3-80
3-14. Comparisons of model predictions to experimental data using a fixed elimination
rate model with hepatic sequestration (A) and an inducible elimination rate model
with (B) and without (C) hepatic sequestration 3-81
3-15. PBPK model simulation of hepatic TCDD concentration (ppb) during chronic
exposure to TCDD at 50, 150, 500, 1,750 ng TCDD/BW using the inducible
elimination rate model compared with the experimental data measured at the
end of exposure 3-82
3-16. Model predictions of TCDD blood concentration in 10 veterans (A-J) from Ranch
Hand Cohort 3-83
3-17. Time course of TCDD in blood (pg/g lipid adjusted) for two highly exposed
Austrian women 3-84
3-18. Observed vs. Emond et al. (2005) model simulated serum TCDD concentrations
(pg/g lipid) over time (In = natural log) in two Austrian women 3-85
3-19. Comparison of the dose dependency of TCDD elimination in the Emond model
vs. observations of nine Ranch Hand veterans and two highly exposed Austrian
patients 3-86
3-20. Sensitivity analysis was performed on the inducible elimination rate 3-87
3-21. Experimental data (symbols) and model simulations (solid lines) of (A) blood,
(B) liver and (C) adipose tissue concentrations of TCDD after oral exposure to
150 ng/kg-day, 5 days/week for 17 weeks in mice 3-88
3-22. Comparison of PBPK model simulations with experimental data on liver
concentrations in mice administered a single oral dose of 0.001-300 jag TCDD/kg.... 3-89
3-23. Comparison of model simulations (solid lines) with experimental data (symbols)
on the effect of dose on blood (cb), liver (cli) and fat (cf) concentrations following
repetitive exposure to 0.1-450 ng TCDD/kg, 5 days/week for 13 weeks in mice 3-90
3-24. Comparison of experimental data (symbols) and model predictions (solid lines) of
(A) blood, (B) liver and (C) adipose tissue concentrations of TCDD after oral
exposure to 1.5 ng/kg-day, 5 days/week for 17 weeks in mice 3-91
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES (continued)
3-25. Comparison of experimental data (symbols) and model predictions (solid lines) of
(A) blood concentration, (B) liver concentration, (C) adipose tissue concentration
(D) feces excretion (% dose) and (E) urinary elimination (% dose) of TCDD after
oral exposure to 1.5 ng/kg-day, 5 days/week for 13 weeks in mice 3-92
3-26. Comparison of experimental data (symbols) and model predictions (solid lines) of
(A) blood concentration, (B) liver concentration, (C) adipose tissue concentration
(D) feces excretion (% dose) and (E) urinary elimination (% dose) of TCDD after
oral exposure to 150 ng/kg-day, 5 days/week for 13 weeks in mice 3-93
3-27. PBPK model simulations (solid lines) vs. experimental data (symbols) on the
distribution of TCDD after a single acute oral exposure to A-B) 0.1, C-D) 1.0
and E-F) 10 jag of TCDD/kg of body weight in mice 3-94
3-28. PBPK model simulation (solid lines) vs. experimental data (symbols) on the
distribution of TCDD after a single dose of 24 [j,g/kgBW on GD 12 in mice 3-95
3-29. Comparison of the near-steady-state body burden simulated with CADM and
Emond models for a daily dose ranging from 1 to 10,000 ng/kg-day in rats and
humans 3-96
3-30. TCDD serum concentration-time profile for lifetime, less-than-lifetime and
gestational exposure scenarios, with target concentrations shown for each; profiles
generated with Emond human PBPK model 3-97
3-31. TCDD serum concentration-time profile for lifetime, less-than-lifetime and
gestational exposure scenarios, showing continuous intake levels to fixed target
concentration; profiles generated with Emond human PBPK model 3-98
4-1. EPA's process to select and identify candidate PODs from key epidemiologic
studies for use in the noncancer risk assessment of TCDD 4-48
4-2. EPA's process to select and identify candidate PODs from key animal bioassays
for use in noncancer dose-response analysis of TCDD 4-49
4-3. Exposure-response array for ingestion exposures to TCDD 4-50
4-4. Candidate RfD array 4-51
5-1. Mechanism of altered gene expression by AhR 5-96
5-2. TCDD's hypothesized modes of action in site-specific carcinogenesis 5-97
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxiii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES (continued)
5-3. EPA's process to select and identify candidate OSFs from key animal bioassays
for use in the cancer risk assessment of TCDD 5-98
5-4. Representative endpoints for each of the hypothesized key events following
AhR activation for TCDD-induced liver tumors 5-99
5-5. Representative endpoints for two hypothesized key events following AhR
activation for TCDD-induced lung tumors 5-100
5-6. Multistage model fit to cholangiocarcinoma response data (NTP, 2006) with
comparison to linear model fit 5-101
5-7. Weibull model fits to Kociba/G&S liver tumor response data with alternative
dose metrics 5-101
6-1. Back-casted vs. predicted TCDD serum levels for a worker subset 6-42
6-2. Distribution of in vivo unweighted REP values in the 2004 database 6-43
6-3. Plot of observed rat cholagiocarcinoma incidence and central estimate of Hill
Model fit to the data vs. human equivalent dose 6-44
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxiv DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS
2,4,5-T
2,4,5-trichlorophenoxyacetic acid
2,4-D
2,4-dichlorophenoxyacetic acid
AA
ascorbic acid
ACOH
acetanilide-4-hydroxylase
AHH
aryl hydrocarbon hydroxylase
AhR
aryl hydrocarbon receptor
AhR-/-
AhR-deficient
AIC
Akaike Information Criterion
ANL
Argonne National Laboratory
ANOVA
analysis of variance
APE
airborne particulate extract
ASAT
aspartate aminotransferase
AUC
area under the curve
bHLH-PAS
basic helix-loop-helix, Per-Arnt-Sim
Bmax
equilibrium maximum binding capacity
BMD
benchmark dose
BMDL
benchmark dose lower confidence bound
BMDS
Benchmark dose software
BMI
body mass index
BMR
benchmark response
BPS
balanopreputial separation
BROD
benzyloxy resoufin-O-deethylase
b-TSH
blood thyroid-stimulating hormone
BW
body weight
C
cerebellum
CADM
concentration- and age-dependent elimination model
Cc
cerebral cortex
CI
confidence interval
CSAF
chemical-specific adjustment factor
CSLC
cumulative serum lipid concentration
Cx
connexin
CYP
cytochrome P450
Da:HED
ratio of administered dose to HED
DEN
diethylnitrosamine
df
degrees of freedom
DLC
dioxin-like compound
DRE/XRE
dioxin/xenobiotic response elements
DRL
differential reinforcement of low rate
DSA
delayed spatial alteration
e2
17P-estradiol
EDX
effective dose eliciting x percent response
EGFR
epidermal growth factor receptor
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxv DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
EPA
Environmental Protection Agency
ER
estrogen receptor
EROD
7-ethoxyresorufin-O-deethylase
ERa
estrogen receptor alpha
EU
European Union
FFA
free fatty acid
FR
fixed-ratio
FSH
follicle stimulating hormone
FT4
free thyroxine
GD
gestation day
GSH
glutathione stimulating hormone
GSH-Px
glutathione stimulating hormone peroxidase
GST
glutathione-»Y-transferase
H
hippocampus
HCH
hexachl orocy cl ohexane
HED
human equivalent dose
HQ
hazard quotient
HR
hazard ratio
Hsp90
heat shock protein 90
IARC
International Agency for Research on Cancer
IGF
insulin-like growth factor
IL
interleukin
ILSI
International Life Sciences Institute
i.p.
intraperitoneal
IRIS
Integrated Risk Information System
KABS
oral absorption parameters
LASC
lipid-adjusted serum concentration
LD50
lethal dose eliciting x percent response
LED
lower confidence effective dose
LEDX
lower bound of the 95% confidence interval on the dose that yields an x% effect
LH
luteinizing hormone
LOAEL
lowest-observed-adverse-effect level
LOAELhed
HED estimate based on LOAELs
LOEL
lowest-observed-adverse level
MCH
mean corpuscular hemoglobin
MCMC
Markov Chain Monte Carlo
MCV
mean corpuscular volume
MOA
mode of action
MOE
margin of exposure
MROD
7-methoxyresorufin-O-deethylase
MTD
maximum tolerated dose
NAS
National Academy of Sciences
NIOSH
National Institute for Occupational Safety and Health
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxvi DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
NOAEL
no-observed-adverse-effect level
NOEL
no-ob served-effect level
NRC
National Research Council
NTP
National Toxicology Program
OR
odds ratio
OSF
oral slope factor
PA
permeability x area
PAI2
plasminogen activator inhibitor 2
PBMC
peripheral blood mononuclear cells
PBPK
physiologically based pharmacokinetic
PCB
polychlorinated biphenyl
PCDD
polychlorinated dibenzo-/>dioxin
PCDF
polychlorinated dibenzofuran
PEPCK
phosphoenolpyruvate carboxykinase
PF
adipose tissue:blood partition coefficient
PHAH
polyhalogenated aromatic hydrocarbons
PK
pharmacokinetic
PND
postnatal day
POD
point of departure
PP
phosphotyrosyl protein
PRA
probabilistic risk assessment
PRE
body:blood partition coefficient
PROD
7-pentoxyresorufin-O-deethylase
RAR
retinoic acid receptor
REP
relative potency
RfC
reference concentration
RfD
reference dose
RL
reversal learning
RL
risk level
RR
rate ratios
RR
relative risk
RT-PCR
reverse transcription polymerase chain reaction
RXR
retinoid X receptor
S
saline
SA
superoxide anion
SAhRM
SRM for AhRs
S-D
Sprague-Dawley
SD
standard deviation
SIR
standardized incidence ratio
SMR
standardized mortality ratio
SOD
superoxide dismutase
SRBC
sheep red blood cell
SSB
single-strand break
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxvii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS AND ACRONYMS (continued)
SWHS
Seveso Women's Health Study
T4
thyroxine
TBARS
thiobarbituric acid-reactive substances
TCB
3,3',4,4'-tetrachlorobiphenyl
TCDD
2,3,7,8 -T etrachl orodib enzo-p-di oxin
TCP
2,4,5-trichlorophenol
TEF
toxicity equivalence factor
TEQ
toxicity equivalence
TGFa
transforming growth factor a
TK
toxicokinetic
TNF-a
tumor necrosis factor alpha
TOTTEQ
total toxicity equivalence
TSH
thyroid stimulating hormone
TT4
total thyroxine
TWA
time-weighted average
U.S. NRC
U.S. Nuclear Regulatory Commission
UDP
uridine diphosphate
UDPGT
UDP-glucoronosyl transferase
LIED
upper confidence bound for the effective dose
UF
uncertainty factor
UFa
interspecies extrapolation factor
UFd
database factor
UFh
human interindividual variability
UFl
LOAEL-to-NOAEL UF
UFS
subchronic-to-chronic UF
UGT
UDP-glucuronosyltransferase
UGT1
uridine diphosphate glucuronosyltransferase I
Vd
volume of distribution
WHO
World Health Organization
ZS@Z
zero slope at zero dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxviii DRAFT—DO NOT CITE OR QUOTE
-------
PREFACE
This document contains the U.S. Environmental Protection Agency's (EPA), Office of
Research and Development, National Center for Environmental Assessment's response to the
National Academy of Sciences (NAS, 2006) review of Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds
(U.S. EPA, 2003). Given that the key recommendations of the NAS refer to issues related to
TCDD dose-response assessment and quantitative uncertainty analysis, EPA's response focuses
on understanding human dose response for TCDD (both cancer and noncancer endpoints), and
the feasibility of conducting quantitative uncertainty analysis in TCDD dose-response
assessment.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxix DRAFT—DO NOT CITE OR QUOTE
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS
PRIMARY AUTHORS
National Center for Environmental Assessment, U.S. Environmental Protection Agency,
Cincinnati, OH
Belinda Hawkins
Janet Hess-Wilson
Glenn Rice
Jeff Swartout
Linda K. Teuschler
CONTRIBUTING AUTHORS
National Center for Environmental Assessment, U.S. Environmental Protection Agency,
Cincinnati, OH
Scott Wesselkamper
Michael Wright
Bette Zwayer
National Health and Environmental Effects Research Laboratory, U.S. Environmental Protection
Agency, Research Triangle Park, NC
Hi sham El-Masri
Argonne National Laboratory, Argonne, IL
Margaret MacDonell
Emory University, Atlanta, GA
Kyle Steenland
Resources for the Future, Washington, DC
Roger M. Cooke
University of Montreal, Montreal, Canada
Kannan Krishnan
CONTRIBUTORS
National Center for Environmental Assessment, U.S. Environmental Protection Agency,
Washington, DC
Karen Hogan
Leonid Kopylev
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxx DRAFT—DO NOT CITE OR QUOTE
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
CONTRIBUTORS continued
Argonne National Laboratory, Argonne, IL
Maryka H. Bhattacharyya Mary E. Finster
Andrew Davidson David P. Peterson
Clark University, Worcester, MA
Dale Hattis
Colorado State University, Fort Collins, CO
Raymond Yang
Bruce Allen Consulting, Chapel Hill, NC
Bruce C. Allen
ICF International, Durham, NC
Robyn Blain
Rebecca Boyles
Patty Chuang
Cara Henning
Baxter Jones
Penelope Kellar
Mark Lee
Nikki Maples-Reynolds
Amalia Marenberg
Garrett Martin
Margaret McVey
Chandrika Moudgal
Bill Mendez
Ami Parekh
Andrew Shapiro
Audrey Turley
National Toxicology Program, Research Triangle Park, NC
Michael Devito
Penn State University, University Park, PA
Jack P. Vanden Heuvel
Risk Sciences International, Ottawa, Ontario
Jessica Dennis Salomon Sand
Dan Krewski Natalia Shilnikova
Greg Paoli Paul Villenueve
University of California-Berkeley, Berkeley, CA
Brenda Eskenazi
University of California-Irvine, Irvine, CA
Scott Bartell
University of Montreal; BioSimulation Consulting, Newark, DE
Claude Emond
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxi DRAFT—DO NOT CITE OR QUOTE
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
INTERNAL REVIEWERS
National Center for Environmental Assessment,
Glinda Cooper, Washington, DC
Ila Cote, Research Triangle Park, NC
Lynn Flowers, Washington, DC
Martin Gehlhaus, Washington, DC
Kate Guyton, Washington, DC
Samantha Jones, Washington, DC
Matthew Lorber, Washington, DC
S. Environmental Protection Agency
Eva McLanahan, Research Triangle Park, NC
Susan Rieth, Washington, DC
Reeder Sams, Research Triangle Park, NC
Paul Schlosser, Research Triangle Park, NC
Jamie Strong, Washington, DC
John Vandenberg, Research Triangle Park, NC
Paul White, Washington, DC
U.
ACKNOWLEDGMENTS
National Center for Environmental Assessment, U.S. Environmental Protection Agency
Jeff Frithsen, Washington, DC Maureen Johnson, Washington, DC
Annette Gatchett, Cincinnati, OH Peter Preuss, Washington, DC
Andrew Gillespie, Cincinnati, OH Linda Tuxen, Washington, DC
Marie Nichols-Johnson, Cincinnati, OH
Colorado State University, Fort Collins, CO
William H. Farland
ECFlex, Inc., Fairborn, OH
Dan Heing
Heidi Glick
Amy Prues
Lana Wood
IntelliTech Systems, Inc., Fairborn, OH
Cris Broyles
Luella Kessler
Debbie Kleiser
Stacey Lewis
Linda Tackett
National Institute of Environmental Health Sciences, Research Triangle Park, NC
Linda S. Birnbaum
Christopher J. Portier
National Toxicology Program, Research Triangle Park, NC
Nigel Walker
2009 Dioxin Workshop Participants
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
EXECUTIVE SUMMARY
OVERVIEW
The U.S. Environmental Protection Agency (EPA) is committed to the development of
risk assessment information of the highest scientific integrity for use in protecting human health
and the environment. Scientific peer review is an integral component of the process EPA uses to
generate high quality toxicity and exposure assessments of environmental contaminants. To this
end, EPA asked the National Academy of Sciences (NAS) to review its comprehensive human
health risk assessment external review draft entitled, Exposure and Human Health Reassessment
of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds (U.S. EPA, 2003;
"2003 Reassessment"). This current document, EPA's Response to "Health Risks from Dioxin
and Related Compounds: Evaluation of the EPA Reassessment" Published by the National
Research Council of the National Academies, directly and technically responds to key comments
and recommendations pertaining to TCDD dose-response assessment published by the NAS in
their review (NAS, 2006a). This document only addresses issues pertaining to TCDD dose-
response assessment.
In response to the recommendations presented in the 2006 NAS review, EPA
Administrator Jackson released EPA's "Science Plan for Activities Related to Dioxins in the
Environment' ("Science Plan") on May 26, 2009.1 There are five key components of the
Science Plan that pertain to EPA's response to the NAS comments on TCDD dose-response
assessment:
1. EPA will release a draft report that responds to the recommendations and comments
included in the NAS review of EPA's 2003 Reassessment.
• EPA's National Center for Environment Assessment (NCEA), in the Office of
Research and Development (ORD), will prepare a limited response to key comments
and recommendations in the NAS report (draft response to comments report).
• The draft response will focus on dose-response issues raised by the NAS and will
include an analysis of relevant new key studies.
• The draft response will be provided for public review and comment and independent
external peer review.
Available online at http://www.epa.gov/dioxin/scienceplan.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxiii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
• The draft response will also include an evaluation of some of the significant
recommendations that are difficult for EPA to address given the current state of
science, and a detailed rationale for these conclusions.
• The peer review will be conducted by EPA's Science Advisory Board (SAB), an
independent review body chartered under the Federal Advisory Committee Act.
• The draft response to comments report will be completed and released for public and
peer review by December 31, 2009.
2. EPA will provide the draft response to comments report for internal and external review.
3. The SAB will review the science content of the response to comments report.
4. EPA will review impacts of the draft response to comments report on its 2003
Reassessment.
5. EPA will release the final response to comments report and focus on completion of the
2003 Reassessment.
This document responds to and addresses key NAS comments relating to TCDD dose-
response assessment. Three separate EPA activities address additional NAS comments
pertaining to toxicity equivalence factors (TEFs) and background exposure levels. Information
on the application of the dioxin TEFs is published elsewhere by EPA for both ecological (U.S.
EPA, 2008a) and human health (U.S, EPA, 2009a) risk assessment. As a consequence, EPA
does not directly address TEFs herein, but makes use of the concept of toxicity equivalence
(TEQ)2 as applicable to the analysis of exposure dose in epidemiologic studies. Furthermore,
addressing the NAS recommendations pertaining to the assessment of human exposures to
TCDD and other dioxins, information on updated background levels of dioxin in the U.S.
population has been recently reported by EPA (Lorber et al., 2009).
The NAS identified three key recommendations requiring substantial improvement to
support a scientifically robust characterization of human responses to exposures to TCDD.
These three key areas are (1) improved transparency and clarity in the selection of key data sets
for dose-response analysis, (2) further justification of approaches to dose-response modeling for
cancer and noncancer endpoints, and (3) improved transparency, thoroughness, and clarity in
quantitative uncertainty analysis. NAS also encouraged EPA to calculate a Reference Dose
(RfD), and provided numerous specific comments on various aspects of EPA's 2003
2Toxicity equivalence (TEQ) is the product of the concentration of an individual dioxin like compound in an
environmental mixture and the corresponding TCDD TEF for that compound. These products are summed to yield
the TEQ of the mixture.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxiv DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Reassessment. The three key recommendations specifically pertain to dose-response assessment
and uncertainty analysis. Therefore, EPA's response to the NAS in this document is focused on
these issues. EPA thoroughly considered the recommendations of the NAS and responds with
scientific and technical evaluation of TCDD dose-response data via:
• an updated literature search that identified new TCDD dose-response studies (see
Section 2);
• a kickoff workshop that included the participation of external experts in TCDD health
effects, toxicokinetics, dose-response assessment and quantitative uncertainty analysis;
these experts discussed potential approaches to TCDD dose-response assessment and
considerations for EPA's response to NAS (see Appendix A);
• detailed study inclusion criteria and processes for the selection of key studies (see
Section 2.3) and epidemiologic and animal bioassay data for TCDD dose-response
assessment (see Section 2.4.1/Appendix B and Section 2.4.2, respectively);
• kinetic modeling to quantify appropriate dose metrics for use in TCDD dose-response
assessment (see Section 3);
• dose-response modeling for all appropriate noncancer and cancer data sets (see
Section 4.2/Appendix E and Section 5.2.3/Appendix F, respectively);
• thorough and transparent evaluation of the selected TCDD data for use in the derivation
of an RfD and an oral slope factor (OSF) (see Sections 4.2 and 5.2.3, respectively);
• the development of an RfD (see Section 4.3);
• the development of a revised OSF (see Section 5.3) with an updated cancer weight of
evidence determination for TCDD based on EPA's 2005 Cancer Guidelines (U.S. EPA,
2005) (see Section 5.1.2);
• consideration of nonlinear dose-response approaches for cancer, including illustrative
RfDs for cancer precursor events and tumors (see Section 5.2.3.4) ; and
• discussion of the feasibility and utility of quantitative uncertainty analysis for TCDD
dose-response assessment (see Section 6).
Each of the activities listed above is briefly described in this Executive Summary, and is
described in detail in the related sections of this document.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxv DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
PRELIMINARY ACTIVITIES UNDERTAKEN BY EPA TO ENSURE THAT THIS
TECHNICAL RESPONSE REFLECTS THE CURRENT STATE-OF-THE-SCIENCE
As part of the development of this document, EPA undertook two activities that included
public involvement: an updated literature search and a scientific expert workshop. The adverse
health effects associated with TCDD exposures are documented extensively in epidemiologic
and toxicologic studies. As such, the database of relevant information pertaining to the dose-
response assessment of TCDD is vast and constantly expanding. Responding directly to the
NAS recommendation to use the most current and up-to-date scientific information related to
TCDD, EPA, in collaboration with Argonne National Laboratory (ANL), developed an updated
literature database of peer-reviewed studies on TCDD toxicity, including in vivo mammalian
dose-response studies and epidemiologic studies. An initial literature search for studies
published since the 2003 Reassessment was conducted to identify studies published between the
year 2000 and October 31, 2008. EPA published the initial literature search results in the Federal
Register in November 2008 and invited the public to review the list and submit additional peer-
reviewed relevant studies. Additional studies identified by the public and through continued
work on this response have been incorporated into the final set of studies for TCDD dose-
response assessment (updated through October 2009). EPA believes that the implementation of
this rigorous search strategy ensures that the most current and relevant studies were considered
for the technical response to NAS and TCDD dose-response assessment included herein.
To assist in responding to the NAS, EPA, in collaboration with ANL, convened a
scientific expert workshop ("Dioxin Workshop") in February 2009 that was open to the public.
The primary goals of the Dioxin Workshop were to identify and address issues related to the
dose-response assessment of TCDD and to ensure that EPA's response to the NAS focused on
the key issues while reflecting the most meaningful science. EPA and ANL assembled expert
scientists and asked them to identify and discuss the technical challenges involved in addressing
the NAS comments, discuss approaches for addressing these key recommendations, and to assist
in the identification of important published and peer-reviewed literature on TCDD. The
workshop was structured into seven scientific topic sessions as follows: (1) quantitative dose-
response modeling issues, (2) immunotoxicity, (3) neurotoxicity and nonreproductive endocrine
effects, (4) cardiovascular toxicity and hepatotoxicity, (5) cancer, (6) reproductive and
developmental toxicity, and (7) quantitative uncertainty analysis of dose-response. External
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxvi DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
co-chairs (i.e., scientists who were not members of EPA or ANL) were asked to facilitate the
sessions and then prepare summaries of discussions occurring in each session. The session
summaries formed the basis of a final workshop report (U.S. EPA, 2009c, Appendix A of this
document). Some of the key outcomes from the workshop include the following
recommendations:
• to further develop study selection criteria for evaluating the suitability of developing
dose-response models based on animal bioassays and human epidemiologic studies;
• to use kinetic modeling to identify relevant dose metrics and dose conversions between
test animal species and humans, and between human internal dose measures and human
intakes;
• to consider newer human or animal (e.g., NTP, 2006) publications when evaluating
quantitative dose-response models for cancer;
• to consider both linear and nonlinear modeling in the cancer dose-response analysis.
The discussions held during the Dioxin Workshop helped inform, guide, and focus EPA's
response to NAS.
EPA'S APPROACH TO CONSIDERING TRANSPARENCY AND CLARITY IN THE
SELECTION OF KEY STUDIES AND DATA SETS FOR DOSE-RESPONSE
MODELING
One of the key NAS recommendations to EPA was to utilize a clear and transparent
process for the selection of key studies and data sets for dose-response assessment. EPA agrees
with the NAS and believes that clear delineation of the study selection process and decisions
regarding key studies and data sets will facilitate communication of critical decisions made in the
TCDD dose-response assessment. EPA developed detailed processes and TCDD-specific
criteria for the selection of key dose-response studies. These criteria are based on common
practices and current guidance for point of departure (POD) identification and RfD and OSF
derivation while also considering issues specifically related to TCDD. Following the selection of
key studies, EPA employed additional processes to further select and identify cancer and
noncancer datasets from these key studies for use in dose-response analysis of TCDD.
Figure ES-1 presents EPA's study evaluation process for the epidemiologic studies
considered for TCDD dose-response assessment, including specific study inclusion criteria (see
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxvii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Section 2.3). EPA applied TCDD-specific epidemiologic study inclusion criteria to all
epidemiologic studies published on TCDD and dioxin-like compounds (DLCs) that had been
identified in the TCDD literature database (see Section 2.4.1, Appendix B). The studies were
initially evaluated using five considerations (see Figure ES-1) that provide the most relevant
kinds of information needed for quantitative human health risk analyses. Additionally, EPA
examined whether the human exposures were primarily to TCDD and if the TCDD exposures
could be quantified so that dose-response analyses could be conducted. Finally, EPA required
that the effective dose be estimable: (1) for cancer, information is required on long-term
exposures, and (2) for noncancer, information is required regarding the appropriate time window
of exposure that is relevant for a specific, nonfatal health endpoint. Therefore, the study should
include an appropriate latency period between TCDD exposure and the onset of the effect. Only
studies meeting these criteria were included in EPA's TCDD dose-response analyses (see
Section 2.4.3).
Figure ES-2 presents EPA's study evaluation process for mammalian bioassays
considered for TCDD dose-response assessment, including the specific study inclusion criteria
(see Section 2.3). EPA applied TCDD-specific in vivo mammalian bioassay study inclusion
criteria to all of the bioassay studies of TCDD that had been identified in the TCDD literature
database (see Section 2.4.2). After ascertaining that a study had been published in the peer-
reviewed literature, EPA applied dose requirements to the lowest tested average daily doses in
each study, with specific requirements for cancer (<1 (j,g/kg-day) and noncancer (<30 ng/kg-day)
studies to ensure that only low-dose TCDD bioassays would be considered. These dose
requirements were used to eliminate those studies that would not be selected for development of
an RfD or an OSF because the lowest doses tested were too high relative to other TCDD
bioassays. EPA also required that the bioassays exposed the animals via the oral route to TCDD
only and that the purity of the TCDD was specified. Finally, the studies were evaluated using
four considerations (see Figure ES-2) regarded as providing the most relevant information for
development of quantitative human health risk analyses from animal bioassay data. Only the
bioassay studies meeting these criteria and considerations were included in EPA's TCDD dose-
response analyses (see Section 2.4.3).
Applying the study inclusion criteria for both epidemiologic and mammalian bioassay
datasets resulted in a list of key noncancer and cancer studies that were considered for
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxviii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
quantitative dose-response analyses of TCDD. Endpoints from these studies that were not
considered to be toxicologically relevant were eliminated from consideration (see
Section 4.2.1.1, Appendix G). The study/endpoint dataset combinations from the remaining
studies were then subjected to dose-response assessment, and PODs for use in developing RfDs
or OSFs were identified. PODs included no-observed-adverse-effect levels (NOAELs), lowest-
observed-adverse-effect levels (LOAELs) or lower bound benchmark dose levels (BMDLs). The
most sensitive PODs were selected as candidates for derivation of the RfD and OSF.
USE OF KINETIC MODELING TO ESTIMATE TCDD DOSES
NAS recommended that EPA utilize state-of-the-science approaches to finalize the 2003
Reassessment. Although NAS concurred with EPA's use of first-order body burden models in
the 2003 Reassessment, analyses of recent TCDD literature and comments by experts at the
Dioxin Workshop suggested that the understanding of TCDD kinetics had increased significantly
since the release of EPA's 2003 Reassessment. These advances led to the development of
several pharmacokinetic models for TCDD (e.g., Emond et al., 2004, 2005, 2006; Aylward et al.,
2005a) and resulted in EPA's incorporation of TCDD kinetics in the dose-response assessment
of TCDD.
The evaluation of internal dose in exposed humans and other species is facilitated by an
understanding of pharmacokinetics (i.e., absorption, distribution, metabolism, and excretion).
TCDD pharmacokinetics are influenced by three distinctive features: (1) TCDD is highly
lipophilic, (2) TCDD is slowly metabolized, and (3) TCDD induces binding proteins in the liver.
The overall impact of these factors results in preferential storage of TCDD in adipose tissue, a
long half-life of TCDD in blood due to slow metabolism, and sequestration in liver tissue when
binding induction becomes significant. As these kinetic features control target tissue levels of
dioxin, they become important in relating toxicity in animals to possible effects in humans.
Consideration of pharmacokinetic mechanisms is critical to the selection of the dose
metrics of relevance to dose-response modeling of TCDD. Earlier assessments for TCDD,
including the 2003 Reassessment, used estimates of body burden as the dose metric for
extrapolation between animals and humans. These body burden calculations used a simple one-
compartment kinetic model based on the assumption of a first-order decrease in the levels of
administered dose as a function of time. However, the assumption of a constant half-life value
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xxxix DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
for the clearance of TCDD from long-term or chronic exposure is not well-supported
biologically given the dose-dependant elimination observed in rodents and humans. The
dynamic disposition and redistribution of TCDD between blood, fat, and liver as a function of
time and dose is better described using biologically-based models. Additionally, these models
provide estimates for other dose metrics (e.g., serum or tissue levels) that are more biologically
relevant to response than body burden estimated based on an assumption of first-order
elimination over time.
EPA considered the following possible dose metrics for TCDD: administered dose,
absorbed dose, first-order body burden, serum or whole blood concentration, tissue
concentration, and functional-related metrics of relevance to the mode of action (MOA) (e.g.,
receptor occupancy) (see Section 3.3.4.1). After careful evaluation of these dose metrics, EPA
chose to use TCDD concentration in whole blood as the dose metric for assessing TCDD dose
response in this document; blood concentration reflects both the body burden and the dose to
target tissues. EPA used the time-weighted average whole-blood concentration over the relevant
exposure periods for all continuous dosing protocols, dividing the area under the time-course
concentration curve (AUC) by the exposure duration.3
Several biologically-based kinetic models for TCDD exist in the literature. The more
recent pharmacokinetic models explicitly characterize the concentration-dependent elimination
of TCDD (Carrier et al., 1995a, b; Emond et al., 2004, 2005, 2006; Aylward et al., 2005a). The
biologically-based pharmacokinetic models describing the concentration-dependent elimination
(i.e., the pharmacokinetic models of Aylward et al. [2005a] and Emond et al. [2005, 2006]) are
relevant for application to simulate the TCDD dose metrics in humans and animals exposed via
the oral route. The rationale for considering the application of the Aylward et al. (2005a) and
Emond et al. (2004, 2005, 2006) models was largely based on the fact that both models reflect
research results from recent peer-reviewed publications, and both models are formulated with
dose-dependent hepatic elimination consistent with the physiological understanding of TCDD
kinetics.
3For the Seveso cohort, which had a high single exposure followed by low-level background exposures leading to a
gradual decline in the internal TCDD concentrations, EPA estimated dose as the mean of the peak exposure and the
average exposure over a defined critical exposure window (see Section 4.2.2).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xl DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Of the two selected models, the pharmacokinetic model developed by Emond et al.
(2004, 2005, 2006) is more physiologically-based, as compared to the Aylward et al. (2005a)
model, and models the blood compartment directly in the rat, mouse, and human; there are also
gestational and life-time nongestational forms of the Emond PBPK model. Therefore, in this
document, EPA chose the Emond rodent physiologically-based pharmacokinetic (PBPK) model
to estimate blood TCDD concentrations based on administered doses (see Section 3.3.4,
Appendix C). EPA assumed that the same blood TCDD levels that led to effects in animals
would also lead to effects in humans; therefore, the Emond human PBPK model was used to
estimate the lifetime average daily oral doses (consistent with the chronic RfD and OSF) that
would correspond to the blood TCDD concentrations estimated to have occurred during the
animal bioassays. EPA used the same Emond human PBPK model to estimate the lifetime
average daily doses that would correspond to the TCDD blood or tissue concentrations reported
in the epidemiological studies. These estimates are the Human Equivalent Doses (HEDs) that
are used to develop candidate RfDs and OSFs for TCDD.
Because TCDD elimination is inducible in the Emond model, ratios of daily averaged
intake to long-term blood concentrations are not linear. OSFs based on internal dose measures
(i.e., blood or fat concentrations) will be linear only with respect to blood or fat TCDD
concentrations; however, ingested TCDD doses are not linear with TCDD fat or blood
concentrations in the Emond PBPK model. Thus, an OSF that is linear with TCDD in the fat is
not linear with the ingested TCDD dose. A consequence, then, of using the Emond PBPK model
is that risk-specific TCDD intake rates corresponding to target risk levels need to be tabulated for
use in human health risk assessment.
DERIVATION OF AN RfD FOR TCDD
The NAS specifically recommended that EPA derive an RfD for TCDD. Through a
transparent study selection process, EPA identified key studies from both human epidemiologic
and animal bioassay studies. To select candidate PODs for its RfD methodology, EPA applied
additional processes to the key human epidemiologic studies and animal bioassays. Figure ES-3
(exposure-response array) shows all of the candidate PODs graphically in terms of human-
equivalent intake (ng/kg-day). The human study endpoints are shown at the far left of the figure
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xli DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
and, to the right, the rodent endpoints are arranged by the following study categories: less than 1
year, greater than 1 year, reproductive, and developmental.
For each noncancer epidemiologic study that EPA selected as key, EPA evaluated the
dose-response information developed by the study authors to determine whether the study
provided noncancer effects and TCDD-relevant exposure data for a toxicologically-relevant
endpoint. If such data were available, EPA identified a NOAEL or LOAEL as a candidate POD.
Then, EPA used the Emond human PBPK model to estimate the continuous oral daily intake
(ng/kg-day) that would lead to the relevant blood TCDD concentrations associated with the
candidate POD that could be used in the derivation of an RfD. If all of this information was
available, then the result was included as a candidate POD.
Through this process, EPA identified health effects from the following four
epidemiologic studies to be considered as the basis for the RfD: Eskenazi et al. (2002)
(reproductive—increased length of menstrual cycle), Alaluusua et al. (2004) (developmental—
tooth development), Mocarelli et al. (2008) (reproductive—decreased sperm concentrations), and
Baccarelli et al. (2008) (developmental—increased thyroid-stimulating hormone levels in
neonates). All four studies are from the Seveso cohort, whose members were exposed
environmentally to high peak concentrations of TCDD as a consequence of an industrial
accident. This complicated the estimation of average daily doses associated with these specific
endpoints, however EPA was able to calculate candidate PODs for derivation of an RfD from
each of these human studies (see Section 4.2.3).
Figure ES-4 summarizes the strategy employed for identifying and selecting candidate
PODs from the key animal bioassays EPA identified for use in noncancer dose-response analysis
of TCDD (see Section 4.2.4). For each noncancer endpoint, EPA first evaluated the
toxicological relevance of each endpoint, rejecting those judged not to be relevant for RfD
derivation. Next, initial PODs (NOAELs, LOAELs, and BMDLs) based on the first-order body
burden metric, and expressed as continuous human-equivalent oral daily doses (HEDs), were
determined for all relevant endpoints. Because there were very few NOAELs, and BMDL
modeling was largely unsuccessful due to data limitations, the next stage of evaluation was
carried out using LOAELs only. Endpoints not observed at the LOAEL (i.e., reported at higher
doses) with BMDLs greater than the LOAEL were eliminated from further analysis, as they
would not be considered as candidates for the final POD on either a BMDL or NOAEL/LOAEL
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xlii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
basis (i.e. the POD would be higher than the PODs of other relevant endpoints). In addition, all
endpoints with HEDs for LOAELs (LOAELreds) beyond a 100-fold range of the lowest
identified LOAELHed were eliminated from further consideration, as they would not be potential
POD candidates either (i.e. the POD would be higher than the PODs of other relevant endpoints).
For the remaining endpoints, EPA then determined final potential PODs (NOAELs, LOAELs,
and BMDLs) based on TCDD blood concentrations obtained from the Emond rodent PBPK
models. HEDs were then estimated for each of these PODs using the Emond human PBPK
model. From these HEDs, a PODred was selected for each study as the basis for the candidate
RfD, to which appropriate uncertainty factors were applied following EPA guidelines. The
resulting candidate RfDs were then considered in the final selection process for the RfD. Other
endpoints occurring at slightly higher doses representing additional effects associated with
TCDD exposure (beyond the 100-fold LOAEL range) were evaluated, modeled, and included in
the final candidate RfD array to examine endpoints not evaluated by studies with lower PODs.
In addition, BMD modeling based on administered dose was performed on all endpoints for
comparison purposes.
For selection of the POD to serve as the basis of the RfD, EPA gave the epidemiologic
studies the highest consideration because human data are preferred in the derivation of an RfD,
given that the underlying epidemiologic and animal bioassay data all met EPA's selection
criteria. This preference for epidemiologic study data also is consistent with reccomendations of
panelists at the Dioxin Workshop. The lower end of the candidate POD distribution is dominated
by mouse studies, with mouse studies comprising six of the first seven animal bioassays yielding
the lowest PODs. EPA has less confidence in the values derived from these mouse bioassays
than the values derived from rat and human studies. The primary reason that the mouse
LOAELhedS are low is the large toxicokinetic interspecies extrapolation factor applied to mouse
data in the Emond PBPK model. In addition, each of these first seven rodent studies has other
qualitative limitations and uncertainties that make them poor candidates as the basis for the RfD.
Most of the other rodent studies yielding POD values higher than the first seven animal
bioassay studies and lower than the human studies of Mocarelli et al (2008) and Baccarelli et al.
(2008) are of small size, using 10 or fewer animals per dose group and are considered too
uncertain on which to base the final RfD. However, two of the rat bioassays—Bell et al (2007)
and NTP (2006)—were very well designed and conducted, using 30 or more animals per dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xliii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
group (see Section 4.3.4). Bell et al (2007) evaluated several reproductive and developmental
endpoints initiating TCDD exposures well before mating and continuing them through gestation.
NTP (2006) is the most comprehensive evaluation of TCDD chronic toxicity in rodents to date,
evaluating dozens of endpoints at several time points in all major tissues. It is EPA's judgement
that the toxicokinetic extrapolation of the results of these two studies to humans, however, is still
less certain than the use of human data. Despite the overall strength of the Bell and NTP studies,
EPA considers the human data to be a better basis for the TCDD RfD.
The most relevant human PODs are based on the Baccarelli et al. (2008) and Mocarelli et
al. (2008) studies, which exhibited similar LOAELs of 0.024 and 0.020 ng/kg-day, respectively.
For Baccarelli et al. (2008), EPA defined a LOAEL as the group mean of 39 ppt TCDD in
neonatal plasma which corresponds to thyroid-stimulating hormone (TSH) values above
5 |iU/mL, Baccarelli et al. did not estimate the equivalent oral intake associated with TCDD
serum concentrations, rather they provided neonatal serum TCDD concentrations for the groups
above and below 5 |iU/mL. EPA estimated the maternal intake at the LOAEL from a maternal
serum-TCDD/TSH regression model presented in Baccarelli et al. by estimating the maternal
TCDD lipid adjusted serum concentration (LASC) at which neonatal TSH exceeded 5 |iU/mL.
EPA then used the Emond PBPK model to estimate the continuous daily TCDD intake that
would result in this TCDD LASC. The resulting predicted maternal daily intake rate established
the LOAEL (0.024 ng/kg-day). EPA did not defined a NOAEL because it is not clear what
maternal intake should be assigned to the group below 5 |iU/mL.
For Mocarelli et al. (2008), EPA defined a LOAEL as the lowest exposed group mean of
68 ppt (lst-quaitile) corresponding to decreased sperm concentrations, decreased motile sperm
counts, and decreased serum estradiol in men who were 1-9 years old at the time of the Seveso
accident (initial TCDD exposure event). TCDD LASC levels were measured within
approximately one year of the initial exposure event. Because effects were only observed in men
who were under 10 years of age at the time of exposure, EPA assumed a maximum 10-year
critical exposure window for elicitation of these effects. EPA has estimated a continuous daily
oral intake of 0.020 ng/kg-day associated with the designated LOAEL from the lowest exposure
group (68 ppt), (see Section 4.2.3.2). The reference group is not designated as a NOAEL
because there is no clear zero-exposure measurement for any of these endpoints, particularly
considering the contribution of background exposure to DLCs, which futher complicates the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xliv DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
interpretation of the reference group response as a true "control" response (see discussion in
Section 4.4). However, males less than 10 years old can be designated as a sensitive population
by comparison to older males who were not affected.
The two human studies, Baccarelli et al. (2008) and Mocarelli et al. (2008), have similar
LOAELs of 0.024 and 0.020 ng/kg-day, respectively. Together, these two studies constitute the
best foundation for establishing a POD for the RfD, and are designated as co-principal studies.
Therefore, increased TSH in neonates (Baccarelli et al., 2008) and male reproductive effects
(decreased sperm count and motility, increased estradiol) are designated as cocritical effects.
Although the exposure estimate used in determination of the LOAEL for Mocarelli et al. (2008)
is more uncertain than the Baccarelli et al. (2008) exposure estimate, the slightly lower LOAEL
of 0.020 ng/kg-day from Mocarelli et al. is designated as the POD.
EPA used a composite UF of 30 for both studies. EPA applied a factor of 10 for UFL to
account for lack of a NOAEL. EPA also applied a factor of 3 (10°5) for UFH to account for
human interindividual variability because the effects were elicited in sensitive populations. A
further reduction to 1 was not made because the sample sizes in these two epidemiologic studies
were relatively small, which, combined with uncertainty in exposure estimation, may not fully
capture the range of interindividual variability. The resulting RfD for TCDD in standard units is
7 x 10-10 mg/kg-day (Table 4-7 details the RfD derivation).
WEIGHT-OF-EVIDENCE STATEMENT FOR CARCINOGENICITY
The NAS recommended that EPA update its cancer classification for TCDD and the
weight-of-evidence (WOE) statement to reflect the current state of the science and incorporate
the latest EPA Cancer Guidelines (U.S. EPA, 2005). Several notable new studies addressing
TCDD's carcinogenic potential have been published since the release of EPA's 2003
Reassessment, including several new studies of the Seveso epidemiologic cohort and an NTP
2-year cancer bioassay in female rats (NTP, 2006).
Under the 2005 Guidelines for Carcinogen Risk Assessment (U.S. EPA, 2005) TCDD is
characterized as carcinogenic to humans, based on the available data as of 2009 (see
Section 5.1.2). When evaluating the carcinogenic potential of a compound, EPA employs a
WOE approach in which all available information is evaluated and considered. In the case of
TCDD, EPA based the classification on numerous lines of evidence, including: multiple
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xlv DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
occupationally- and accidentally-exposed epidemiologic cohorts showing an association between
TCDD exposure and certain cancers or increased mortality from all cancers; extensive evidence
of carcinogenicity at multiple tumor sites in both sexes of multiple species of experimental
animals; consensus that the mode of TCDD's carcinogenic action in animals involves aryl
hydrocarbon receptor (AhR)-dependent key precursor events and proceeds through modification
of one or more of a number of cellular processes; the human AhR and rodent AhR are similar in
structure and function, and human and rodent tissue and organ cultures respond to TCDD in a
similar manner and at similar concentrations; and general scientific consensus that AhR
activation is anticipated to occur in humans and to progress to tumors.
DERIVATION OF CANDIDATE OSFs FROM EPIDEMIOLOGIC STUDIES AND
ANIMAL BIOASSAYS
There are several well-studied occupationally-exposed epidemiologic cohorts showing an
association between TCDD and increased all-cancer mortality, and several epidemiologic
cohorts exposed to TCDD as a consequence of industrial accidents showing an association
between TCDD and cancer or cancer mortality (see Section 5.2.3.1). The 2003 Reassessment
included cancer dose-response analyses based on the following three occupational cohorts: the
NIOSH cohort, an occupational cohort subject to chronic TCDD exposures (Steenland et al.,
2001); the Hamburg cohort, an occupational cohort also subject to chronic TCDD exposures
(Becher et al., 1998); and the BASF cohort, an occupational cohort subject to peak TCDD
exposures through clean-up following an industrial accident (Ott and Zober, 1996). In this
document, EPA determined that each of these studies met the epidemiologic study inclusion
criteria. Thus, after further evaluating the OSFs presented in the 2003 Reassessment for these
three studies, EPA accepted those OSF estimates and retained them as candidate OSFs in this
document. EPA also determined that two additional studies met the epidemiologic study
inclusion criteria: Cheng et al. (2006) (NIOSH cohort) and Warner et al. (2002) (Seveso cohort,
a cohort exposed environmentally as a consequence of an industrial accident). EPA was unable
to derive a credible OSF from the data presented by Warner et al. (2002) but did derive an OSF
from Cheng et al. (2006), as detailed in Text Box ES-1. In Table ES-1, EPA presents estimates
of OSFs for specific TCDD intake rates based on target risk levels of 1 x io~3, 1 x io~4, 1 x io~5,
1 x io~6, and 1 x io-7 based on Cheng et al. (2006).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xlvi DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
Text Box ES-1
OSF Calculations Using Cheng et al. (2006) Information
Following the estimation of dose using the AUC values developed from the kinetic modeling, Cheng
and colleagues derived dose-response estimates for the NIOSH cohort. For exposures lagged 15 years, the
regression coefficient of the linear slope derived by Cheng et al. (2006) was 3.3 x 10 6 per ppt-year lipid-
adjusted serum TCDD (the standard error of this regression coefficient was 1.4 x 10 6). The upper 5% of the
exposure range (individuals >252,950 ppt-year lipid adjusted serum TCDD) was excluded in estimating this
slope. Because this exclusion reduces the upper portion of the response where the slope is shallow, this likely
better represents the slope in the region of the curve where the fatal cancer risk is increasing with dose, which is
the equivalent of dropping the highest dose in an animal bioassay.
To develop an OSF for TCDD, EPA used information from Cheng et al. (2006) in its calculations as
follows:
• Upper 95th percentile slope (fi95) of 6.0 x 10 6 per ppt-year lipid adjusted serum TCDD.
• Background cancer mortality risk estimate (11,) of 0.112.
• Risk in the exposed group associated with a 1% extra risk of fatal cancer (Rexp) of 0.12088.
• Incremental cancer mortality risk in the exposed population based on a 1% extra risk (R,,) of
8.9 x 10 ' using the equation, RD = Rexp - R0
• Cumulative TCDD concentration in the fat compartment for a 1% extra risk (AUC0i) using the
following formula corresponding to equations used in Cheng et al. (2006):
AUCoi = LN(RC + Ro/Ro)//395
Cheng et al. (2006) appear to assume that the TCDD concentration in fat is the same as ppt-yr lipid
adjusted serum concentration.
• OSF associated with 1% extra risk [OSF(AUC0i)], calculated to be 7.92 x 10 by dividing 0.01 by the
AUCoi. OSF(AUCoi) is linear with the TCDD concentration in fat. Ingested TCDD doses, however,
are not linear with the predicted TCDD fat concentrations in the Emond pharmacokinetic model.
Thus, the OSF(AUC0i) that is linear with TCDD in the fat is not linear with ingested TCDD dose.
EPA calculated estimates of OSFs for specific TCDD intake rates based on target risk levels (RLs) of 1 x 10 3.
1 x 10 ', 1 x i(T5, 1 x 1() 6. and 1 x 10 7, using the following calculations:
• Area under the TCDD fat concentration curve associated with a target risk level (AUCrl) (ppt-yr) [i.e.,
RL/7.92 x 10"7 (ppt-yr)1].
• Lifetime averaged concentration of TCDD in the fat compartment associated with the target risk level
(FATrl) (ng/kg). The AUCrl estimates were then further divided by 70 years to identify (FATrl)
(ng/kg). This step essentially reverses the integration undertaken to calculate AUC0i.
• Continuous daily TCDD intake (DRL)(ng/kg-day) associated with a target risk level over a lifetime.
Using the Emond pharmacokinetic model, EPA estimated the Drl necessary to achieve the FATrl.
• Oral slope factor at the target risk level (OSFrl) (per mg/kg-day). At the target risk levels, the
associated OSFrlS range from 3.7 x 105 to 1.3 x io6 per mg/kg-day. These are calculated as
OSFrl = RL/Drl x 10s
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xlvii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
EPA also identified candidate OSFs for TCDD from key animal bioassays (see
Section 5.2.3.2). Based on the inclusion criteria, EPA selected four key rodent cancer bioassays
suitable for quantitative dose-response assessment. These included Kociba et al. (1978), NTP
(1982), and Toth et al. (1979) that were evaluated in the 2003 Reassessment, and the new NTP
(2006) rat chronic bioassay. EPA conducted dose-response modeling for each tumor type
separately (individual tumor models) as well as composite tumor incidence dose estimates
(multiple tumor models). The tumor types that EPA analyzed are shown in Table ES-2.
For each in vivo animal cancer study that qualified for TCDD dose-response assessment,
EPA selected the species/sex/tumor dataset combinations characterized as having statistically
significant increases in tumor incidences, then used the Emond rodent PBPK model to estimate
blood concentrations corresponding to each study's average daily administered dose for use in
dose-response modeling. BMDLois were then estimated for the blood concentration by two
different methodologies: (1) using the multistage cancer model for each species/sex/tumor
combination within each study, and (2) using a Bayesian Markov Chain Monte Carlo framework
that assumes independence of tumors, modeling all tumors together for each species/sex
combination within each study. The final selected models were subjected to goodness-of-fit tests
and visual inspection of fit to the raw data. Thus, for each sex/species combination within each
study, EPA generated a BMDLoi for each single tumor type and another BMDLoi for the
combined tumors. Using the Emond human PBPK model, BMDLhedS were then calculated for
each of the BMDLois, and using a linear extrapolation, OSFs were calculated by
OSF = 0.01/BMDLhed The highest OSF for a species/sex combination for either a single tumor
type or all combined tumors was selected as a candidate OSF. The OSF candidates from the key
animal bioassays are shown in Table ES-2.
DERIVATION OF TCDD ORAL SLOPE FACTOR AND RISK ESTIMATES
EPA was able to derive OSFs for tumor incidence data from four animal cancer
bioassays, as well as cancer mortality data from four epidemiological cohort studies that were
selected for TCDD dose-response modeling using the study inclusion criteria (see Section 5.3).
For the animal data, OSFs based on individual tumors were developed for 25 study/sex/endpoint
combinations, and the results ranged from 1.8 x 104 to 5.9 x 106 (per mg/kg-day). The OSFs
based on combined tumors were developed for seven study/sex combinations, and the results
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xlviii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
ranged from 3.2 x 105 to 9.4 x 106 (per mg/kg-day). EPA also developed OSFs based on four
epidemiologic studies from three cohorts, ranging from 3.75 x 105 to 2.5 x 106 (per mg/kg-day).
EPA has chosen to use the human data over the animal data as recommended by expert
panelists at EPA's 2009 Dioxin Workshop (U.S. EPA, 2009c) and in the 2005 Cancer Guidelines
(U.S. EPA, 2005). OSFs derived from the human data are consistent with the animal bioassay
results; human OSFs fall within the same range as the animal bioassay OSFs.
Among the human studies, the occupational TCDD exposures in the NIOSH and
Hamburg cohorts are assumed to be reasonably constant over the duration of occupational
exposure. In contrast, the TCDD exposure pattern for the Seveso and BASF accidents is acute,
high dose, followed by low-level background exposure. Such exposure patterns similar to those
experienced by the BASF and Seveso cohorts have been shown to yield higher estimates of risk
when compared to constant exposure scenarios with similar total exposure magnitudes (Kim et
al. 2003; Murdoch and Krewski, 1988; Murdoch et al, 1992). Thus, EPA has judged that the
NIOSH and Hamburg cohort response data are more relevant than the BASF and Seveso data for
assessing cancer risks from continuous ambient TCDD exposure in the general population.
The NIOSH (Steenland et al. 2001; Cheng et al. 2006) and Hamburg (Becher et al. 1998)
cohort studies report cumulative TCDD levels in the serum for cohort members. The most
significant difference among the Cheng et al. (2006) analysis and those of Steenland et al. (2001)
and Becher et al. (1998) is the method used to back-extrapolate exposure concentrations based
on serum TCDD measurements. Steenland et al. (2001) and Becher et al. (1998) back-
extrapolated exposures and body burdens using a first-order model with a constant half-life. In
contrast, Cheng et al. (2006) back-extrapolated body burdens using a kinetic modeling approach
that incorporated concentration- and age-dependent elimination kinetics.
Although all three of these are high-quality studies, the kinetic modeling used by Cheng
et al. (2006) is judged to better reflect TCDD pharmacokinetics, as currently understood, than the
first-order models used by Steenland et al. (2001) and Becher et al. (1998). EPA believes that
the representation of physiological processes provided by Cheng et al (2006) is more realistic
than the assumption of simple first-order kinetics and this outweighs the attendant modeling
uncertainties. Furthermore, the use of kinetic modeling is consistent with recommendations both
by the NAS and the Dioxin Workshop panel.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 xlix DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Therefore EPA has selected the results from the Cheng et al. (2006) study for derivation
of the TCDD OSF (see Section 5.3). Table ES-1 shows the oral slope factors at specific target
risk levels (OSFRLs) which range from 3.7 x 105 to 1.3 x 106 per (mg/kg-day). EPA
recommends the use of an OSF of 1.3 x 106 per (mg/kg-day) when the target risk range is 10 5 to
10~7.
CONSIDERATION OF NONLINEAR DOSE-RESPONSE APPROACHES FOR
CANCER
The NAS focused much of its review on EPA's derivation of a cancer slope factor,
commenting extensively on the extrapolation of dose-response modeling below the POD. The
NAS questioned EPA's choice of a linear, nonthreshold model for extrapolating risk associated
with exposure levels below the POD, concluding that the current scientific evidence was
sufficient to justify the use of nonlinear methods when extrapolating below the POD for dioxin
carcinogenicity.
While, based on the 2005 Cancer Guidelines, EPA deemed linear extrapolation to be
most appropriate for TCDD, EPA carefully considered the NAS recommendation to provide risk
estimates using both linear and nonlinear methods. In this document, EPA has evaluated the
information available for identifying a threshold and for estimating the shape of the dose-
response curve below the POD (see Section 5.2.3.4). EPA presents a hypothetical sublinear
dose-response modeling example of rodent carcinogenicity. EPA also presents two illustrative
examples of RfD development (i.e., nonlinear method) for carcinogenic effects of TCDD, using
data derived from animal bioassays. EPA derives illustrative RfDs for cancer based on
combined tumor response and also on hypothesized key events in TCDD's MOA for female rat
liver and lung tumors. EPA identifies a number of limitations that prevent making strong
conclusions based on the nonlinear dose-response modeling exercises.
FEASIBILITY OF QUANTITATIVE UNCERTAINTY ANALYSIS
EPA also addresses the third key recommendation of the NAS, specifically, improving
transparency, thoroughness, and clarity in quantitative uncertainty analysis (see Section 6). In
summary, NAS suggested that EPA should
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
• describe and define (quantitatively to the extent possible) the variability and
uncertainty for key assumptions used for each key endpoint-specific risk
assessment (choices of data set, POD, model, and dose metric),
• incorporate probabilistic models to the extent possible to represent the range of
plausible values,
• clearly state it when quantitation is not possible and explain what would be
required to achieve quantitation (NAS, 2006a, p. 9).
Although the NAS summarized the shortfalls in the 2003 Reassessment categorically, the
elaborations within their report often contain the qualification "if possible" and do not take a
position with regard to the feasibility of many suggestions. With appreciation for the extent of
information available for dioxin, EPA's goal herein was to examine the feasibility of a
data-driven quantitative uncertainty analysis for TCDD dose-response assessment.
In examining feasibility of quantitative uncertainty analysis, EPA recognized that
different kinds of uncertainty require different statistical treatment. Cognitive uncertainty
concerns uncertainty that can be expressed as probabilities and may be operationalized using
either frequentist or Bayesian approaches. For example, classical statistical methods yield
distributions on model parameters which reflect sample fluctuations, assuming that the model is
true. This type of uncertainty can be taken into account in the BMDL estimation. Also, for
TCDD epidemiologic data, the dose reconstruction often involves assumptions that may be
amenable to data-driven uncertainty analysis if sufficient data can be retrieved; back-
extrapolated TCDD levels, biological half-life, body fat, and background levels are example
variables that could be included in such an analysis. In addition, a Monte Carlo analysis has
been examined to develop quantitative uncertainty distributions for the RfD (e.g., Swartout et al.,
1998). Given a set of animal bioassay data, quantifying dose-response uncertainty may be
approached in different ways. The differences reflect different types of uncertainty that are
captured. A recent evaluation enumerates the following possible methodologies (Bussard et al.,
2009):
Benchmark Dose Modeling (BMD): Choose the 'best' model, and
assess uncertainty assuming this model is true. Supplemental results can compare
estimates obtained with different models, and sensitivity analyses can investigate
other modeling issues.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 li DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
Probabilistic Inversion with Isotonic Regression (PI-IR): Define
model-independent 'observational' uncertainty, and look for a model that captures
this uncertainty by assuming the selected model is true and providing for a
distribution over its parameters.
5
6
7
8
Non-Parametric Bayes (NPB): Choose a prior mean response (potency)
curve (potentially a "non-informative prior") and a precision parameter to express
prior uncertainty over all increasing dose-response relations, and update this prior
distribution with the bioassay data.
9
10
11
12
13
14
15
Bayesian Model Averaging (BMA) (as considered here): Choose an
initial set of models, and then estimate the parameters of each model with
maximum likelihood. Use classical methods to estimate parameter uncertainty,
given the truth of the model. Determine a probability weight for each model
using the Bayes Information Criterion (BIC), and use these weights to average the
model results.
16 The first of the above methods involves standard classical statistical methods and captures
17 sampling uncertainty conditional on the truth of the model used. The other methods are "exotic"
18 in the sense that they attempt to capture uncertainty that is not conditional on the truth of a given
19 model. In this response document, EPA has not applied such methods, but recognizes that
20 quantitative uncertainty analysis is possible in these cases.
21 In contrast to cognitive uncertainty, Volitional uncertainty concerns uncertainty regarding
22 choices on the best course of action to take; volitional uncertainty cannot be analyzed by
23 sampling from a probability distribution and, thus, is not amenable to a complete quantitative
24 uncertainty analysis. Some of the choices made in TCDD dose-response assessment that are
25 volitional include: choice of occupational cohort data set or bioassay data set; choice of PODs
26 (e.g., EDoi, ED0s, and EDi0); choice of species, strain, or sex within an animal bioassay; and
27 choice of dose metric (e.g., administered doses, blood concentrations, lipid-adjusted serum
28 concentrations). These volitional uncertainties cannot be quantified by sampling an input
29 distribution. However, EPA believes that NAS was requesting that dose-response modeling
30 results be shown for specific choices of interest to TCDD assessment. To this end, for the cancer
31 dose-response modeling, BMDLs are reported for 1, 5, and 10% extra risk levels, and a
32 comparison is provided of linear and nonlinear dose-response assessments for cancer. For the
33 noncancer dose-response modeling, different model forms are run and contrasted. Finally,
34 TCDD kinetic doses from the Emond et al. (2005, 2006) PBPK model that is primarily used in
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 lii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
the technical analysis in this document are compared with those predicted by the Aylward et al.
(2005a) model.
Uncertainty quantification is an emerging area in science. There are many examples of
highly vetted and peer-reviewed uncertainty analyses based on structured expert judgment.
Under this process, experts in effect synthesize a wide diversity of information in generating
their subjective probability distributions. Where considerable data exist for an environmental
pollutant, such as for the well-studied TCDD, it is natural to ask whether these extensive data can
be leveraged more directly in uncertainty quantification. This is an area where research could be
focused. Additional research topics relevant to dioxin that could further inform health
assessments include population variability of biokinetic constants and threshold mechanisms for
the mass action model. Further data and improved methodologies in these areas, combined with
developments illustrated elsewhere in this report, will help reduce or better quantify uncertainties
and strengthen EPA's understanding of potential health implications of environmental TCDD
exposures.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 liii DRAFT—DO NOT CITE OR QUOTE
-------
1 Table ES-1. Comparison of lipid-adjusted serum concentrations, fat
2 concentrations, risk specific dose estimates and equivalent oral slope factors
3 based on upper 95th percentile estimate of regression coefficienta of all fatal
4 cancers reported by Cheng et al. (2006) for risk levels of 1 x 10~3,1 x 10~4,
5 lx 10"5,1 x 10"6, and 1 x 10"7
6
Risk level
(RL)
AUCrl,
(ppt-yr)
FATrL
(ng/kg)
Risk specific dose
(Drl) (ng/kg-day)
Equivalent oral slope
factors (OSFrl) per
(mg/kg-day)
1 x 1(T3
1.26 x 10~3
1.803 x 101
2.73 x 10~3
3.7 x 105
1 x l(T4
1.26 x 102
1.803 x 10
1.23 x 10-4
8.1 x io5
1 x 10~5
1.26 x 101
1.803 x 10_1
8.57 x 10~6
1.2 x 106
1 x l(T6
1.26 x 10
1.803 x 10~2
7.77 x 10~7
1.3 x 106
1 x 10~7
1.26 x KT1
1.803 x 10-3
7.62 x 10-8
1.3 x 106
7
8 aBased on regression coefficient of Cheng et al. (2006; Table III), excluding observations in the upper 5% range
9 (>252,950 ppt-year lipid adjusted serum TCDD) of the exposures; where reported (3 = 3.3 x 10 6 ppt-years and
10 standard error =1.4 / 10 6, Upper 95th percentile estimate of regression coefficient (p95) calculated to be:
11 6.04 x 10~6 = (3.3 x 10~6) + 1.96 x (1.4 x 10~6).
-------
1 Table ES-2. Tumor points of departure and oral slope factors using blood
2 concentrations
3
Study
Sex/species: tumor sites
HMD l,i iin'
(ng/kg-day)
OSF
(per mg/kg-day)
NTP, 1982
Male mice: liver adenoma and carcinoma,
lung
1.1E-03
9.4E+6
Toth et al., 1979
Male mice: liver tumors
1.9E-03
5.2E+6
NTP, 1982
Female mice: liver adenoma and carcinoma,
thyroid adenoma, subcutaneous
fibrosarcoma, all lymphomas
5.3E-03
1.9E+6
NTP, 1982
Female rats: liver neoplasitc nodules, liver
adenoma and carcinoma, adrenal cortex
adenoma or carcinoma, thyroid follicular cell
adenoma
5.7E-03
1.8E+6
Kociba et al.,
1978
Female rats: liver adenoma carcinoma, oral
cavity, lung
7.3E-03
1.4E+6
NTP, 1982
Male rats: thyroid follicular cell adenoma,
adrenal cortex adenoma
9.6E-03
1.0E+6
NTP, 2006
Female rats: liver cholangiocarcinoma,
hepatocellular adenoma, oral mucosa
squamous cell carcinoma, lung cystic
keratinizing epithelioma, pancreas adenoma,
carcinoma
2.3E-02
4.4E+5
Kociba et al.,
1978
Male rats: adrenal cortex adenoma, tongue
carcinoma, nasal/palate carcinoma
3.1E-02
3.2E+5
4
5 "BMDL|[|.[,s are from the multiple tumor analyses, with the exception of Toth et al. (1979) which is the result of
6 modeling a single tumor site.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 lv DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
No
Yes
No
Study
in peer-reviewed
_ literature? _
Yes
^ Exposure
primarily to TCDD
and quantified?
No
Yes
Long-term
exposures and
latency information
available for cancer
^Xassessmenf?/
Exposure \
windows and
latency information
available for RfD
N\assessment2/
No
No
Yes
Yes
Most
elements of the five
considerations
satisfied?
List of available epidemiologic studies on TCDD and DLCs
Key study included
for TCDD cancer
dose-response
assessment
Study excluded
from TCDD
dose-response
assessment
Key study included
for TCDD noncancer
dose-response
assessment
Evaluate study using five considerations:
Methods used to ascertain health outcomes are clear and unbiased.
Confounding exposures or other sources of bias are controlled for.
Dose-response is apparent between TCDD and adverse health effect(s).
Exposure methods are described including exposure duration and latency.
Statistical precision and power are sufficient.
Figure ES-1. EPA's process to evaluate available epidemiologic studies using
study inclusion criteria for use in the dose-response analysis of TCDD. EPA
applied its TCDD-specific epidemiologic study inclusion criteria to all studies published
on TCDD and DLCs. The studies were initially evaluated using five considerations
regarded as providing the most relevant kind of information needed for quantitative
human health risk analyses. For each study that satisfied most of these considerations
and was published in the peer-reviewed literature, EPA then examined whether the
exposures were primarily to TCDD and if the TCDD exposures could be quantified so
that dose-response analyses could be conducted. Finally, EPA required that the effective
dose and oral exposure be estimable: (1) for cancer, information is required on long-term
exposures, (2) for noncancer, information is required regarding the appropriate time
window of exposure that is relevant for a specific, nonfatal health endpoint, and (3) for
all endpoints, the latency period between TCDD exposure and the onset of the effect is
needed. Only studies meeting these criteria were included in EPA's TCDD dose-
response analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 lvi DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
Study
in peer-reviewed
literature?
Yes
No
Lowest
dose tested for
cancer endpoint <1
s. ijg/kg-day? ^
Lowest dose
tested for noncancer
endpoint <30
ng/kg-day?
No
No
Yes
Yes
No
No
Yes
Oral
exposure to TCDD
only with purity
specified?
Yes^T^^
Were
elements of the four
considerations
satisfied?
Study excluded
from TCDD
dose-response
assessment
List of available in vivo mammalian bioassay studies on TCDD
Key study included
for TCDD cancer and/or noncancer
dose-response assessment
Evaluate study further using four considerations:
Strain, gender, and age of test species is identified.
Testing protocol, including duration and timing of dosing is clear.
Study design is consistent with standard toxicological practices.
Magnitude of animal responses is outside range of normal variability.
Figure ES-2. EPA's process to evaluate available animal bioassay studies using
study inclusion criteria for use in the dose-response analysis of TCDD. EPA
evaluated all available in vivo mammalian bioassay studies on TCDD. Studies had to be
published in the peer-reviewed literature. Next, to ensure working in the low-dose range
for TCDD dose-response analysis, EPA applied dose requirements to the lowest tested
average daily doses in each study, with specific requirements for cancer (<1 (j,g/kg-day),
and noncancer (<30 ng/kg-day) studies. Third, EPA required that the animals were
exposed via the oral route to only TCDD and that the purity of the TCDD was specified.
Finally, the studies were evaluated using four considerations regarded as providing the
most relevant kind of information needed for quantitative human health risk analyses
from animal bioassay data. Only studies meeting all of these criteria and considerations
were included in EPA's TCDD dose-response analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 lvii DRAFT—DO NOT CITE OR QUOTE
-------
CD
a:
<
O
o
CO
~ ^ 0
c
0)
E
Q.
O
0)
>
0)
Q
2.1
cc
3
¦a
n
s>
0
c
ro
~ a-
CM
T—
O
o
+
+
LU
LU
o
o
T—
T—
o
o
+
LU
O
(p-6>j/6u '8sop ju0|BAmb0-UBiunH)
ajey uoijssBui aaoj.-8'Z'e'Z
(y) ezoos "lets nag
(lAl) 9002 "|B )9 n
(y) 9661- ' |B }9Z)UBips
(y) ooos'ib }9 u|ujv
(y) 1.003 "|B)9 0>|BSHO
(y) 9002 ' |B)9 U9U!H9!|/\|
(y) 1-003 ' |B)9 !>|SMO>|JB|/\|
(lAl) q'B800S 'ZOOS ' |B }9 J9||9»
(y) 1-002 ' |B)9 U9U!BHB»
(y) 2002 "|B}9 0f0H
(y)6Z6l. ' |B)9 ABJjniAl
(y) zoos "ib}9 ins
(y) 2002 JnL|iB|AI PUB no)B"|
(y) 9003 'dlN
(lAl) 6Z6I. "|B)9 moi
(y) 8Z6I. "|B)9 Bqpo»
(lAl) 2861- 'dlN
(y) 1861 ' |B)9 !UO)UBQ
(y) S66I- "|B)9 IIBM9S
(y) BS66I. ' |B }9 U9|96j|g UBA
(lAl) 8002 ' |B }9 Z0!M0|B|ms
(y) S66I. "|B)9 09S
(y) ooos'iBtggjoHBj
(y) Z002'|B}9 nno
(lAl) 986U "|B)9 9}NM
(y) SOOS' IB }9 uouojq
(y) Z66I. "ib)9 n
(H) 8002 ' |B)9 !||9JB000|/\|
(H) 8002 ' |B)9 !||9JB00Bg
(H) K)02'|B}9 Bnsnn|B|v
T—
CM
CO
o
O
o
o
LU
LU
LU
LU
o
o
o
o
T—
T—
T—
T—
Q
P
U
H
o
u
5
o
a
*
co
c/)
W
5
£X)
77;/s document is a draft for review purposes only and does not constitute Agency policy m
1/15/10 lviii DRAFT: DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
No
^ Is the ^
endpoint observed
\at the LOAELV
Is the BMDL less
than the LOAEL?
No
No
Yes
Yes
Is the endpoint less
than the minimum
LOAEL X 100?
No
Yes
Is the
endpoint under consideration
toxicologically
relevant?
Exclude endpoint
as a POD candidate
Include NOAEL/LOAEL/BMDL
as a POD candidate
List of key noncancer animal studies for
quantitative dose-response analysis of TCDD
Estimate a Human Equivalent Dose (HED)
corresponding to each blood concentration NOAEL, LOAEL, or BMDL
using the Emond human PBPK model
Determine NOAEL, LOAEL, and BMDL (if possible) human equivalent dose
(HED) based on 1s,-order body burden for each study/endpoint combination
Determine a NOAEL, LOAEL, and BMDL (if possible) for each
study/endpoint combination, based on blood concentrations from the
Emond rodent PBPK model
Figure ES-4. EPA's process to select and identify candidate PODs from key animal
bioassays for use in noncancer dose-response analysis of TCDD. For each noncancer
endpoint found in the studies that qualified for TCDD dose-response assessment using
the study inclusion criteria, EPA first determined if the endpoint was toxicologically
relevant. If so, EPA determined the NOAEL, LOAEL, and BMDL Human Equivalent
Dose (HED) based on lst-order body burdens for each endpoint. These potential PODs
were examined for statistical relevance and included when the endpoint was observed at
the LOAEL. If the BMDL was less than the LOAEL, and if the endpoint was less than
the minimum LOAEL x 100, EPA then calculated NOAELs, LOAELs, or BMDLs based
on blood concentrations from the Emond rodent PBPK model. Then, for all of the
candidate PODs, HEDs were estimated using the Emond human PBPK model. Finally,
the lowest group of the toxicologically relevant candidate PODs was selected for final
use in derivation of an RfD.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 lix DRAFT—DO NOT CITE OR QUOTE
-------
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
1. INTRODUCTION
Dioxins and dioxin-like compounds (DLCs), including polychlorinated dibenzo-dioxins,
polychlorinated dibenzofurans, and polychlorinated biphenyls are structurally and
toxicologically related halogenated dicyclic aromatic hydrocarbons.4 Dioxins and DLCs are
released into the environment from several industrial sources such as chemical manufacturing,
combustion, and metal processing; from individual activities including the burning of household
waste; and from natural processes such as forest fires and volcanic activity. Dioxins and DLCs
are widely distributed throughout the environment and typically occur as chemical mixtures.
Additionally, they do not readily degrade; therefore, levels persist in the environment, build up in
the food chain, and accumulate in the tissues of animals. Human exposure to these compounds
occurs primarily through the ingestion of contaminated foods (Lorber et al., 2009).
The health effects from exposures to dioxins and DLCs have been documented
extensively in epidemiological and toxicological studies. 2,3,7,8-Tetrachlorodibenzo-p-dioxin
(TCDD) is one of the most toxic members of this class of compounds and has a robust
toxicological database. Characterization of TCDD toxicity is critical to the risk assessment of
mixtures of dioxins and DLCs because it has been selected repeatedly as the "index chemical" to
serve as the basis for standardization of the toxicity of components in a mixture of dioxins and
DLCs. The dose-response information for TCDD is used to evaluate risks from exposure to
mixtures of DLCs (Van den Berg et al., 1998, 2006; also see the World Health Organization's
Web site for the dioxin toxicity equivalence factors [TEFs]),5 therefore, it is imperative to
correctly assess the dose response of TCDD and understand the uncertainties and limitations
therein.
In 2003, the U.S. Environmental Protection Agency (EPA) produced an external review
draft of the multiyear comprehensive reassessment of dioxin exposure and human health effects
entitled, Exposure and Human Health Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin
(TCDD) and Related Compounds (U.S. EPA, 2003). This draft report, herein called the "2003
Reassessment," consisted of (1) a scientific review of information relating to sources of and
exposures to TCDD, other dioxins, and DLCs in the environment; (2) detailed reviews of
4For further information on the chemical structures of these compounds, see U.S. EPA (2003, 2008a).
5Available at http://www.who.int/ipcs/assessment/tef_update/en/.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
scientific information on the health effects of TCDD, other dioxins, and DLCs; and (3) an
integrated risk characterization for TCDD and related compounds.
In 2004, EPA asked the National Research Council of the National Academy of Sciences
(NAS) to review the 2003 Reassessment. The NAS Statement of Task was as follows
The National Academies' National Research Council will convene an expert committee that will
review EPA's 2003 draft reassessment of the risks of dioxins and dioxin-like compounds to
assess whether EPA's risk estimates are scientifically robust and whether there is a clear
delineation of all substantial uncertainties and variability. To the extent possible, the review will
focus on EPA's modeling assumptions, including those associated with the dose-response curve
and points of departure; dose ranges and associated likelihood estimates for identified human
health outcomes; EPA's quantitative uncertainty analysis; EPA's selection of studies as a basis
for its assessments; and gaps in scientific knowledge. The study will also address the following
aspects of EPA's 2003 Reassessment: (1) the scientific evidence for classifying dioxin as a
human carcinogen; and (2) the validity of the nonthreshold linear dose-response model and the
cancer slope factor calculated by EPA through the use of this model. The committee will also
provide scientific judgment regarding the usefulness of toxicity equivalence factors (TEFs) in
the risk assessment of complex mixtures of dioxins and the uncertainties associated with the use
of TEFs. The committee will also review the uncertainty associated with the 2003
Reassessment's approach regarding the analysis of food sampling and human dietary intake
data, and, therefore, human exposures, taking into consideration the Institute of Medicine's
report Dioxin and Dioxin-Like Compounds in the Food Supply: Strategies to Decrease
Exposure. The committee will focus particularly on the risk characterization section of EPA's
2003 Reassessment report and will endeavor to make the uncertainties in such risk assessments
more fully understood by decision makers. The committee will review the breadth of the
uncertainty and variability associated with risk assessment decisions and numerical choices,
including, for example, modeling assumptions, including those associated with the dose-
response curve and points of departure. The committee will also review quantitative uncertainty
analyses, as feasible and appropriate. The committee will identify gaps in scientific knowledge
that are critical to understanding dioxin reassessment (NAS, 2006a, p. 43, Box 1-1).
In 2006, the NAS published its review of EPA's 2003 Reassessment entitled Health Risks from
Dioxin and Related Compounds: Evaluation of the EPA Reassessment (NAS, 2006a).
1.1. SUMMARY OF KEY NAS (2006a) COMMENTS ON DOSE-RESPONSE
MODELING IN THE 2003 REASSESSMENT
While recognizing the effort that EPA expended to prepare the 2003 Reassessment, the
NAS committee identified three key areas that they believe require substantial improvement to
support a scientifically robust risk assessment. These three key areas are
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
• transparency and clarity in selection of key data sets for analysis;
• justification of approaches to dose-response modeling for cancer and noncancer
endpoints; and
• transparency, thoroughness, and clarity in quantitative uncertainty analysis.
In their Public Summary, the NAS made the following overall recommendations to aid
EPA in addressing their key concerns:
• EPA should compare cancer risks by using nonlinear models consistent with a receptor
mediated mechanism of action and by using epidemiological data and the new National
Toxicology Program (NTP) animal bioassay data (NTP, 2006). The comparison should
include upper and lower bounds, as well as central estimates of risk. EPA should clearly
communicate this information as part of its risk characterization (NAS, 2006a, p. 9).
• EPA should identify the most important data sets to be used for quantitative risk
assessment for each of the four key end points (cancer, immunotoxicity, reproductive
effects, and developmental effects). EPA should specify inclusion criteria for the studies
(animal and human) used for derivation of the benchmark dose (BMD) for different
noncancer effects and potentially for the development of RfD (reference dose) values and
discuss the strengths and limitations of those key studies; describe and define
(quantitatively to the extent possible) the variability and uncertainty for key assumptions
used for each key end-point-specific risk assessment (choices of data set, POD [point of
departure], model, and dose metric); incorporate probabilistic models to the extent
possible to represent the range of plausible values; and assess goodness-of-fit of
dose-response models for data sets and provide both upper and lower bounds on central
estimates for all statistical estimates. When quantitation is not possible, EPA should
clearly state it and explain what would be required to achieve quantitation (NAS, 2006a,
p. 9).
• When selecting a BMD as a POD, EPA should provide justification for selecting a
response level (e.g., at the 10%, 5%, or 1% level). In either case, the effects of this
choice on the final risk assessment values should be illustrated by comparing point
estimates and lower bounds derived from selected PODs (NAS, 2006a, p. 9).
• EPA should continue to use body burden as the preferred dose metric but should also
consider physiologically based pharmacokinetic modeling as a means to adjust for
differences in body fat composition and for other differences between rodents and
humans (NAS, 2006a, p. 9).
• Although EPA addressed many sources of variability and uncertainty qualitatively, the
committee noted that the 2003 Reassessment would be substantially improved if its risk
characterization included more quantitative approaches. Failure to characterize
variability and uncertainty thoroughly can convey a false sense of precision in the
conclusions of the risk assessment (NAS, 2006a, p. 5).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Importantly, the NAS encouraged EPA to calculate an RfD as the 2003 Reassessment
does not contain an RfD derivation. The committee suggested that
... estimating an RfD would provide useful guidance to risk managers to help
them (1) assess potential health risks in that portion of the population with intakes
above the RfD, (2) assess risks to population subgroups, such as those with
occupational exposures, and (3) estimate the contributions to risk from the major
food sources and other environmental sources of TCDD, other dioxins, and DLCs
for those individuals with high intakes (NAS, 2006a, p. 6).
The NAS made many thoughtful and specific recommendations throughout their review;
additional NAS recommendations and comments pertaining to the dose-response assessment of
TCDD will be presented and addressed in various sections throughout this document.
1.2. EPA'S SCIENCE PLAN
In response to recommendations presented in the 2006 NAS review, EPA Administrator
Jackson released EPA's "Science Plan for Activities Related to Dioxins in the Environment'
("Science Plan") on May 26, 2009.6 There are five key components of the Science Plan that
pertain to EPA's response to the NAS comments on TCDD dose-response assessment:
1. EPA will release a draft report that responds to the recommendations and comments
included in the NAS review of EPA's 2003 Reassessment.
• EPA's National Center for Environmental Assessment in the Office of Research and
Development, will prepare a limited response to key comments and recommendations
in the NAS report (draft response to comments report).
• The draft response will focus on dose-response issues raised by the NAS and will
include an analysis of relevant new key studies.
• The draft response will be provided for public review and comment and independent
external peer review.
• The draft response will also include an evaluation of some of the significant
recommendations that are difficult for EPA to address given the current state of
science, and a detailed rationale for these conclusions.
• The peer review will be conducted by EPA Science Advisory Board, an independent
review body chartered under the Federal Advisory Committee Act.
6Available at http://www.epa.gov/dioxin/scienceplan.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
• The draft response to comments report will be completed and released for public and
peer review by December 31, 2009.
2. EPA will provide the draft response to comments report for internal and external review.
3. The Science Advisory Board will review the science content of the response to comments
report.
4. EPA will review impacts of the draft response to comments report on its 2003
Reassessment.
5. EPA will release the final response to comments report and focus on completion of the
2003 Reassessment.
This document comprises EPA's report that responds to the recommendations and
comments on TCDD dose-response assessment included in the NAS review of EPA's 2003
Reassessment. This document focuses on TCDD only.
1.3. OVERVIEW OF EPA'S RESPONSE TO NAS (2006a) "HEALTH RISKS FROM
DIOXIN AND RELATED COMPOUNDS: EVALUATION OF EPA's 2003
REASSESSMENT"
In their key recommendations, the NAS commented that EPA should thoroughly justify
and communicate approaches to dose-response modeling, increase transparency in the selection
of key data sets, and improve the communication of uncertainty (particularly quantitative
uncertainty). They also encouraged EPA to calculate an RfD. These main areas of improvement
refer to issues specifically related to TCDD dose-response assessment (and uncertainty analysis);
therefore, as noted in the Science Plan, EPA's response to the NAS is particularly focused on
these issues.
EPA thoroughly considered the recommendations of the NAS and responds with
scientific and technical evaluation of TCDD dose-response data via:
• an updated literature search that identified new TCDD dose-response studies (see Section
2);
• a kickoff workshop that included the participation of external experts in TCDD health
effects, toxicokinetics, dose-response assessment and quantitative uncertainty analysis;
these experts discussed potential approaches to TCDD dose-response assessment and
considerations for EPA's response to NAS (see Appendix A);
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
• detailed study inclusion criteria and processes for the selection of key studies (see Section
2.3) and epidemiologic and animal bioassay data for TCDD dose-response assessment
(see Section 2.4.1/Appendix B and Section 2.4.2, respectively);
• kinetic modeling to quantify appropriate dose metrics for use in TCDD dose-response
assessment (see Section 3);
• dose-response modeling for all appropriate noncancer and cancer data sets (see
Section 4.2/Appendix E and Section 5.2.3/Appendix F, respectively);
• thorough and transparent evaluation of the selected TCDD data for use in the derivation
of an RfD and an oral slope factor (OSF) (see Sections 4.2 and 5.2.3, respectively);
• the development of an RfD (see Section 4.3);
• the development of a revised OSF (see Section 5.3) with an updated cancer weight of
evidence determination for TCDD based on EPA's 2005 Cancer Guidelines (U.S. EPA,
2005) (see Section 5.1.2);
• consideration of nonlinear dose-response approaches for cancer, including illustrative
RfDs for cancer precursor events and tumors (see Section 5.2.3.4) ; and
• discussion of the feasibility and utility of quantitative uncertainty analysis for TCDD
dose-response assessment (see Section 6).
Each of these activities is described in detail in subsequent sections of this document.
It should be noted that three separate EPA activities address additional TCDD issues,
specifically related to the application of dioxin TEFs and to TCDD and DLC background
exposure levels. Information on the application of the dioxin TEFs is published elsewhere by
EPA for both ecological (U.S. EPA, 2008a) and human health risk assessment (U.S. EPA,
2009a). As a consequence, EPA does not directly address TEFs herein, but makes use of the
concept of toxicity equivalence7 as applicable to the analysis of exposure dose in
epidemiological studies. Furthermore, this document does not address the NAS
recommendations pertaining to the assessment of human exposures to TCDD and other dioxins.
Information on updated background levels of dioxin in the U.S. population has been recently
reported (Lorber et al., 2009).
'Toxicity equivalence (TEQ) is the product of the concentration of an individual DLC in an environmental mixture
and the corresponding TCDD TEF for that compound. These products are summed to yield the TEQ of the mixture.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
1.3.1. TCDD Literature Update
EPA has developed a literature database of peer-reviewed studies on TCDD toxicity,
including in vivo mammalian dose-response studies and epidemiological studies. An initial
literature search for studies published since the 2003 Reassessment was conducted by the U.S.
Department of Energy's Argonne National Laboratory (ANL) through an Interagency Agreement
with EPA. ANL used the online National Library of Medicine database (PubMed) and identified
studies published between the year 2000 and October 31, 2008. Supporting references published
since the release of the 2003 Reassessment were also identified. Supporting studies were
classified as studies pertaining to TCDD kinetics, TCDD mode-of-action, in vitro TCDD studies,
and TCDD risk assessment approaches. The literature search strategy explicitly excluded studies
addressing (1) analytical/detection data and cellular screening assays; (2) environmental fate,
transport and concentration data; (3) dioxin-like compounds and toxic equivalents;
(4) nonmammalian dose-response data; (5) human exposure analyses only, including body
burden data; and (6) combustor or incinerator or other facility-related assessments absent
primary dose-response data. EPA published the initial literature search results in the Federal
Register on November 24, 2008 (73 FR 70999; November 24, 2008) and invited the public to
review the list and submit additional peer-reviewed in vivo mammalian dose-response studies for
TCDD, including epidemiological studies that were absent from the list (U.S. EPA, 2008b). The
literature search results and subsequent submissions were used during a 2009 scientific
workshop, which was open to the public and featured a panel of experts on TCDD toxicity and
dose-response modeling (discussed below). Additional studies identified during the workshop
and those collected by EPA scientists during the development of this report through October
2009 have been incorporated into the final set of studies for TCDD dose-response assessment.
1.3.2. EPA's 2009 Workshop on TCDD Dose Response
To assist EPA in responding to the NAS, EPA, and ANL convened a scientific workshop
(the "Dioxin Workshop") on February 18-20, 2009, in Cincinnati, Ohio. The goals of the
Dioxin Workshop were to identify and address issues related to the dose-response assessment of
TCDD and to ensure that EPA's response to the NAS focused on the key issues and reflected the
most meaningful science. The Dioxin Workshop included seven scientific sessions: quantitative
dose-response modeling issues, immunotoxicity, neurotoxicity and nonreproductive endocrine
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
effects, cardiovascular toxicity and hepatotoxicity, cancer, reproductive and developmental
toxicity, and quantitative uncertainty analysis of dose-response. During each session, EPA asked
a panel of expert scientists to perform the following tasks:
• Identify and discuss the technical challenges involved in addressing the NAS comments
related to the dose-response issues within each specific session topic and the TCDD
quantitative dose-response assessment.
• Discuss approaches for addressing the key NAS recommendations.
• Identify important published, independently peer-reviewed literature—particularly
studies describing epidemiological studies and in vivo mammalian bioassays expected to
be most useful for informing EPA's response.
The sessions were followed by open comment periods during which members of the
audience were invited to address the expert panels. The session's Panel Co-chairs were asked to
summarize and present the results of the panel discussions—including the open comment
periods. The summaries incorporated points of agreement as well as minority opinions. Final
session summaries were prepared by the session Panel Co-chairs with input from the panelists,
and they formed the basis of a final workshop report (U.S. EPA, 2009c, Appendix A of this
report). Because the sessions were not designed to achieve consensus among the panelists, the
summaries do not necessarily represent consensus opinions; rather reflect the core of the panel
discussions. Some of the key discussion points from the workshop that influenced EPA's
development of this document are listed below (see Appendix A for detail):
• In the development of study selection criteria, more relevant exposure-level (i.e., dose)
decision points using tissue concentrations could be defined.
• A linear approach to body-burden estimation, which was utilized in the 2003
Reassessment (U.S. EPA, 2003), does not fully consider key toxicokinetic issues related
to TCDD—e.g., sequestration in the liver and fat, age-dependent elimination, and
changing elimination rates over time. Thus, kinetic/mechanistic modeling could be used
to quantify tissue-based metrics. In considering human data, lipid-adjusted serum levels
may be preferable over body burden, although the assumptions used in the back
calculation of the body burden in epidemiologic cohorts are of concern. In considering
rat bioassay data, lipid-adjusted body-burden estimates may be preferable.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
• New epidemiological studies on noncancer endpoints have been published since the 2003
Reassessment that may need to be considered (e.g., thyroid dysfunction literature from
Wang et al. [2005] and Baccarelli et al. [2008]).
• The 1% of maximal response (ED0i) that was utilized in the 2003 Reassessment has not
typically been used in dose-response assessment. Some alternative ideas were as follows:
(1) the POD should depend on the specific endpoint; (2) for continuous measures, the
benchmark response (BMR) could be based on the difference from control and consider
the adversity level; and (3) for incidence data, the BMR should be set to a fixed-risk
level.
• The quantitative dose-response modeling for cancer could be based on human or animal
data. There are new publications in the literature for four epidemiological cohort studies
(Dutch cohort, NIOSH cohort, BASF accident cohort, and Hamburg cohort). The
increase in total cancers could be considered for modeling human cancer data. However,
non-Hodgkin's lymphoma and lung tumors are the main TCDD-related cancer types seen
from human exposure. In reviewing the rat data, the NTP (2006) data sets are new and
can be modeled. Although the liver and lungs are the main target organs, modeling all
cancers, as well as using tumor incidence in lieu of individual rats as a measure, should
be considered.
• Both linear and nonlinear model functions should be considered in the cancer
dose-response analysis because there are data and rationales to support use of either
below the POD.
• For quantitative uncertainty analysis, consider the impacts of choices among plausible
alternative data sets, dose metrics, models, and other more qualitative choices. Issues to
consider include how much difference these choices make and, also, how much relative
credence should be put toward each alternative as a means to gauge and describe the
landscape of imperfect knowledge with respect to possibilities for the true dose response.
This may be difficult to do quantitatively because the factors are not readily expressed as
statistical distributions. However, the rationale for accepting or questioning each
alternative in terms of the available supporting evidence, contrary evidence, and needed
assumptions, can be delineated.
1.3.3. Overall Organization of EPA's Response to NAS Recommendations
The remainder of this document is divided into five sections that address the three
primary areas of concern resulting from the NAS (2006a) review. Section 2 describes EPA's
approach to the recommendation for transparency and clarity during selection of key data
sets—including criteria for the selection of key dose-response studies, evaluations of the
important epidemiologic studies and animal bioassays, and a summary of the key studies used
for subsequent dose-response modeling. Sections 3, 4, and 5 present EPA's response to the NAS
recommendation to better justify the approaches used in dose-response modeling of TCDD.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-9 DRAFT—DO NOT CITE OR QUOTE
-------
1 Section 3 discusses the toxicokinetic modeling EPA conducted to support the dose-response
2 analyses. Section 4 presents EPA's approach to noncancer data set selection, dose-response
3 modeling, derivation of an RfD for TCDD, and contains a qualitative discussion of the
4 uncertainties associated with the RfD. Section 5 presents an updated cancer weight-of-evidence
5 summary, EPA's approach to cancer data set selection, dose-response modeling, derivation of an
6 OSF for TCDD, and a qualitative discussion of the uncertainties associated with the OSF,
7 including an evaluation of alternative approaches to cancer assessment of TCDD. Finally,
8 Section 6 discusses the feasibility of conducting a quantitative uncertainty analysis of TCDD
9 dose response.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 1-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
2. TRANSPARENCY AND CLARITY IN Till SELECTION OF KEY DATA SETS
FOR DOSE-RESPONSE ANALYSIS
This section focuses on addressing transparency and clarity in the study selection process
and on identifying key data sets for 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) dose-response
analysis. Section 2.1 summarizes the National Academy of Sciences (NAS) committee's
comments specifically regarding this issue. Section 2.2 presents U.S. Environmental Protection
Agency's (EPA's) response to those comments and describes EPA's approach to ensuring
transparency and clarity in selection of studies for subsequent dose-response analyses.
Section 2.3 describes the TCDD-specific study inclusion criteria and evaluation process EPA
used in this document for determining the eligibility of both epidemiologic and experimental
animal studies for TCDD dose-response analysis. Section 2.4.1 summarizes epidemiologic data
and evaluates the suitability of these data for TCDD dose-response analyses. Section 2.4.2
summarizes animal bioassay data that have met the study inclusion criteria for TCDD
dose-response assessment. Finally, Section 2.4.3 identifies key TCDD epidemiologic and animal
bioassay studies that were determined using the study inclusion criteria. Then it previews the
evaluation processes used to further identify study/endpoint combination data sets for developing
TCDD toxicity values for noncancer (see Section 4) and cancer (see Section 5) effects.
2.1. SUMMARY OF NAS COMMENTS ON TRANSPARENCY AND CLARITY IN
THE SELECTION OF KEY DATA SETS FOR DOSE-RESPONSE ANALYSIS
The NAS committee proposed that EPA develop a clear and readily understandable
methodology for evaluating and including epidemiologic and animal bioassay data sets in
dose-response evaluations. The NAS committee recommended the development and application
of transparent initial criteria to judge whether or not specific epidemiologic or animal bioassay
studies be included in TCDD dose-response analysis.
Specific NAS comments on the topic of study evaluation and inclusion criteria include
EPA should specify inclusion criteria for the studies (animal and human) used for
derivation of the benchmark dose (BMD) for different noncancer effects and
potentially for the development of RfD values and discuss the strengths and
limitations of those key studies (NAS, 2006, p. 27).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-1 DRAFT—DO NOT CITE OR QUOTE
-------
1 .. .in its [EPA's] evaluation of the epidemiological literature of carcinogenicity, it
2 did not outline eligibility requirements or otherwise provide the criteria used to
3 assess the methodological quality of other included studies (NAS, 2006, p. 56.)
4 With regard to EPA's review of the animal bioassay data, the committee
5 recommends that EPA establish clear criteria for the inclusion of different data
6 sets (NAS, 2006, p. 191).
7 .. .the committee expects that EPA could substantially improve its assessment
8 process if it more rigorously evaluated the quality of each study in the database
9 (NAS, 2006, p. 56).
10 EPA could also substantially improve the clarity and presentation of the risk
11 assessment process for TCDD.. by using a summary table or a simple summary
12 graphical representation of the key data sets and assumptions... (NAS, 2006,
13 p. 56).
14
15 2.2. EPA'S RESPONSE TO NAS COMMENTS ON TRANSPARENCY AND CLARITY
16 IN THE SELECTION OF KEY DATA SETS FOR DOSE-RESPONSE ANALYSIS
17 EPA agrees with the NAS committee regarding the need for a transparent and clear
18 process for selecting studies and key data sets for TCDD dose-response analyses. The
19 delineation of the study selection process and decisions regarding key data sets will facilitate
20 communication regarding critical decisions made in the TCDD dose-response assessment. In
21 keeping with the NAS committee's recommendation to use a transparent process and improve
22 clarity and presentation of the risk assessment process for TCDD, Figure 2-1 overviews the
23 approach that EPA has used in this document to develop a final list of key cancer and noncancer
24 studies for quantitative dose-response analysis of TCDD. The steps in Figure 2-1 are further
25 explained below.
26
27 Literature search for in vivo mammalian and epidemiologic TCDD studies
28 (2000-2008): EPA conducted a literature search to identify peer-reviewed, dose-response
29 studies for TCDD that have been published since the 2003 Reassessment. This search
30 included in vivo mammalian and epidemiological studies of TCDD from 2000 to 2008.
31 Additional details describing the conduct of this literature search are presented in
32 Section 1.3.1 of this document.
33 Federal Register Notice—Web publication of literature search for public comment:
34 In November 2008, EPA published a list of-500 citations from results of this literature
35 search (U.S. EPA, 2008b) and invited the public to review this preliminary list of
36 dose-response citations for use in TCDD dose-response assessment. EPA requested that
37 interested parties identify and submit peer-reviewed studies for TCDD that were absent
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
from this list. All additional studies submitted to EPA were included in the final TCDD
literature database considered by EPA for TCDD dose-response analysis.
Initial study inclusion criteria development for TCDD in vivo mammalian
bioassays: EPA developed an initial set of draft criteria for evaluating the extensive
TCDD database of in vivo mammalian bioassays. These initial inclusion criteria had
three purposes. First, they provided a transparent and rigorous evaluation of the scientific
quality of each study in EPA's database, a deficiency in the 2003 Reassessment identified
by the NAS committee. Second, given the vast TCDD mammalian bioassay database,
they provided a transparent method for initially screening studies to be considered for
TCDD dose-response analyses. Third, they served as a starting point for discussions of
study inclusion criteria by expert panelists who were convened by EPA for its scientific
workshop on TCDD dose-response analysis (the Dioxin Workshop), described next (also
see the workshop report in Appendix A).
Dioxin Workshop and expert refinement of TCDD in vivo mammalian bioassay
inclusion criteria: In February 2009, EPA convened "A Scientific Workshop to Inform
EPA's Response to NAS Comments on the Health Effects of Dioxin in EPA's 2003
Dioxin Reassessment." The goals of this 3-day public and scientific workshop were to
identify and address issues related to the dose-response assessment of TCDD. Sessions at
the workshop examined toxicities associated with TCDD, issues related to developing
dose-response estimates based on these data and associated uncertainties. At the
workshop, EPA presented the draft set of study inclusion criteria for evaluating the
extensive TCDD in vivo mammalian bioassay literature and asked workshop panelists to
discuss these criteria and make recommendations for their revision. Further details on
this workshop are presented in Section 1.3.2 of this document, and the complete report
from this workshop is available in Appendix A, including detailed summaries of the
panels' comments on the inclusion criteria in relation to the various toxic endpoints that
were discussed.
Final development of inclusion criteria for TCDD in vivo mammalian studies: Based
on discussions at the Dioxin Workshop, the initial draft inclusion criteria for evaluating
the TCDD mammalian bioassay literature were revised and are presented in Section 2.3.2
(see Figure 2-3). An initial criterion is that studies for consideration must be publically
available and published in a peer-reviewed scientific journal. Because the methodology
EPA uses to develop reference doses (RfDs) and cancer oral slope factors (OSFs) relies
on identification of studies reporting potential adverse effects at low doses (relative to the
overall database), another important criterion shown in Section 2.3.2 identifies a
maximum value for the lowest TCDD dose tested in a bioassay. This maximum value
was used to eliminate those studies that could not be selected for development of an RfD
or an oral slope factor because tested doses were too high relative to other TCDD
bioassays.
Development of inclusion criteria for epidemiologic studies: Following the Dioxin
Workshop, EPA determined that an evaluation process was also needed for inclusion of
epidemiologic studies for TCDD dose-response assessment. These criteria were
developed and are detailed in Section 2.3.1 (see Figure 2-2). Analogous to animal
bioassay data, epidemiologic studies for consideration must also be publically available
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
and published in a peer-reviewed scientific journal. In addition to assessing the
methodological considerations relative to epidemiologic cohorts and studies (e.g.,
statistical power and precision of estimates, consideration of latency periods), a key
consideration for use of a data set in TCDD dose-response modeling is characterization
of the exposure assessment methodology, and specifically, whether this methodology
allowed assignment of individual-level exposures within a study.
Final literature collection (October 2009): Additional literature was collected as it was
identified by EPA following the Dioxin Workshop through October 2009 to ensure the
consideration of all recently published data for this report.
Studies screened using inclusion criteria: The two sets of TCDD-specific study
inclusion criteria presented in Section 2.3 were used to evaluate all studies included in the
2003 Reassessment, studies identified in the 2000-2008 literature search, studies
identified through public comment and submission, and studies collected in 2009 as
identified by EPA during the development of this document. Section 2.4 presents results
of EPA's evaluation of epidemiologic and mammalian bioassay literature for both cancer
and noncancer endpoints.
Final list of key cancer and noncancer studies for quantitative dose-response
analysis of TCDD: Application of the study inclusion criteria concludes in Section 2.4
with development of a list of key noncancer and cancer studies that were considered for
quantitative dose-response analyses of TCDD in Sections 4 and 5, respectively. In those
sections, points of departure (PODs) are developed and evaluated for all study/endpoint
combinations from these final key study lists, and key data sets and PODs for the
development of TCDD toxicity values are identified.
2.3. STUDY INCLUSION CRITERIA FOR TCDD DOSE-RESPONSE ANALYSIS
One of the three major recommendations made by the NAS (2006a) committee was that
EPA should provide greater clarity and transparency on the selection of studies that were used in
the quantitative dose-response modeling of TCDD in the 2003 Reassessment. In this section,
EPA describes TCDD-specific study inclusion criteria that have been developed to evaluate
epidemiologic studies and animal bioassays for TCDD dose-response assessment. These criteria
reflect EPA's goal of developing an RfD and a cancer OSF for TCDD through a transparent
study selection process; they are intended to be used by EPA for TCDD dose-response
assessment only. These criteria were applied to each of the -500 studies listed in Preliminary
Literature Search Results and Request for Additional Studies on 2,3,7,8-Tetrachlorodibenzo-p-
Dioxin (TCDD) Dose-Response Studies (U.S. EPA, 2008b); studies identified and submitted by
the public and by participants in the Dioxin Workshop (U.S. EPA, 2009c); studies included in
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
the 2003 Reassessment, and other relevant published studies collected by EPA scientists through
October 2009.
EPA has undertaken different approaches for epidemiologic versus in vivo animal
bioassay study evaluation and key data set selection. The significant differences between animal
and human health effects data and their use in EPA risk assessment support development of
separate criteria for study inclusion and different approaches to study evaluation. For the vast
majority of compounds on EPA's Integrated Risk Information System (IRIS), cancer and
noncancer toxicity values have been derived using animal bioassay data; therefore, approaches to
dose-response modeling and POD selection from in vivo mammalian bioassays have been
standardized and codified (U.S. EPA, 2008c). The study criteria shown below and in Figure 2-3
for animal bioassay data reflect EPA's preferences for TCDD-specific study inclusion, some of
which are based on common practices and guidance for POD selection and RfD and OSF
derivation. Far fewer IRIS toxicity values have been derived from human data, although some
examples do exist. For example, benzene, beryllium and compounds, chromium IV, and
1,3-butadiene have RfDs, Reference Concentrations, Inhalation Unit Risks and/or OSFs all based
on occupational cohort data and the methyl mercury RfD is based on high fish consuming
cohorts (U.S. EPA, 2009b). The modeling and interpretation of such human data have been
conducted on a case-by-case basis because each cohort is uniquely defined and has its own set of
exposure conditions, significant confounders, and biases that may need to be considered in
dose-response modeling. For TCDD, not all data are from occupational cohorts, but include
cohorts exposed for relatively short time periods to high concentrations as a consequence of
industrial accidents, a scenario that has not commonly been used to establish chronic EPA
toxicity values.
Because of these differences in data characteristics, divergent selection approaches are
used in this document to present and evaluate the epidemiologic studies (see Section 2.3.1) and
the in vivo animal bioassays (see Section 2.3.2). In Section 2.4.1, all of the available
epidemiologic studies on TCDD are summarized and evaluated for suitability for dose-response
modeling using the TCDD-specific study inclusion criteria below and shown in Figure 2-2; only
studies meeting the inclusion criteria are presented as key studies in Section 2.4.3 (see Tables 2-4
and 2-5 for the cancer and noncancer endpoints, respectively). In Section 2.4.2, because
summarizing and showing the evaluation of the thousands of available animal bioassays on
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
TCDD was prohibitive, only studies first meeting the in vivo animal bioassays study inclusion
criteria below (and shown in Figure 2-3) are summarized. These studies are also presented as
key studies in Section 2.4.3 (see Tables 2-6 and 2-7 for cancer and noncancer endpoints,
respectively).
2.3.1. Study Inclusion Criteria for TCDD Epidemiologic Studies
This section identifies the process EPA used to select epidemiologic studies for defining
candidate PODs for TCDD dose-response modeling. These criteria are based on EPA's
approaches for deriving OSFs and RfDs. A discussion of the considerations used in selecting
epidemiologic data for quantitative dose-response modeling is valuable, particularly given EPA's
preference to use high-quality human studies over animal studies because such human studies are
regarded as providing the most relevant information needed for quantitative human health risk
analyses (U.S. EPA, 2005). As described by Hertz-Picciotto (1995), key components needed for
the use of an epidemiologic study as a basis for quantitative risk assessment include issues
regarding exposure assessment (a well-quantified exposure assessment with exposures linked to
individuals) and study quality ("strong biases," for example with respect to inclusion criteria for
membership in the cohort and follow-up procedures "ruled out or unlikely" and "confounding
controlled or likely to be limited"). The strength of the association, either within the full study or
within a high exposure subgroup, can also be considered in the evaluation of suitability for
dose-response modeling (Hertz-Picciotto, 1995). Stayner et al. (1999), however, note that even
weak associations could be useful in terms of providing an estimate of a potential upper bound
for a quantitative risk estimate.
EPA's method for applying the TCDD study inclusion criteria to epidemiologic data is
detailed below and in Figure 2-2. Based on the framework discussed above, EPA evaluated the
available epidemiologic cohorts and studies based on the five following considerations:
1. The methods used to ascertain health outcomes are clearly identified and unbiased, with
high sensitivity and specificity.
2. The risk estimates generated from the study are not susceptible to important biases
arising from an inability to control for potential confounding exposures or other sources
of bias arising from either study design or statistical analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
3. The study demonstrates an association between TCDD and an adverse health effect
(assuming minimal misclassification of exposure and absence of important biases) with
some suggestion of an exposure-response relationship.
4. The exposure assessment methodology is clearly described and can be expected to
provide adequate characterization of exposure, with assignment of individual-level
exposures within a study (e.g., based on biomarker data, or based on a
job-exposure-matrix approach). Limitations and uncertainties in the exposure assessment
are considered.
5. The size and follow-up period of a cohort study are large enough and long enough,
respectively, to yield sufficiently precise estimates for use in development of quantitative
risk estimates and to ensure adequate statistical power to limit the possibility of not
detecting an association that might be present (i.e., to avoid Type II Errors due to failing
to reject the null hypothesis when the null hypothesis is true). Similar considerations
regarding sample size and statistical precision and power apply to case-control studies.
Three specific study inclusion criteria were used to select studies for further evaluation
and potential TCDD quantitative dose-response assessment
• The study is published in the peer-reviewed scientific literature and includes an
appropriate discussion of strengths and limitations.
• The exposure is primarily to TCDD, rather than dioxin-like compounds (DLCs), and is
properly quantified so that dose-response relationships can be assessed.
• The effective dose and oral exposure must be reasonably estimable. The measures of
exposure must be consistent with the current biological understanding of dose. For
TCDD dose-response assessment, it is critical that reported dose is consistent with a dose
that is likely to be toxicologically relevant. The timing of the measurement of effects
(i.e., the response) also must be consistent with current biological understanding of the
effect and its progression.
For cancer endpoints, EPA assumes that cumulative TCDD dose estimates are
toxicologically relevant measures. Thus, cancer studies must provide information
about long-term TCDD exposure levels. Further, EPA reasons that measures of
cancer occurrence or death need to allow for examination of issues of latency
between the end of effective exposure and cancer detection or death.
For noncancer endpoints, exposure estimates and analysis must allow for examination
of issues of latency and other issues regarding the appropriate time window of
exposure relevant for specific endpoints. Also, to be consistent with the RfD
methodology, the response must be to a nonfatal endpoint.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Those studies that met the aforementioned considerations and inclusion criteria (see
Sections 2.4.1, 2.4.3, and Appendix B) were then subjected to further consideration for
quantitative dose-response analyses.
2.3.2. Study Inclusion Criteria for TCDD In Vivo Mammalian Bioassays
This section identifies the criteria EPA applied to select nonhuman in vivo mammalian
studies for defining candidate PODs for use in TCDD dose-response modeling. These inclusion
criteria are based on EPA's approaches for deriving OSFs and RfDs from bioassay data (U.S.
EPA, 2005). EPA agrees with the NAS committee regarding the utility of an oral RfD and the
need for reevaluation of the OSF for TCDD, specifically in light of data that have been published
since the 2003 Reassessment was released. RfDs and OSFs are generally derived using data sets
that demonstrate the occurrence of adverse effects, or their precursors, in low-dose range for that
chemical. RfDs and OSFs are derived from a health protective perspective for chronic
exposures. Thus, when a group of studies is available on a chemical for which a number of
effects are observed at various doses across those studies, the studies using the lowest exposures
that show effects will typically drive the RfD and OSF derivations, all other considerations being
equal. Studies conducted at higher exposures relative to other available studies are used as
supporting evidence for the final RfD or OSF since they were conducted at doses too high to
impact the numeric derivations of toxicity values. EPA expresses RfDs and OSFs in terms of
average daily doses, usually as mg/kg-day and per mg/kg-day, respectively. Thus, the study
inclusion criteria for the animal bioassay data presented in this section include requirements that
average daily exposures in the studies are within a low dose range where, relative to other
studies, they could be considered for development of a toxicity value. These low-dose
requirements do not imply that TCDD studies conducted at higher doses are of poor quality,
simply that they are not quantitatively useful in the development of toxicity values because other
studies with lower exposures will drive the RfD and OSF derivations under current EPA
practice. Because EPA has identified -2,000 studies on TCDD that may be considered for this
purpose, the development and application of these study inclusion criteria has been critical to
moving the risk assessment process forward.
EPA's method for applying study inclusion criteria for mammalian bioassays is detailed
below and in Figure 2-3. The first study inclusion criterion is that the study is published in the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
peer-reviewed scientific literature. Then, two specific study inclusion criteria were used to select
studies for further evaluation and potential TCDD quantitative dose-response analyses and
identification of candidate PODs:
1. The lowest dose level tested is <1 [j,g/kg-day for cancer studies and <30 ng/kg-day for
noncancer studies.
2. The study design consists of orally administered TCDD-only doses, and specifies the
purity and matrix used to administer the doses.
Then, EPA evaluated the remaining in vivo animal studies based on the following four
considerations.
1. The study tests mammalian species, identifying the strain, gender, and age of the tested
animals.
2. The study clearly documents testing protocol, including dosing frequency, duration, and
timing of dose administration relative to age of the animals.
3. The overall study design is consistent with standard toxicological principles and
practices. The control group or groups are appropriate, given the testing protocol, and are
well characterized. Clinical and pathological examinations conducted during the study
are endpoint-appropriate, particularly for negative findings.
4. The magnitude of animal responses is outside the range of normal variability exhibited by
control animals (e.g., greater than or less than one standard deviation).
Those studies that met the aforementioned considerations and inclusion criteria (see
Sections 2.4.2 and 2.4.3) were then subjected to dose-response analysis.
The criteria for dose requirements, although somewhat arbitrary, are intended to be
reasonable cutoffs that restrict the number of studies that would need to be modeled while
ensuring that all study/data set combinations that could be candidates for the cancer slope factor
or RfD were modeled. Thus, the dose range under consideration allows for liberal ranges of
no-observed-adverse-effect levels (NOAELs), lowest-observed-adverse-effect levels (LOAELs),
and benchmark dose lower confidence bound (BMDLs) for assessment of both cancer and
noncancer effects.
For cancer studies, the dose requirements were selected based on an initial evaluation of
available average daily doses administered in TCDD animal bioassays in which adverse effects
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
were observed. For example, in cancer studies, a sample of the relatively low ranges of tested
average daily doses include 1-1000 ng/kg-day (Toth et al., 1979), 1-100 ng/kg-day (Kociba et
al., 1978), 1.43-286 ng/kg-day (NTP, 1982) and 2.14-71.4 ng/kg-day (NTP, 2006) with
statistically significant increases in tumor incidence via pair-wise or trend tests found in all of
these studies. The entire range of each these studies is <1 |ig/kg-day. The linearized multi-stage
model used by EPA to estimate OSFs is most appropriately applied to studies from which PODs
can be estimated as closely as possible to the experimental data. Thus, given the dose ranges in
these studies that are available for modeling, the restriction to <1 [j,g/kg-day for cancer was
considered to be a reasonable cutoff.
For noncancer studies, dose ranges are more complex and vary according to study
endpoint. Examples of the lowest administered doses (not including developmental studies) that
might be considered as NOAELs or LOAELs in POD determinations for noncancer endpoints
include 1 ng/kg-day (Toth et al., 1979), 1.43 ng/kg-day (Cantoni et al., 1981), 1.07 ng/kg-day
(Smialowicz et al., 2008) 1.43 ng/kg-day (NTP, 1982) and 2.14 ng/kg-day (NTP, 2006). Most of
the lowest tested doses in the TCDD studies have been designated as LOAELs (see Section 4.1).
Given the available database, it is likely that the same composite uncertainty factor (e.g., of 300;
3 for UFa [interspecies], 10 for UFH [intraspecies], and 10 for UFL [LOAEL to NOAEL]) would
be applied to any animal noncancer LOAEL used to derive an RfD for TCDD. This implies that
any study that has a LOAEL of 30 ng/kg-day or more would result in a candidate RfD that is
more than an order of magnitude higher than the example doses of 1-2 ng/kg-day shown here.
BMDLs that might be derived from such data also would not be expected to be lower than these
example doses of 1-2 ng/kg-day. Thus, a tested dose <30 ng/kg-day is considered to be a
reasonable cutoff where the lowest tested dose would never be used as a POD to derive an RfD
given that much lower tested doses (associated with adverse effects) are available from other
studies of acceptable quality.
2.4. EVALUATION OF KEY STUDIES FOR TCDD DOSE RESPONSE
2.4.1. Evaluation of Epidemiological Cohorts for Dose-Response Assessment
2.4.1.1. Cancer
In the 2003 Reassessment, EPA selected three cohort studies from which to conduct a
quantitative dose-response analysis: the National Institute for Occupational Safety and Health
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
(NIOSH) cohort (Steenland et al., 2001), the BASF cohort (Ott and Zober, 1996), and the
Hamburg cohort (Becher et al., 1998). Although these studies were deemed suitable for
quantitative dose-response analysis, the criteria EPA used to reach this conclusion were unclear.
In this section, the study selection criteria and methodological considerations presented in
Section 2.3 are systematically applied to evaluate a number of studies to determine their
suitability for inclusion in dose-response modeling. In addition to the three cohorts used in
previous TCDD quantitative risk assessment, considerations are applied to other relevant TCDD
epidemiological data sets that were identified through a literature review for epidemiological
studies of TCDD and cancer. Study summaries and suitability for quantitative dose-response
analysis evaluations are discussed below.
2.4.1.1.1. Cancer cohorts.
2.4.1.1.1.1. The NIOSH cohort.
In 1978, the NIOSH undertook research that identified workers employed by U.S.
chemical companies that made products contaminated with TCDD between 1942 and 1982.
TCDD was generated in the production of 2,4,5-trichlorophenol and subsequent processes. This
chemical was used to make 2,4,5-trichlorophenoxyacetic acid (2,4,5-T), which was a major
component of the widely-used defoliant, Agent Orange. The NIOSH cohort is the largest cohort
of occupational workers studied to date and has been the subject of a series of investigations
spanning more than two decades. It is important to note that this cohort consists mostly of male
workers that were exposed to TCDD via daily occupational exposure, as compared to an acute
accidental exposure scenario seen with other cohorts. The investigations have progressed from a
comparison of the mortality patterns of the cohort to the U.S. general population to
dose-response modeling using serum-derived estimates of TCDD that have been
back-extrapolated several decades. Analyses of cancer data from the NIOSH cohort that are
addressed in this section include Fingerhut et al. (1991), Steenland et al. (1999, 2001), Cheng et
al. (2006), and Collins et al. (2009).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-11 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.1.1. Fingerhut et al., 1991.
2.4.1.1.1.1.1.1. Study summary.
The investigation of Fingerhut and her colleagues published nearly two decades ago
attracted widespread attention (Fingerhut et al., 1991). This retrospective study examined
patterns of cancer mortality for 5,172 workers who comprised the NIOSH cohort, which
combined workers from the company-specific cohorts of Dow Chemical (Ott et al., 1987; Cook,
1981) and the Monsanto Company (Zack and Gaffey, 1983; Zack and Suskind, 1980). These
workers were employed at 12 plants producing chemicals contaminated with TCDD. Almost all
workers in the cohort (97%) had production or maintenance jobs with processes involving
TCDD contamination. On average, workers were employed for 2.7 years specifically in
processes that involved TCDD contamination, and overall, were employed for 12.6 years. The
mortality follow-up began in 1940 and extended until the end of 1987. Vital status was
determined using records from the Social Security Administration, the Internal Revenue Service,
or the National Death Index. The ascertainment of vital status in the cohort was nearly complete,
with less than 1% of the cohort not followed up until death or the end of the study period.
Comparisons of mortality were made relative to the U.S. male general population and
expressed using the standardized mortality ratio (SMR) metric and 95% confidence intervals
(CIs). Life-table methods were used to generate person-years of risk accrued by cohort members
at each plant. Person-years and corresponding deaths were tabulated across age, race, and year
of death strata, which permitted the SMRs to be examined for potential confounding from these
three characteristics. No unadjusted SMRs were presented in the paper. Cross-classification of
person-years and deaths was also done across several exposure-related groupings, including
duration of employment, years since first exposure, years since last exposure, and duration of
exposure. Employment duration was categorized as <5, 5- <10, 10- <15, 15- <20, 20- <25,
25- <30, and >30 years. The variable "years since first exposure" (<10, 10- <20, and >20 years)
was used to evaluate associations in relation to different latency periods. The analysis was
jointly stratified by duration of employment and for varying latency intervals to evaluate whether
cohort members with higher cumulative TCDD levels had higher cancer mortality rates than
those cohort members with lower cumulative levels.
Overall, the cohort of workers had slightly elevated cancer mortality than the general
population (SMR = 1.15, 95% CI = 1.02-1.30). Comparisons to the general population,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
however, yielded no statistically significant excess for any site-specific cancer. Cancer mortality
was examined for the subset of workers that worked for at least one year and had a latency
interval of at least 20 years (n = 1,520). The 1-year cut-point was selected based on analyses of
serum levels in a subset of 253 workers which revealed that every worker employed for at least
one year had a lipid-adjusted serum level that exceeded the mean (7 ppt). Relative to the
U.S. general population, statistically significant excesses in cancer mortality were observed for
all cancers (SMR = 1.46, 95% CI = 1.21-1.76), cancers of the respiratory system (SMR = 1.42,
95% CI = 1.03-1.92), and for soft tissue sarcoma (SMR = 9.22, 95% CI = 1.90-26.95) among
this subset of 1,520 male workers. The elevated SMR for soft tissue sarcoma, however, was
based on only three cases in this subset.
SMRs also were generated across joint categories of duration of exposure and period of
latency for deaths from all cancer sites (combined), and cancer of the trachea, bronchus, and
lung. Increased SMRs were observed in strata defined by longer exposure and latency, but no
statistically significant linear trends were found.
2.4.1.1.1.1.1.2. Study evaluation.
This cohort was the largest of four the International Agency for Research on Cancer
(IARC) considered in its 1997 classification of TCDD as a Group 1 human carcinogen (IARC,
1997). Duration of employment in processes that involved TCDD contamination was used as a
surrogate measure of cumulative exposure. In using this exposure metric, Fingerhut et al. (1991)
assumed that TCDD exposures were equivalent at all production plants. Doses for individual
cohort members were not reconstructed for these analyses, although they were in subsequent
analyses of this cohort.
Workers in this cohort also were exposed to other chemicals, which could lead to bias
due to confounding if these exposures were associated with both TCDD exposure and the health
outcomes being examined. At one plant, workers were exposed to 4-aminobiphenyl. Previous
investigators also reported that workers at another plant were exposed to 2,4,5-T and
2,4-dichlorophenoxyacetic acid (2,4-D) (Bond et al., 1988, 1989; Ott et al., 1987). Subsequent
analyses revealed only modest correlations between duration of employment (a surrogate
measure of exposure to other chemicals) and cumulative TCDD exposure. This suggests that
confounding due to other coexposures is unlikely to have biased the results. In addition,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
subsequent analyses of this cohort indicate that after excluding workers exposed to
pentachlorophenols from analyses, positive associations between cumulative TCDD and all
cancer mortality persisted (Steenland et al., 1999). Removal of workers who died from bladder
cancer also did not substantially change the dose-response association between TCDD and
cancer mortality from all other sites combined. This finding suggests that exposures to
4-aminobiphenyl did not confound the association between cancer mortality and TCDD
exposure. Overall, there is little evidence of confounding by these co-exposures among this
cohort, however, exposure to other possible confounders, such as dioxin-like compounds, was
not examined.
The study collected no information on smoking behavior of the workers, and therefore,
the SMRs do not account for any differences in the prevalence of smoking that might have
existed between the workers and the general population. For several reasons, however, the
inability to take into account smoking is unlikely to have been an important source of bias. First,
mortality from other smoking-related causes of death such as nonmalignant respiratory disease
were not more common in the cohort than in the general population (SMR = 0.96,
95% CI = 0.54-1.58). Second, stratified analyses of workers with at least a 20-year latency
(assuming this subset shared similar smoking habits) revealed that excesses were apparent only
among those who were exposed for at least 1 year. Specifically, when compared to the general
population, the SMR among workers exposed for at least 1 year with a latency of 20 years was
1.46, (95% CI = 1.21-1.76) while those exposed for less than 1 year had an SMR of 1.02
(95%) CI = 0.76-1.36). Third, for comparisons of cancer mortality between blue-collar workers
and the general population, smoking is unlikely to explain cancer excesses of greater than
10-20%) (Siemiatycki et al., 1988). Finally, the investigators found no substantial changes in the
results for lung cancer when risks were adjusted for smoking histories obtained in 1987 from
223 workers employed at two plants. These data were used to adjust for the expected number of
lung cancer deaths expected in the entire cohort (Fingerhut et al., 1991). Following this
adjustment, the expected number of lung cancer deaths resulted in only a small change in the
SMR for lung cancer in the overall cohort from 1.11 (95% CI = 0.89-1.37) to 1.05
(95% CI = 0.85-1.30). Similarly, there was little change in the SMR for lung cancer in higher
exposure subcohort (SMR = 1.39, 95% CI = 0.99-1.89; to SMR = 1.37, 95% CI = 0.98-1.87).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
The use of death certificate information from the National Death Index is appropriate for
identifying cancer mortality outcomes. For site-specific cancers such as soft tissue sarcoma,
however, the coding of this underlying cause of death is more prone to misclassification (Percy
et al., 1981). Indeed, a review of tissues from four men concluded to have died from soft-tissue
sarcoma determined that two deaths had been misclassified (Fingerhut et al., 1991). A review of
hospital data revealed that two other individuals had soft tissue sarcomas that were not identified
by death certificate information. The use of death certificate information to derive SMRs for
cancer as a whole is likely not subject to significant bias; the same might not hold true, however,
for some site-specific cancers such as soft tissue sarcoma.
Using the SMR metric to compare an occupational cohort with the general population is
subject to what is commonly referred to as the "healthy worker effect" (Li and Sung, 1999; Choi,
1992). The healthy worker effect is a bias that arises because those healthy enough to be
employed have lower morbidity and mortality rates than the general population. The healthy
worker effect is likely to be larger for occupations that are more physically demanding
(Aittomaki et al., 2005; Checkoway et al., 1989), and the healthy worker effect is considered to
be of little or no consequence in the interpretation of cancer mortality (McMichael, 1976;
Monson, 1986). Few cancers are associated with a prolonged period of poor health that would
affect employability long before death. Also recognized is that, as the employed population
ages, the magnitude of the healthy worker effect decreases as the absolute reduction in mortality
becomes relatively smaller in older age groups (McMichael, 1976). The mortality follow-up of
occupational cohorts generally spans several decades, which should minimize the associated
healthy worker effect in such studies. Bias could also be introduced in that workers who are
healthier might be more likely to stay employed and therefore accrue higher levels of exposure.
In the NIOSH cohort, however, mortality was ascertained for those who could have left the
workforce or retired by linking subjects to the National Death Index. Although internal cohort
comparisons can minimize the potential for the healthy worker effect for the reasons presented
above, for cancer outcomes, the SMR statistic is a valuable tool for characterizing whether
occupational cohort are more likely to die of cancer than the general population. Moreover,
stratified analyses across categories of duration of exposure, or latency periods within a cohort
can yield important insights about which workers are at greatest risk. Perhaps most important,
subsequent analyses of the NIOSH cohort that presented risk estimates derived from external
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
comparisons using the SMR were remarkably consistent with rate ratios derived using an internal
referent (Steenland et al., 1999).
2.4.1.1.1.1.1.3. Suitability of data for TCDD dose-response modeling.
This cohort meets most of the identified considerations for conducting a quantitative
dose-response analysis for mortality from all cancer sites combined. The NIOSH cohort is the
largest cohort of TCDD-exposed workers, exposure characterization at an individual level is
possible but not available in this particular study, the follow-up period is long enough to evaluate
latent effects, and the study is not subject to important sources of bias that would materially alter
findings. For the purpose of quantitative dose-response modeling, it is important to note that
subsequent studies of this cohort adopted methods that greatly improved the characterization of
TCDD exposure in this cohort and increased the follow-up interval (Steenland et al., 2001;
Cheng et al., 2006). As such, for all practical purposes, due consideration for dose-response
modeling should focus on the more recently developed data sets.
For quantitative dose-response modeling for individual cancer sites, the data are much
more limited. A statistically significant positive association with TCDD was noted only for soft-
tissue sarcoma among those with more than 1 year of exposure and 20 years of latency
(SMR = 9.22, 95% CI = 1.90-26.95). However there were only three deaths from soft tissue
sarcoma among this exposed component of the cohort, and four deaths in total in the overall
cohort. Also, misclassification of outcome for soft-tissue sarcoma through death registries is
well recognized and supported with additional review of tissue from two of the men.
Specifically, tissues from the four men who died of soft-tissue sarcoma revealed that only two of
these cases were coded correctly.
The design of this initial publication of the NIOSH cohort did not allow for exposures to
other possible confounders, such as dioxin-like compounds and other occupational exposures, to
be examined and accounted for in the analyses. TCDD exposures were based on duration of
employment and effective doses could not be estimated based on the TCDD exposure measure
(i.e., duration of employment). Therefore, dose-response modeling was not conducted for this
study.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.1.2. Steenlandet al., 1999.
2.4.1.1.1.1.2.1. Study summary.
A subsequent analysis of the NIOSH cohort extended the follow-up interval of Fingerhut
et al. (1991) by 6 years (i.e., from 1940-1993) and improved characterization of TCDD exposure
(Steenland et al., 1999). A key distinction from the work of Fingerhut et al. (1991) was the
exclusion of several workers that had been included in the previous mortality analyses. The
authors excluded 40 workers who were either female, had never worked in TCDD-exposed
departments, or had missing date of birth information. An additional 238 workers were excluded
as occupational data for characterizing duration of exposure were lacking, preventing their use in
a subcohort dose-response analysis. This subcohort was further reduced by excluding workers
from four plants (n = 591) because the information on the degree of TCDD contamination in
work histories was limited, preventing the characterization of TCDD levels by job type.
Thirty-eight additional workers were excluded from the eight remaining plants because TCDD
contamination could not be estimated. Finally, 727 workers were excluded because they had
been exposed to pentachlorophenol. In total, exposures were assigned to 3,538 (69%) members
of the overall cohort, a cohort substantially reduced from the 5,172 on which Fingerhut et al.
(1991) reported. Steenland et al. (1999) also evaluated the mortality experience of a subcohort
of 608 workers with chloracne who had no exposure to pentachlorophenol.
For each worker, a quantitative exposure score for each day of work was calculated based
on the concentration of TCDD (j_ig/g) present in process materials, the fraction of the day
worked, and a qualitative contact level based on estimates of the amount of TCDD exposure via
dermal absorption or inhalation. The authors derived a cumulative measure of TCDD exposure
by summing the exposure scores across the working lifetime history for each worker. The
authors validated this cumulative exposure metric indirectly by comparing values obtained for
workers with and without chloracne. Such a validation is appropriate, given that chloracne is
considered a clinical sign of exposure to dioxin. The median exposure score among those with
chloracne was 11,546 compared with 77 among those without (Steenland and Deddens, 2003).
Cancer mortality was compared using two approaches. As in Fingerhut et al. (1991),
external comparisons were made to the U.S. general population using the SMR statistic. The
authors adjusted the SMR statistics for race, age, and calendar time. They also applied life-table
methods to characterize risks across the subcohort of 3,538 workers with exposure data by
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
categorizing the workers into seven cumulative exposure groups. The cut-points for these
categories were selected so that the number of deaths in each category was nearly equal to
optimize study power. Life-table analyses were extended further to consider a 15-year lag
interval, which in a practical sense means that person-years at risk would not begin to accrue
until 15 years after the first exposure occurred. The person-years and deaths that occurred in the
first 15 years were included in the lowest exposure grouping. The Cox proportional hazards
model was used to characterize risk within the cohort. Cox regression was used to provide an
estimate of the hazard ratios and the 95% CIs for ischemic heart disease, all cancers combined,
lung cancer, smoking related cancers, and all other cancers. The authors also performed Cox
regression analyses using the seven categories of exposure, adjusting the regression coefficients
for year of birth and age. The regression models were run for both untagged and lagged
(15 years) cumulative exposure scores.
Overall, when compared with the U.S. general population, a slight excess of cancer
mortality (from all sites) was noted in the 5,132 cohort study population (SMR = 1.13,
95% CI = 1.02-1.25). This result did not substantially differ from the earlier finding that
Fingerhut et al. (1991) published (SMR= 1.15, 95% CI = 1.03-1.30). Site-specific analyses
revealed statistically significant excesses relative to the U.S. general population for bladder
cancer (SMR = 1.99, 95% CI = 1.13-3.23) and for cancer of the larynx (SMR = 2.22,
95% CI = 1.06-4.08). In the chloracne subcohort (n = 608), a borderline statistically significant
excess relative to the general population was found for all cancer sites combined (SMR = 1.25,
95% CI = 0.98-1.57) and for lung cancer (SMR = 1.45, 95% CI = 0.98-2.07). The authors also
found statistically significant excesses for connective and soft tissue sarcomas (SMR = 11.32,
95%) CI = 2.33-33.10) and for lymphatic and hematopoietic malignancies (SMR = 3.01,
95% CI = 1.43-8.52).
External comparisons made by grouping workers into septiles of cumulative TCDD
exposure and generating an SMR for each septile using the U.S. population as the referent group
suggested a dose-response relationship. For all cancer sites combined, workers in the highest
exposure score category had an SMR of 1.60 (95% CI = 1.15-1.82); increases also were
observed in the sixth (SMR = 1.34) and fifth (SMR = 1.15) septiles. The two-sided^-value
associated with the test for trend for cumulative TCDD exposure was statistically significant
(p = 0.02). A similar approach for lung cancer revealed virtually the same pattern. The
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
incorporation of a 15-year latency for the analyses of all cancer deaths, in general, produced
slightly higher SMRs across the septiles, although a slight attenuation of effect was noted in the
highest septile (SMRuniagged = 1.60 vs. SMRiagged = 1.54). For a 15-year lag, the lung cancer
SMRs were mixed compared to the untagged results with some septile exposure categories
increasing and others decreasing relative to the lowest exposure group.
For the internal cohort comparisons using Cox regression analyses higher hazard ratios
were found among workers in the higher exposure categories than in the lowest septile. The
linear test for trend, however, was not statistically significant (p = 0.10). The associations across
the septiles for the untagged exposure for the internal cohort comparisons were not as strong as
for the external cohort comparisons. The opposite was true, however, for cumulative exposures
lagged 15 years.
Relative to the lowest septile, stratified analyses revealed increased hazard ratios in the
upper septiles of the internal cohort comparisons for both smoking- and nonsmoking-related
forms of cancer. The test for linear trend was statistically significant for all other cancers (after
smoking-related cancers were excluded). These analyses suggest that the overall cancer findings
were not limited to an interaction between TCDD and smoking. Additional sensitivity analyses
by the authors indicated the findings for smoking-related cancers were largely unaffected by the
exclusion of bladder cancer cases. This observation suggests that the exposure to
4-aminobiphenyl, which occurred at one plant and might have contributed to an increased
number of bladder cancers, did not substantially bias the dose-response relationship between
TCDD and all cancers combined.
The investigators also evaluated the dose-response relationship with a Cox regression
model separately for each plant using internal cohort comparisons and found some heterogeneity.
This finding is not unexpected particularly given the relatively small number of cancer deaths at
each plant, and given that exposures were quite low for one plant at which no positive
association was found. The variability among plants was taken into account by modeling plant
as a random effect measure in the Cox model, which produced little change in the slope
coefficient (P = 0.0422 vs. P = 0.0453).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.1.2.2. Study evaluation.
This study represents a valuable extension of that by Fingerhut et al. (1991). Internal
comparisons were performed to help minimize potential biases associated with using an external
comparison group (e.g., healthy worker effect, and differences in other risk factors between the
cohort and the general population). That similar dose-response relationships were found for
internal and external comparison populations suggests that the bias due to the health worker
effect in the cohort might be minimal for cancer mortality. More importantly, the construction of
the cumulative exposure scores provides an improved opportunity to evaluate dose-response
relationships compared with the length of exposure and duration of employment metrics that
Fingerhut et al. (1991) used.
A potential limitation of the NIOSH study was the inability to account for cigarette
smoking. If cigarette smoking did contribute to the increased cancer mortality rates in this and
other cohorts, increased cancer mortality from exposure to TCDD would be expected only for
smoking-attributable cancers. This study demonstrates associations with TCDD for both
smoking- and nonsmoking-related cancers, including a stronger association for
nonsmoking-related cancers. Therefore, the data provide evidence that associations between
TCDD and cancer mortality are not likely due to cigarette smoking.
The findings regarding latency should be interpreted cautiously as the statistical power in
the study to compare differences across latency intervals was limited. Caution also should be
heeded, given that latency intervals can vary on an individual basis as they are often
dose-dependent (Guess and Hoel, 1977). The evaluation of whether TCDD acts as either an
initiating or promoting agent (or both) is severely constrained by the reliance on cancer mortality
data rather than incidence data. This constraint is due to the fact that survival time can be quite
lengthy and can vary substantially across individual and cancer subtype. For example, the 5-year
survival among U.S. males for all cancer sites combined ranged between 45 and 60% (Clegg et
al., 2002). When only mortality data are available, evaluating the time between when individuals
are first exposed and when they are diagnosed with cancer is nearly impossible.
Starr (2003) suggested that Steenland et al. (1999) focused too heavily on the exposures
that incorporated a 15-year period of latency and that those who experienced high exposures
would inappropriately contribute person-years to the lowest exposure group "irrespective of how
great the workers' actual cumulative exposure scores may have been." Most cancer deaths
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-20 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
would, however, typically occur many years postemployment. Given that the follow-up interval
of the cohort was long and the average exposure duration was 2.7 years, at the time of death,
person-years for those with high cumulative exposures would be captured appropriately. The
median 5-year survival for all cancers is approximately 50% (Clegg et al., 2002), so applying a
minimum latency of 5 years when using cancer mortality rather than cancer incidence data is
needed to assure that the exposure metric is capturing exposures that occur before diagnoses.
Increasing this latency period, for example to 10 or 15 years, would eliminate consideration of
exposures that occur in the period between tumor occurrence and tumor detection (diagnosis),
and allows for an appropriate focus on exposures that act either early or late in the pathogenic
process. The IARC recommendation also underscores the relevance of including a period of
latency when estimating cancer risks related to TCDD. Specifically, IARC indicated that if the
association of TCDD with cancer is causal, effects might become apparent only at high
exposures and with adequate latency; they suggested that a latency interval of 15 years could be
too short (IARC, 1997). EPA considers the Steenland et al. (1999) presentation to be balanced in
that they provided the range in lifetime excess risk estimated across the various models used.
The authors' finding that the models with a 15-year lag provided a statistically significant
improvement in fit based on the chi-square test statistic should not be readily dismissed.
2.4.1.1.1.1.2.3. Suitability of data for TCDD dose-response modeling.
This study meets most of the epidemiological considerations for conducting a
quantitative dose-response analysis for mortality from all cancer sites combined. This study
excludes a large number of workers who were exposed to pentachlorophenol, thus eliminating
the potential for bias from this exposure and used an improved methodology for assigning TCDD
exposures to the workers. However, given that exposures to other dioxin-like compounds were
not described, it is unclear if the exposures among this cohort were primarily to TCDD.
Therefore, dose-response modeling was not pursued for this study, but was for the subsequent
NIOSH study by Steenland et al. (2001), which did examine exposure to dioxin-like compounds.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-21 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.1.3. Steenlandet al., 2001.
2.4.1.1.1.1.3.1. Study summary.
In 2001, Steenland et al. published a risk analysis using the NIOSH cohort that for the
first time incorporated serum measures in the derivation of TCDD exposures for individual
workers. The authors applied the same exclusion criteria to the entire cohort of workers across
the 12 plants in the Steenland et al. (1999) study, which left 3,538 workers for which risk
estimates could be calculated. Cumulative TCDD serum levels were estimated on an individual
basis for all 3,538 workers by developing predictive models that used a subset of 170 workers for
which both serum measures and TCDD exposures scores were available (Steenland et al., 2001).
Unlike previous analyses of the NIOSH cohort that considered several different mortality
outcomes, the analyses presented in Steenland et al. (2001) focused exclusively on mortality
from all cancers sites combined. The authors observed 256 cancer deaths in the cohort during
the follow-up interval that extended from 1942 until the end of 1993. All risks estimated in the
Steenland et al. (2001) study were based on internal cohort comparisons.
Characterization of TCDD exposure levels among the workers was based on serum
measures obtained in 1988 from 199 workers who were employed in one of the eight plants. The
researchers restricted the development of the model to include only those workers whose
measured serum levels were deemed to be greater than the upper range of background levels
(10 ppt), which resulted in 170 workers.
The authors developed a regression model that could estimate the level of TCDD at the
time of last exposure for the 170 workers. The model was developed based on the estimated
half-life of TCDD, the known work history of each worker, a pharmacokinetic model for the
storage and excretion of TCDD, and exposure scores for each job held by each worker over time.
The resulting equation follows
yiast exposure >>1988 exp(kAt) (Eq. 2-1)
The first-order elimination rate constant QC) was based on a half-life of 8.7 years
previously reported for the Ranch Hands cohort (Michalek et al., 1996). The background rate of
TCDD exposure was assumed to be 6.1 parts per trillion (ppt), which was based on the median
level in a sample of 79 unexposed workers in the NIOSH cohort (Piacitelli et al., 1992). This
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
value was subtracted when TCDD values were back-extrapolated, and then added again after the
back-extrapolation was completed. A background level of 5 ppt also was used in some of the
analyses with minimal demonstrable effects on the results. Sensitivity analyses also were
incorporated to consider a 7.1-year half-life estimate that had been developed for the earlier
Ranch Hands study (Pirkle et al., 1989).
After back-extrapolating to obtain TCDD serums levels at the time of last exposure, the
investigators estimated cumulative (or "area under the curve") TCDD serum levels for every
cohort member. This estimation procedure was the same method Flesch-Janys et al. (1998)
applied to the Hamburg cohort to derive a coefficient for relating serum levels to exposure
scores. The "area under the curve" approach integrates time-specific serum levels over the
employment histories of the individual workers. The slope coefficient was estimated using a
no-intercept linear regression model. This model is based on the assumption that a cumulative
score of zero is associated with no serum levels above background.
Cox regression was also used to model the continuous measures of TCDD. A variety of
exposure metrics were considered that took into account different lags, nonlinear relationships
(e.g., log-transform and cubic spline), as well as threshold and nonthreshold exposure metrics.
Categorical analyses were used to evaluate risks across TCDD exposure groups, while different
shapes of dose-response curves were evaluated through the use of lagged and untagged
continuous TCDD measures. Categorical analyses of TCDD exposure were conducted using the
Cox regression model to derive estimates of relative risk (RR) as described by hazard ratios and
95% CIs. The reference group in this analysis was those workers in the lowest septile
cumulative exposure grouping (<335 ppt-years). The septiles were chosen based on cumulative
serum levels that considered no lag and also a 15-year lag.
The investigators also conducted dose-response analyses using the toxicity equivalence
(TEQ) approach. The TEQ is calculated as the sum of all exposures to dioxins and furans
weighted by the potency of each specific compound. In this study, TCDD was assumed to be
account for all dioxin exposures in the workplace. For background TEQ levels, the investigators
used a value of 50 ppt in the dose-response modeling. This is based on the assumption that
TCDD accounted for 10% of the toxicity of all dioxins and furans (WHO, 1988), and is
equivalent to using a background level of 5 ppt/yr that was used in the derivation of cumulative
serum TCDD levels. A statistically significant dose-response pattern was observed for all cancer
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
mortality and TCDD exposure based on log of cumulative TEQs with a 15-year lag. A
comparison of the overall model chi-square values indicated that the fit of this model was not as
good as that for TCDD.
The hazard ratios among workers grouped by categories of cumulative TCDD exposure
(lagged 15 years) suggested a dose-response relationship. Steenland et al. (2001) found
statistically significant excesses in the higher exposure categories compared to the lowest septile.
The RR was 1.82, 95% CI = 1.18-2.82 for the sixth septile (7,568-20,455 ppt-years) and 1.62,
95% CI = 1.03-2.56) for the seventh septile (>20,455 ppt-years). Cox regression indicated that
log TCDD serum concentrations (lagged 15 years) was positively associated with cancer
mortality (P = 0.097, standard error (P) = 0,032, p < 0.003). A statistically significant
improvement in fit was observed when a 15-year lag interval was incorporated into the model
compared to a model with no such lag [Model % with 4 degrees of freedom (df) = 7.5], Results
were similar when using a half-life of 7.1 years rather than 8.7 years. The excess lifetime risk of
death from cancer at age 75 for TCDD intake (per 1.0-picogram per kilogram [pg/kg] of body
weight (BW) per day) was about 0.05-0.9%) above a background lifetime risk of cancer death of
12.4%). The results from the best-fitting models provide lifetime risk estimates within the ranges
derived using data from the Hamburg cohort (Becher et al., 1998).
In both categorical and continuous analyses of TCDD based on a linear exposure metric,
the dose-response pattern tailed off at high exposures suggesting nonlinear effects. This
phenomenon could be due to saturation effects (Stayner et al., 2003) or, alternatively, could have
resulted from increased exposure misclassification of higher exposures (Steenland et al., 2001).
As the authors highlighted, some of the highest exposures might have been poorly estimated as
they occurred in workers exposed to short-term high exposures during the clean-up of a spill.
The choice of a linear model to develop data from a single time point can also result in exposure
misclassification in those individuals that have differences in the length of exposure (Emond et
al., 2005). Misclassification would be less likely at low concentrations where dose-dependent
elimination is minimal.
2.4.1.1.1.1.3.2. Study evaluation.
An important consideration in the Steenland et al. (2001) study was the use of a small
subset of workers (n = 170) to infer exposures for the remainder of the cohort. This subset
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
comprised surviving members of the cohort (in 1988), and therefore, their age distribution would
have differed from the rest of the cohort. Furthermore, these workers were employed at a single
plant, at which the work histories were less detailed than at other plants; thus, the development of
the exposure scores differed between this plant and that of the others. Also, many of the workers
at this plant had the same job title and were employed during the same calendar period. The use
of serum data from this subset adds a level of uncertainty that is not readily characterized.
Despite this limitation, the use of these sera data to derive cumulative measures for all cohort
workers has merit given the strong correlation observed between the exposure scores, and TCDD
serum levels estimates at the time of last exposure (Spearman r = 0.90).
The authors performed an extensive series of sensitivity analyses and considered several
alternative exposure metrics to the simple linear model. The lifetime excess risk above
background was nearly twice as high for the log cumulative serum measures with a 15-year lag
when compared to the piecewise linear models with no lag. An important observation was that
the exposure metric based on cumulative serum (lagged 15 years) did not fit the data as well as
the cumulative exposure score used in earlier analyses (Steenland et al., 1999). A priori, one
would expect that a better fit would be obtained with serum-based measures because serum
levels are a better measure of relevant biological dose. As the authors noted, inaccuracies
introduced in estimating the external-based exposure scores could have contributed to a poorer
fit of the data. Alternatively, exposure misclassification error could be introduced if serum
samples based on the 170 workers were not representative of the entire cohort. Although the
serum-based measures did not fit the data as well as the exposures scores, the authors regarded
them as providing a reasonable fit. Moreover, the serum-based measures enabled better
characterization of risk in units (pg/kg-day) that can be used in regulation exposures.
2.4.1.1.1.1.3.3. Suitability of data for TCDD dose-response modeling.
This study meets all of the epidemiological considerations for conducting a quantitative
dose-response analysis for mortality from all cancer sites combined. As mentioned previously,
the NIOSH cohort is the largest assembled to date for which TCDD-related risks of cancer
mortality can be estimated. The use of serum-based measures provides an objective measure of
TCDD exposure. Repeated measures in other study populations have provided reasonable
estimates of the half-life of TCDD, which permitted back-extrapolation of exposures.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
The authors have made extensive efforts to evaluate a wide variety of nonlinear and
linear models with varying lengths of latency and log transformations. The model chi-square test
statistics were fairly similar for the log cumulative serum (15-year lag) (Model %2(4df) = 11.3)
model and the piecewise linear model (no lag) (Model %2(5df) = 12.5). These models, however,
produced results with twofold differences in lifetime excess risks. These differences underscore
the importance of characterizing uncertainty in modeling approaches when conducting
dose-response analysis.
The Steenland et al. (2001) study characterizes risk in terms of pg/kg of body weight per
day. Given that tolerable daily intake dioxin levels are typically expressed in pg/kg of body
weight (WHO, 1998), the presentation of risks in terms of these units is an important advance
from the earlier analyses that used exposure scores (Steenland et al., 1999). Many of the
Steenland et al. (2001) findings are consistent with earlier work from this cohort, which is not
surprising given that exposures scores were used to derive serum-based levels for the cohort.
The findings of excess lifetime risks obtained for the best- fitting model are also consistent with
those derived from the Hamburg cohort (Becher et al., 1998). This study meets the
epidemiological considerations noted previously as there is no evidence that the study is subject
to bias from confounding due to cigarette smoking or other occupational exposures. Given the
considerable efforts to measure effective dose to TCDD among the study participants, this study
also meets the requisite dose-response modeling criteria and will be used in quantitative
dose-response analyses of cancer mortality.
2.4.1.1.1.1.4. Cheng et al., 2006.
2.4.1.1.1.1.4.1. Study summary.
Cheng et al. (2006) undertook a subsequent quantitative risk assessment of 3,538 workers
in the NIOSH cohort using serum-derived estimates of TCDD. This dose-response analysis was
published after the 2003 Reassessment document was released. The goal of this study was to
examine the relationship between TCDD and cancer mortality (all sites combined) using a new
estimate of dose that estimated TCDD as a function of both exposure intensity and age using a
kinetic model. This physiologically based pharmacokinetic model has been termed the
"concentration- and age-dependent elimination model" (CADM) and was developed by Aylward
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-26 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
et al. (2005b). This model describes the kinetics of TCDD following oral exposure to humans by
accounting for key processes affecting kinetics by simulating the total concentration of TCDD
based on empirical consideration of hepatic processes (see Section 3.3). An important feature of
this kinetic model is that it incorporates concentration- and age-dependent elimination of TCDD
from the body; consequently, the effective half-life of TCDD elimination varies based on
exposure history, body burden, and age of the exposed individuals. The study was motivated by
the reasoning that back-calculations of TCDD using a first-order elimination model and a
constant half-life of 7-9 years underestimated exposures to TCDD among workers. This
underestimate, in turn, would result in overestimates of the carcinogenic potency of TCDD.
As with the earlier Steenland et al. (2001) analyses, the cohort follow-up period was
extended from 1942 until the end of 1993 and work histories were linked to a job exposure
matrix to obtain cumulative TCDD scores. Two cumulative serum lipid exposure metrics (in
ppt-years) were constructed using the data obtained from the sample of 170 workers. The first
replicated the metric used in a previous analysis of the cohort (Steenland et al., 2001) and was
based on a first-order elimination model with an 8.7-year half-life (Michalek et al., 1996). The
second metric was based on CADM and had two first-order elimination processes (Aylward et
al., 2005a). This metric assumes that the elimination of TCDD in humans occurs at a faster rate
when body concentrations are high and at slower rates in older individuals (Aylward et al.,
2005a, b). The model was optimized using individuals for which serial measures of serum
TCDD were available. These measures were obtained from 39 adults with initial serum levels
between 130 and 144,000 ppt (Aylward et al., 2005b). This group included 36 individuals who
had been exposed in the Seveso accident and 3 exposed in Vienna, Austria. In practice, for
serum levels greater than 1,000 ppt, the effective half-life would be less than 3 years, and for
serum TCDD levels less than 50 ppt, the effective half-life would be more than 10 years
(Aylward et al., 2005b). Results from the model indicate that men eliminate TCDD faster than
women do as demonstrated previously by Needham et al. (1994). These age- and
concentration-dependent processes were assumed to operate independently on TCDD in hepatic
and adipose tissues, and TCDD levels in liver and adipose tissue were assumed to be a nonlinear
function of body concentration. Cheng et al. (2006) calibrated CADM using a dose of 156 ng
per unit of exposure score and assumed a background exposure rate of 0.01 ng/kg-month. The
average TCDD ppt-years derived from CADM with a 15-year lag was 4.5-5.2 times higher than
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
with the first-order elimination model. The two metrics, however, were highly correlated based
on a Pearson correlation coefficient of 0.98 (p < 0.001). Comparisons of fit between the CADM
and first-order elimination model were made using R2 values and presented in Aylward et al.
(2005b).
Cheng et al. (2006) compared the mortality experience of NIOSH workers to the U.S.
general population using the SMR statistic. SMR statistics also were generated separately for
each of the 8 plants and for all plants combined. Cox regression models were used to analyze
internal cohort dose-response. These models used age as the time variable, and penalized
smoothing spline functions of the CADM metric also were considered. The possible
confounding effects of other occupational exposures and other regional population differences
were assessed by repeating analyses after excluding one plant at a time. Lagged and untagged
TCDD exposures were analyzed separately, and stratified analyses compared risk estimates for
smoking- and nonsmoking-related cancers. Cheng et al. (2006) adjusted the slope estimates
derived from the Cox model for potential confounding effects of race and year of birth.
Overall, a statistically significant excess in all cancer mortality in the cohort occurred
relative to the general population (SMR = 1.17, 95% CI = 1.03-1.32). The plant-specific SMRs
ranged from 0.62-1.87, with a statistically significant excess evident only for plant 10
(SMR = 1.87, 95% CI = 1.35-2.52). For lung cancer mortality, the overall SMR was not
statistically significant (SMR = 1.11, 95% CI = 0.89-1.37). A statistically significant excess for
lung cancer also was found for plant 10 (SMR = 2.35, 95% CI = 1.44-3.64). The SMRs between
smoking- (SMR = 1.22, 95% CI = 1.01-1.45) and nonsmoking-related cancers (SMR = 1.12,
95% CI = 0.94-1.33) were comparable.
For the internal cohort analyses of serum-derived measures, the authors were able to
replicate the one-compartmental model used previously (Steenland et al., 2001). As had been
noted by Steenland et al. (2001), an inverse-dose-response pattern was seen for individuals with
high exposures (above 95th percentile); this type of pattern is often seen in occupational studies
(Stayner et al., 2003). Excluding these data produced a stronger association between TCDD and
all-cause mortality. In fact, only when the upper 2.5% or 5% of observations was removed did a
statistically significant positive association become evident with the untransformed data.
Similarly, when the model incorporated a lag of 15 years, a statistically significant association
was noted only for the untransformed TCDD ppt-years with the upper 5% of observations
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
removed. Stratified analyses revealed little difference between smoking- and
nonsmoking-related cancers, and the removal of one plant at a time from the analyses of TCDD
ppt-years changes did not substantially change the slope.
2.4.1.1.1.1.4.2. Study evaluation.
The authors reported that CADM provided an improved fit over the one-compartmental
model, but presented no evidence regarding any formal test of statistical significance. A
comparison of R2 values presented in Aylward et al. (2005b), however, does reveal that the R2
value increased from 0.27 (first-order compartmental model with an 8.7-year half-life) to 0.40
for CADM. TCDD exposures estimated using CADM were approximately fivefold higher than
the one-compartmental model estimates among cohort members with higher levels of exposure.
Differences in exposure estimates between the two metrics were less striking among individuals
with lower TCDD exposures. The net effect was that CADM produced a 6- to 10-fold decrease
in estimated risks compared to estimates previously reported (Steenland et al., 2001).
Nonetheless, the estimates produced by CADM span more than two orders of magnitude under
various assumptions. Further uncertainties arise from between-worker variability of TCDD
elimination rates, possible residual confounding, and the variability associated with the use of
data obtained from other cohorts. Nevertheless, the use of the CADM model to estimate TCDD
exposure is considered a significant advantage over the previous first-order body burden
calculations.
2.4.1.1.1.1.4.3. Suitability of data for TCDD dose-response modeling.
The value of including the NIOSH cohort data has already been established based on
investigations published by Steenland et al. (1999, 2001). The decision to include data from the
quantitative dose-response analysis that Cheng et al. (2006) conducted relates to the added value
that the CADM exposure estimates would provide. The earlier modeling work of Aylward et al.
(2005b) provided some support for a modest improvement of the fit of CADM over the
first-order compartmental model, and they also confirmed previous studies that found that TCDD
elimination rates varied by age and sex. Recent work by Kerger et al. (2006) also demonstrates
that the half-life for TCDD is shorter among Seveso children than the corresponding half-life for
adults, and that body burdens influence the elimination of TCDD in humans. That estimates of
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
half-lives among men have been remarkably consistent, with mean estimates ranging between
6.9 and 8.7 years (Flesch-Janys et al., 1996; Pirkle et al., 1989; Michalek et al., 2002; Needham
et al., 2005), however, is noteworthy. Based on the underlying strengths of the NIOSH cohort
data and efforts by Cheng et al. (2006) to improve estimates of effective dose, these data support
further dose-response modeling.
2.4.1.1.1.1.5. Collins et al., 2009.
2.4.1.1.1.1.5.1. Study summary.
In a recent study, Collins et al. (2009) investigated the relationship between serum TCDD
levels and mortality rates in a cohort of trichlorophenol workers exposed to TCDD. These
workers were part of the NIOSH cohort having accounted for approximately 45% of the
person-years in an earlier analysis (Bodner et al., 2003). The investigators completed an
extensive dioxin serum evaluation of workers employed by the Dow Chemical plant in Midland,
Michigan, that made 2,4,5-trichlorophenol (TCP) from 1942 to 1979 and 2,4,5-T from 1948 to
1982. Collins et al. (2009) developed historical TCDD exposure estimates for all TCP and
2,4,5-T workers. This study represents the largest group of workers from a single plant ever
studied for the health effects of TCDD. Little information on how vital status was ascertained,
either in this paper or in the Bodner et al. (2003) report of mortality in this cohort. Although the
authors indicate that death certificates were obtained from the states in which the employees
died, whether vital status was ascertained from company records or through record linkage to the
National Death Index is unclear.
The follow-up interval for these workers covered the period between 1942 and 2003.
Thus, the study included 10 more years of follow-up than earlier investigations of the entire
NIOSH cohort. Serum samples were obtained from 280 former workers collected during
2004-2005. A simple one-compartment first-order pharmacokinetic model and elimination rates
as estimated from the BASF cohort were used (Flesch-Janys et al., 1996). The "area under the
curve" approach was used to characterize workers' exposures over the course of their working
careers and provided a cumulative measure of exposure. Analyses were performed with and
without 165 of the 1,615 workers exposed to pentachlorophenol to evaluate the impact of these
exposures.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
External comparisons of cancer mortality rates to the general U.S. population were made
using SMRs. Internal cohort comparisons of exposure-response relationships were made using
the Cox regression model. This model used age as the time variable, and was adjusted for year
of hire and birth year. Only those causes of death for which an excess was found based on the
external comparisons or for which previous studies had identified a positive association were
selected for dose-response analyses.
A total of 177 cancer deaths were observed in the cohort. For the external comparison
with the U.S. general population, overall, no statistically significant differences were observed in
all cancer mortality among all workers (SMR = 1.0, 95% CI = 0.8-1.1). Results obtained after
excluding workers exposed to pentachlorophenol were similar (SMR = 0.9, 95% CI = 0.8-1.1).
Statistically significant excesses in the cohort were found for leukemia (SMR = 1.9,
95%) CI = 1.0-3.2) and soft tissue sarcoma (SMR = 4.1, 95%> CI = 1.1-10.5). Although not
statistically significant SMRs for other lymphohemopoietic cancers included non-Hodgkin's
lymphoma SMR = 1.3; 95%CI = 0.6, 2.5) and Hodgkin's disease (SMR = 2.2; 95% CI = 0.2,
6.4).
Internal cohort comparisons using the Cox regression model were performed for all
cancers combined, lung cancer, prostate cancer, leukemia, non-Hodgkin's lymphoma, and
soft-tissue sarcoma. Whether the internal comparisons excluded those workers exposed to
pentachlorophenol is not entirely clear from the text or accompanying table, but presumably they
do not. The RR was 1.002 (95%> CI = 0.991-1.013) for all cancer mortality per 1 ppb-year
increase in cumulative TCDD exposure was not statistically significant. Except for soft tissue
sarcomas, no statistically significant exposure-response trends were observed for any cancer site.
For soft tissue sarcoma, analyses were based on only four deaths.
2.4.1.1.1.1.5.2. Study evaluation.
A key limitation of this study is that SMRs were not derived for different periods of
latency for the external comparison group analysis. The original publication on the NIOSH
cohort found that SMRs increased when a 20-year latency period was incorporated (Fingerhut et
al., 1991), and similar patterns have been observed in other occupational cohorts (Manz et al.,
1991; Ott and Zober, 1996) and among Seveso residents (Consonni et al., 2008). Additionally,
dose-response analyses showed marked increases in slopes with a 15-year latency period
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
(Steenland and Deddens, 2003; Cheng et al., 2006). In this context, the absence of an elevated
SMR for cancer mortality is consistent with previous findings of the NIOSH cohort. While the
cohort did have sufficient follow-up, no evaluation of possible latent effects was presented and
this is a major limitation of this study. Further, the evaluation of the exposure metrics should be
expanded from what was presented in Collins et al. (2009) due to the previous analyses of the
same workers finding positive associations between cancer mortality and TCDD (Steenland et
al., 2001).
Unfortunately, the Collins et al. (2009) study did not include a categorical analysis of
TCDD exposure and cancer mortality. This categorical analysis would have enabled an
evaluation of whether a nonlinear association exists between TCDD exposure and cancer risk.
The analyses of both Cheng et al. (2006) and Steenland et al. (2001) suggest an attenuation of
effects at higher doses, and several investigations have considered log-transformed associations
as a means to address nonlinearity. Also, the earlier plant-specific dose-response analyses of
Steenland et al. (2001) are not consistent with the findings for the Midland plant that Collins et
al. (2009) presented. These differences could be due to differences in the construction of
exposure metrics, additional follow-up, or lagging of exposures.
2.4.1.1.1.1.5.3. Suitability of data for dose-response modeling.
The Collins et al. (2009) study uses serum levels to derive TCDD exposure estimates and
does not appear to be subject to important biases. The reliance on data from one plant offers
some advantages over the multiplant analyses, as heterogeneity in exposure to other occupational
agents would be lower. The number of individuals who provided serum samples (n = 280) is
greater than the 170 individuals used to derive TCDD estimates for the NIOSH cohort. This
study's main limitation is that it lacks rigor in sensitivity analyses to explore the impact that
other exposure metrics and model assumptions had on the study findings. The data used in
quantitative dose-response modeling should retain flexibility to account for latency effects. The
data, as structured for this study, are inadequate for quantitative dose-response analysis given
that no dose-response effects were detected.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-32 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.2. The BASF cohort.
In 1953, dioxin contamination occurred as a result of an autoclave accident during the
production of trichlorophenol at the BASF plant in Ludwigshafen, Germany. A second dioxin
incident occurred in 1988 that was attributed to the blending of thermoplastic polyesters with
brominated flame retardants. Of the two events, the one on November 13, 1953, was associated
with more severe acute health effects, including chloracne that resulted in immediate
hospitalizations for seven workers. These adverse events were not linked to TCDD until 1957
when TCDD was identified as a byproduct of the production of trichlorophenol and was shown
to induce chloracne (Zober et al., 1994). Zober and colleagues (1998) noted that with the 1988
accident, affected individuals did not exhibit clinical symptoms or chloracne, but rather were
identified through "analytical measures." In both instances, efforts were made to limit the
potential for exposure to employees.
2.4.1.1.1.2.1. Thiess andFrentzel-Beyme, 1977 and Thiess et al., 1982.
2.4.1.1.1.2.1.1. Study summary.
A study of the mortality of workers employed at the BASF plant was first presented in
1977 (Thiess and Frentzel-Beyme, 1977) with subsequent updates in both 1982 (Thiess et al.,
1982), and in 1990 (Zober et al., 1990). In the first published paper (Thiess et al., 1982),
74 employees involved in the 1953 accident were traced and their death certificate information
extracted. Of these, 66 suffered chloracne or severe dermatitis. Observed deaths were compared
to the expected number using three external reference groups: the town of Ludwigshafen
(n = 180,000), the district of Rhinehessia-Palatinate (n = 1.8 million), and the Federal Republic
of Germany (n = 60.5 million). Another comparison group was assembled by selecting
age-matched employees taken from other cohorts under study. This additional comparison was
aimed at avoiding potential biases associated with healthy worker effect when using an external
referent.
During a follow-up interval of up to 26 years (1953-1979), 21 individuals died. Of
these, seven deaths were from cancer. The expected number of cancer deaths derived for the
three external comparison groups ranged between 4.1 and 4.2, producing an SMR of 1.7
(p-values ranged between 0.12 and 0.14). Excess mortality was found for stomach cancer based
on the external comparisons (p < 0.05); however, this was based on only three cases. No other
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-33 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
statistically significant excesses were found with the external comparisons made to the other
cohorts of workers.
2.4.1.1.1.2.1.2. Study evaluation.
In the Thiess et al. (1982) study, no TCDD exposures were derived for the workers, thus
no dose-reconstruction was performed. The findings from this study are limited by the small size
of the cohort. The 74 workers followed in this cohort represent the smallest number of workers
across the occupational cohorts (Fingerhut et al., 1991; Steenland et al., 2001; Becher et al.,
1998; Hooiveld et al., 1998; Michalek and Pavuk, 2008; McBride et al., 2009a, b) that have
investigated TCDD exposures and cancer mortality. Mechanisms of follow-up were excellent as
all individuals were traced, and death certificates were obtained from all deceased workers.
Although the study does compare the mortality experience to other occupational cohorts,
the paper provides insufficient information to adequately interpret the associated findings. For
example, a description of these occupations is lacking making it impossible to determine whether
these cohorts were exposed to other occupational carcinogens that might have confounded the
associations between TCDD exposure and cancer mortality.
2.4.1.1.1.2.1.3. Suitability of data for TCDD dose-response modeling.
Subsequent data assembled for the BASF cohort provide more detailed exposure
characterization and also include information for 243 male workers employed at the plant. As
such, this study did not meet the considerations for further dose-response analysis.
2.4.1.1.1.2.2. Zober et al., 1990.
2.4.1.1.1.2.2.1. Study summary.
Zober et al. (1990) also examined the mortality patterns of 247 individuals involved in
the 1953 accident at the BASF plant. As detailed in their paper, the size of the original cohort
was expanded by efforts to locate all individuals who were exposed in the accident or during the
clean-up. Three approaches were followed in assembling the cohort. Sixty-nine cohort members
were identified from the company physician's list of employees exposed as a result of the
accident (Subcohort CI). Sixty-six of these workers were included in the original study
population of workers Thiess et al. (1982) examined. Eighty-four other workers who were
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-34 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
potentially exposed to TCDD due to their involvement in demolitions or operations were added
to the cohort. This group included 43 firemen, 18 plant workers, 7 bricklayers, 5 whitewashers,
4 mechanics, 2 roofers, and 5 individuals in other occupations (Subcohort C2). The cohort was
further augmented through the Dioxin Investigation Program, which sought to locate those who
were involved in the 1953 accident and were still alive in 1986. Current and former workers
enrolled in the study were asked to identify other current or former coworkers (including
deceased or retired) who might have been exposed from the accident. This third component of
94 workers (Subcohort C3) included 27 plant workers, 16 plumbers, 10 scaffolders,
10 professionals, 7 mechanics, 6 transportation workers, 5 bricklayers, 5 laboratory assistant,
3 insulators, and 5 individuals in other occupations. A medical examination was performed for
those identified through the Dioxin Investigation Program, and blood measures were obtained for
28 of these workers.
External comparisons of the workers' mortality experience to the general population of
the Federal Republic of West Germany were made using SMRs. Person-years were tabulated
across strata defined by calendar period, sex, and age group. Sixty-nine deaths including
twenty-three from cancer were detected among the workers during the 34-year follow-up period
(November 17, 1953 through December 31, 1987). Cause-specific death rates for these same
strata were available for the Federal Republic of West Germany. Stratified analyses were
conducted to examine variations in the SMRs according to years since first exposure (0-9,
10-19, and >20 years) for each of the three subcohorts, as well as 114 workers with chloracne.
Although it was consistent in magnitude with findings from the NIOSH cohort, a
statistically significant SMR for all cancer mortality was not observed (SMR = 1.17,
90% CI = 0.80-1.66). The SMRs for each of the three subcohorts varied substantially. For
Subcohorts CI, C2, and C3, the SMRs were 1.30 (90% CI = 0.68-2.26), 1.71
(90% CI = 0.96-2.83), and 0.48 (90% CI = 0.13-1.23), respectively. The SMRs increased
dramatically when analyses were restricted to those with 20 or more years since first exposure in
Subcohort CI (SMR = 1.67, 90% CI = 0.78-3.13) and Subcohort C2 (SMR = 2.38,
90% CI = 1.18-4.29). Meanwhile, in a subgroup analysis of those with chloracne, for the period
of 20 or more years after first exposure, a statistically significant excess in cancer mortality was
noted (SMR = 2.01; 90% CI = 1.22-3.15).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-35 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.2.2.2. Study evaluation.
An important limitation of the study is the manner in which the cohort was constructed.
Subcohort C3 was constructed by identifying individuals who were alive in 1986. This resulted
in 97 active and retired employees who participated in the program, with 94 included in the
analysis. Although these individuals did identify other workers who might have also retired or
died, inevitably, some individuals who had died were not included in the cohort. This would
serve to underestimate the SMRs that were generated with external comparisons to the German
population. Indeed, cancer mortality rates in this subcohort were about half of what would have
been expected based on general population rates (SMR = 0.48, 90% CI = 0.13-1.23).
Additionally, more than half of Subcohort C2 were firemen (43 of 84), who would likely have
been exposed to other carcinogens as a consequence of their employment. Quantitative analyses
of epidemiologic data for firefighters have demonstrated increased cancer risk for several
different forms of cancer (Youakim, 2006). Therefore, potential confounding from other
occupational exposures of the firefighters could have contributed to the higher SMR in
Subcohort C2 cohort and is a concern. Data on cigarette smoking were not available either. No
excess for nonmalignant respiratory disease was found, however, suggesting this might not be an
important source of bias.
2.4.1.1.1.2.2.3. Suitability of data for TCDD dose-response modeling.
As with the Thiess et al. (1982) publication, worker exposure was not estimated. Lack of
exposure estimates precludes a quantitative dose-response analysis using these data. Also, the
study design is not well suited to characterization of risk using the SMR statistic. Mortality is
also likely under-ascertained in the large component of the cohort that was constructed through
the identification of surviving members of the cohort.
2.4.1.1.1.2.3. Ott and Zober, 1996.
2.4.1.1.1.2.3.1. Study summary.
Ott and Zober (1996) extended the analyses of the BASF cohort to include estimates of
individual-level measures of TCDD. The researchers also investigated associations with cancer
mortality and identified incident cancer cases. The cohort follow-up period of 39 years extended
until December 31, 1992, adding 5 years to a previous study (Zober et al., 1990). Ott and Zober
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-36 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
(1996) identified incident cases of cancer using occupational medical records, death certificates,
doctor's letters, necropsy reports, and information from self-reported surveys sent to all
surviving cohort members. Self-reported cancer diagnoses were confirmed by contacting the
attending physician.
This study characterized exposure by two methods: (1) determining chloracne status of
the cohort members and (2) estimating cumulative TCDD (j_ig/kg) levels. In 1989, serum
measures were sought for all surviving members of the 1953 accident, and serum TCDD levels
were quantified for 138 individuals. These serum levels were used to estimate cumulative
TCDD concentrations for all 254 members of the accident cohort. Ott et al. (1993) published a
description of the exposure estimation procedure, which was a regression model that accounted
for the circumstances and duration of individual exposure. The average internal half-life of
TCDD was estimated to be 5.8 years based on repeated serum sampling of 29 individuals. The
regression model allowed for this half-life to vary according to the percentage of body fat, and
yielded half-lives of 5.1 and 8.9 years among those with 20% and 30% body fat, respectively.
Previous analyses of this cohort had used a half-life of 7.0 years (Ott et al., 1993).
TCDD half-life has been reported to increase with percentage of body fat in both
laboratory mammals (Geyer et al., 1990) and humans (Zober and Papke, 1993). Ott and Zober
(1996) contend that observed correlations with chloracne severity and cumulative estimates of
TCDD exposure indirectly validated this exposure metric. Specifically, the mean TCDD
concentration for those without chloracne was 38.4 ppt; for those with moderate and severe
forms of chloracne, the mean was 420.8 ppt and 1,008 ppt, respectively.
Unlike for the NIOSH cohort, individual-level data were collected for other cancer risk
factors. These factors included body mass index at time of first exposure, history of
occupational exposure to P-naphthylamine and asbestos, and history of smoking. Smoking data
were available for 86% of the cohort. SMRs were based on the external referent population of
West Germany. For cancer incidence, Ott and Zober (1996) generated standardized incidence
ratios (SIRs) using incidence rates for the state of Saarland (1970-1991) as the external referent.
They calculated SMRs (and SIRs) for three categories of cumulative TCDD levels: <0.1 |ig/kg,
0.1-0.99 (J,g/kg and >1 |ig/kg. The Cox regression model was used to characterize risk within
the cohort using a continuous measure of TCDD. These analyses considered the potential
confounding influence of age, smoking, and body mass index using a stepwise regression
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-37 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
modeling approach. The Cox modeling employed a stratified approach using the date of first
exposure to minimize possible confounding between calendar period and exposure. The three
first exposure groups were exposure within the first year of the accident, exposure between
1 year after the accident and before 1960, and exposure after 1959. The Cox regression
estimates were presented in terms of conditional risk ratios (i.e., hazard ratios adjusted for body
mass index, smoking and age).
Although no statistically significant excesses relative to the general population were
detected for all cancer mortality, there was some suggestion of an exposure-response
relationship. In the 0.1-0.99 (J,g/kg and >1 (J,g/kg exposure groups, the all cancer SMRs were 1.2
(95% CI = 0.5-2.3) and 1.6 (95% CI = 0.9-2.6), respectively. Higher SMRs for cancer (all sites
combined) were also found with an increased interval since exposure first occurred.
Specifically, when observed versus expected counts of cancer were compared in the time interval
20 years after first exposure, the SMR in the highest exposure group (>1 |ig/kg) was 1.97
(95%) CI = 1.05-5.36). An excess in lung cancer also was noted with the same lag in this
exposure group (SMR = 3.06, 95% CI = 1.12-6.66). For cancer incidence, a statistically
significant increased SIR for lung cancer was observed in the highest exposure category
(SIR = 2.2, 95% CI = 1.0-4.3), but no other statistically significant associations were detected
for any other cancer site. No cases of soft-tissue sarcoma were found among the cohort members
in this analysis.
Based on internal cohort comparisons, Cox regression models also were used to generate
hazard ratios as measures of relative risk for TCDD exposures following adjustment for
smoking, age and body mass index. A statistically significant association between TCDD dose
(per |ig/kg) and cancer mortality was detected (RR = 1.22, 95% CI = 1.00-1.50), but not for
cancer incidence (RR = 1.11, 95% CI = 0.91-1.35). Statistically significant findings were
observed for stomach cancer mortality (RR = 1.46, 95% CI = 1.13-1.89) and incidence
(RR = 1.39, 95% CI = 1.07-1.69).
The Ott and Zober (1996) study also compared the relationship between TCDD exposure
categories and cancer mortality from all sites combined according to smoking status.
Associations were noted between increased exposure to TCDD and mortality from cancer among
smokers, but not among nonsmokers or former smokers.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-38 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.2.3.2. Study evaluation.
The Ott and Zober (1996) study characterizes exposure to TCDD at an individual level.
Therefore, unlike in past studies involving this cohort, these data can provide an opportunity for
conducting quantitative dose-response modeling. As with the more recent studies involving the
NIOSH cohort, serum samples were obtained from surviving cohort members and then used to
back-extrapolate TCDD values for all cohort members. In the BASF cohort, however, serum
data were available for a much higher percentage of cohort members (54%) than in the NIOSH
cohort (5%). An additional study strength was the collection of questionnaire data, which
allowed for the potential confounding from cigarette smoking and body mass index to be
examined.
The Ott and Zober (1996) study also evaluates the relationship between TCDD and
cancer incidence. Most cohort studies of TCDD-exposed workers have relied solely on mortality
outcomes. The availability of incidence data better allows for period of latency to be described,
and moreover, to characterize risks associated with cancers that typically have long survival
periods. The authors provide few details on the expected completeness of ascertainment for
incident cancer cases, which makes determining any associated bias difficult. They do, however,
suggest that nonfatal cancers are more likely to have been missed in the earlier part of the
follow-up. The net result of differential case ascertainment over time makes evaluating
differences in risk estimates across different periods of latency impossible.
The small sample size of the cohort (n = 243 men) likely limited the statistical power to
detect small associations for some of the exposure measures. This also effectively limited the
ability to analyze dose-response relationships quantitatively, particularly across strata such as
time since exposure. For site-specific analyses, the cancer site with the most cancer deaths was
the respiratory system (n = 11). Thus, quantitative dose-response analysis using these cohort
data would be limited to the evaluation of all cancer sites combined.
The most important limitation of this study is related to the construction of the third
component of the cohort. As mentioned earlier, this cohort was assembled by actively seeking
out surviving members of the cohort in the mid-1980s. The mortality experience of this cohort is
much lower than that of the general population over the entire follow-up, a result that is expected
given that the individuals were known to be alive as of 1986. The net result is likely an
underestimate of the SMR.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.2.3.3. Suitability of data for TCDD dose-response modeling.
This study was included in the quantitative dose-response modeling for the 2003
Reassessment (U.S. EPA, 2003). The characterization of exposure data and availability of other
risk factor data at an individual level are appropriate for use in quantitative dose-response
analyses.
2.4.1.1.1.3. The Hamburg cohort.
The Hamburg cohort has been the subject of several cancer risk assessments. As with the
NIOSH and BASF cohorts, analyses have progressed from basic comparisons of mortality
experience to general population rates to more sophisticated internal cohort analyses involving
the reconstruction of TCDD exposures using serum measures. This cohort consists of
approximately 1,600 workers who were employed in the production of herbicides at a plant in
Hamburg, Germany during 1950-1984 (Flesch-Janys et al., 1995; Becher et al., 1998). The
herbicides produced included 2,4,5-T, P-hexachlorocyclohexane and lindane. The production of
TCP and 2,4,5-T was halted in 1954 following a chloracne outbreak. The plant ceased
operations in 1984. Approximately 20 different working areas were identified, which, in turn,
were grouped into five main areas based on putative TCDD exposure levels. One working area
was deemed to be extremely contaminated, having TCDD exposures at least 20-fold higher than
in other areas. In this section, the studies undertaken in this cohort that have examined cancer
mortality are summarized.
2.4.1.1.1.3.1. Manz et al., 1991.
2.4.1.1.1.3.1.1. Study summary.
Manz et al. (1991) investigated patterns of mortality in the Hamburg cohort. The study
population consisted of 1,583 workers (1,184 men, 399 women) who were employed for at least
three months between 1952 and 1989. Casual workers were excluded as they lack sufficient
personal identifying information thereby not allowing for associations with mortality outcomes
to be examined. Vital status was determined using community-based registries of inhabitants
throughout West Germany. Cause of death until the end of 1989 was determined from medical
records for all cancer deaths and classified based on the ninth revision of the International
Classification of Diseases (WHO, 1978). Although Manz et al. (1991) present some data on
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-40 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
cancer incidence for the cohort, the data are incomplete as information was available on only
12 cases; 93 cancer deaths were observed in the cohort.
In this study, the authors used information on production processes to group workers into
categories of low, medium, or high exposure to TCDD. This information was based on TCDD
concentrations in precursor materials, products, waste, and soil from the plant grounds, measured
after the plant closed in 1984. The distribution of workers into the low, medium, and high
exposure groups was 186, 901, and 496, respectively. The authors examined the validity of the
three exposure categories using a separate group of 48 workers who provided adipose tissue
samples. The median exposure of the 37 volunteers in the high group was 137 and 60 ng/kg in
the remaining 11. Information about chloracne in the cohort was incomplete, and, therefore, was
not used as a marker of TCDD exposure. Other surrogate measures of exposure were considered
in this study, including duration of exposure and year of first employment. For the latter
measure, employment that began after 1954 was assumed to result in much lower exposures
given that production of 2,4,5-T and TCP stopped in 1954.
External comparisons of cancer mortality were made by calculating SMRs using the
general population of West Germany as a referent. Comparisons of mortality in the cohort also
were made to a separate cohort of 3,417 gas supply workers to avoid bias from a healthy worker
effect. Vital status and cause of death in the gas supply workers were determined using the same
methods as used in the Hamburg cohort. SMRs were calculated relative to both referent
populations (West Germany and gas supply workers) across low, medium, and high TCDD
exposure groups. The comparison of mortality to the gas supply workers, however, extended
only until the end of 1985, whereas, comparisons to the general population extended until 1989.
Stratified analyses were undertaken to calculate SMRs for each of the three exposure groups for
categories of duration of employment (<20 versus >20 years) and date of entry into the cohort
(<1954 versus >1954).
When compared to the general population, overall cancer mortality was elevated in male
cohort members (SMR = 1.24, 95% CI = 1.00-1.52) but not in females (SMR = 0.80,
95% CI = 0.60-1.05). A two-fold increase in female breast cancer mortality was noted but was
of borderline statistical significance (SMR = 2.15, 95% CI = 0.98-4.09). The SMR among men
was further increased when analyses were restricted to workers who were employed for at least
20 years (SMR = 1.87, 95% CI = 1.11-2.95). Analyses restricted to those in the highest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-41 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
exposure group produced an even higher SMR for those with at least 20 years of employment
(SMR = 2.54, 95% CI = 1.10-5.00). Statistically significant excesses in risk were detected
among those who first worked before 1954, but not afterward. Furthermore, a dose-response
trend was observed across increasing exposure categories in the subset of workers employed
before 1954. The SMRs using the cohort of gas supply workers as the referent group for the low,
medium, and high groups in this subset were 1.41 (95% CI = 0.46-3.28), 1.61
(95%) CI = 1.10-2.44), and 2.77 (95%> CI = 1.59-4.53), respectively. This finding is consistent
with what was known about TCDD exposures levels at the plant, namely, that TCDD
concentrations were much higher between 1951 and 1954, with subsequent declining levels after
1954.
Generally speaking, patterns of excess mortality were similar when the cohort of gas
workers was used as a reference group. The overall SMR for men was 1.39
(95%>CI= 1.10-1.75); and was 1.82 (95%> CI = 0.97-3.11) when analyses were restricted to
workers with 20 or more years of employment. A dose-response trend also was observed across
exposure categories when analyses were restricted to those employed for at least 20 years. In
particular, with these analyses, no cancer deaths were observed among those in the lowest
exposure group, while the SMRs in the middle and high exposure groups were 1.36
(95% CI = 0.50-2.96) and 3.07 (95% CI = 1.24-6.33).
SMRs also were generated for several site-specific cancers relative to the West German
general population and the gas worker cohort. No statistically significant excesses were
observed using the general population reference. In contrast, statistically significant excesses
were observed for lung cancer (SMR = 1.67, 95% CI = 1.09-2.44) and hematopoietic system
cancer (SMR = 2.65, 95% CI = 1.21-5.03) relative to the gas workers cohort.
2.4.1.1.1.3.1.2. Study evaluation.
The Manz et al. (1991) findings indicate an excess of all cancer mortality among the
workers with the highest exposures, particularly those who worked for at least 20 years and were
employed before 1954. The findings across categories of exposure within the subsets of workers
employed for at least 20 years and before 1954, particularly using the cohort of gas supply
workers, are consistent with a dose-response relationship. These elevated cancer mortality rates
found among those employed before 1954 were likely due to higher TCDD exposures. Other
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-42 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
carcinogenic coexposures, such as benzene, asbestos, and dimethyl sulfate, could have occurred
among this population. Given that no substantial changes in the production processes at the
Hamburg plant occurred after 1954, comparable levels of these coexposures would be expected
before and after 1954. Therefore, confounding due to these coexposures is unlikely to explain
the dose-response relationship demonstrated between all cancer mortality and TCDD exposures.
No information, however, was presented on potential exposure to other dioxin-like compounds
which may confound the associations that were detected.
Detailed information on workers' smoking behaviors was not collected. Limited
evidence indicated, however, that smoking prevalence between the Hamburg cohort and the gas
supply workers cohort was quite similar. A nonrepresentative sample of 361 workers in the
Hamburg cohort and the sample of 2,860 workers in the gas supply cohort indicated that the
self-reported smoking prevalence was 73% and 76%, respectively. This suggests that the two
cohorts are comprised predominantly of smokers. The similarity in overall smoking prevalence
indicates that comparisons of cancer mortality between the two groups are not unduly influenced
by an inability to adjust for smoking.
2.4.1.1.1.3.1.3. Suitability of data for TCDD dose-response modeling.
The data compiled for the Manz et al. (1991) study do satisfy many of the considerations
for conducting quantitative dose-response analysis; health outcomes appear to be ascertained in
an unbiased manner, and exposure was characterized on an individual-level basis. However, as
demonstrated in later studies, there was a large dioxin-like compound component that was not
quantified or assessed in this study. Dose-response associations between TCDD and cancer
mortality were detected, with stronger associations observed with increased periods of latency
and for those who first worked when TCDD was at higher levels.
The size of the cohort, although not as large as the NIOSH cohort, does offer sufficient
statistical power to evaluate TCDD-related risk for cancers from all cancer sites. The data are
limited, however, for characterizing cancer risks among women; only 20 cancer deaths occurred
in the 399 women included in the cohort. It is unlikely that the findings are biased by
confounding due to cigarette smoking since dose-response patterns were strengthened when
comparisons were made to the cohort of gas supply workers rather the general population
referent where smoking rates were likely lower. The inability to account for other occupational
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-43 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
exposure when TCDD exposures were much higher (pre-1955) could result in confounding if
these other exposures were related to TCDD and the health outcomes under consideration. This
data set would be suitable for quantitative dose-response modeling if the exposure
characterization of the cohort could be improved using biological measures of dose.
2.4.1.1.1.3.2. Flesch-Janys et al., 1995.
2.4.1.1.1.3.2.1. Study summary.
In 1995, Flesch-Janys et al. published an analysis of the male employees from the
Hamburg cohort that extended the follow-up to 40 years (1952-1992). Inclusion of these three
additional years of follow-up resulted in a sample size of 1,189 male workers.
The authors estimated a quantitative exposure variable for concentrations of TCDD in
blood at the end of exposure (i.e., when employment in a department ended) and above German
median background TCDD levels. The TCDD exposure assessment defined 14 production
departments according to TCDD levels in various products in the plant, in waste products, and in
various buildings. The time (in years) each worker spent in each department then was
calculated. Concentrations of TCDD were determined in 190 male workers using serum
(n = 142) and adipose tissue samples (n = 48). The authors used a first-order kinetic model to
calculate TCDD levels at the end of exposure for the 190 workers with available poly chlorinated
dibenzo-p-dioxin (PCDD) and -furan (PCDF) at various time points. Half-lives were calculated
from an elimination study of 48 workers from this cohort, and the median TCDD background
level was estimated at 3.4 ng/kg blood fat from the German population (Papke et al., 1994;
Flesch-Janys et al., 1994). Using the one-compartment, first-order kinetic model, the half-life of
TCDD was estimated to be 6.9 years (Flesch-Janys, 1997). Increased age and higher body fat
percentage were associated with increased TCDD half-life, while smoking was associated with a
higher decay rate for most of the congeners examined (Flesch-Janys et al., 1996). Cumulative
TCDD exposures were estimated by summing exposures over the time spent in all production
departments and were expressed in terms of ng/kg of blood fat. The authors also applied a
metric of total toxicity equivalence (TOTTEQ) as the weighted sum of all congeners where
weights were TEQs that denoted the toxicity of each congener relative to TCDD.
Similar to previous analyses on this cohort, comparisons were made using an external
referent group of workers from a gas supply company (Manz et al., 1991). In contrast to
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-44 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
previous analyses where SMR statistics were generated using this "external" reference, however,
Flesch-Janys et al. (1995) used Cox regression. The Cox regression models treated the gas
worker cohort as the referent group, and six exposure groups were defined by serum-derived
cumulative TCDD estimates. The groups were determined by using the first four quintiles with
the upper two exposure categories corresponding to the ninth and tenth deciles of the cumulative
TCDD. Internal cohort comparisons used those workers in the lowest quintile as the referent
group, as opposed to the cohort of gas workers. A similar approach was used to model TEQs.
No known TCDD exposures occurred in the gas workers, so they were assigned exposures based
on the median background levels in the general population. RRs were calculated based on
exposure above background levels; in other words, background levels were assumed to be
equivalent across all workers and also for those employed by the gas supply company. The RRs
derived using the Cox model were adjusted for total duration of employment, age, and year when
employment began.
The Cox regression with the cohort of gas workers as the referent exposure group yielded
a linear dose-response relationship between cumulative TCDD exposure and cancer mortality for
all sites combined (p < 0.01). The RRs for all-cancer mortality were 1.59, 1.29, 1.66, 1.60, 1.70,
and 3.30. For four of the six categories (excluding the referent group), the RRs were statistically
significant (p < 0.05); in the highest TCDD exposure category (344.7-3890.2 ng/kg) the RR was
3.30 (95% CI = 2.05-5.31). Similar findings were evident with TOTTEQ. A dose-response
pattern for all cancer mortality (p < 0.01) based on the internal cohort comparisons was also
detected.
The authors performed an additional analysis to evaluate the potential confounding role
of dimethylsulfate. Although no direct measures of dimethylsulfate were available, the
investigators repeated analyses by excluding 149 workers who were employed in the department
where dimethylsulfate was present. A dose-response pattern persisted for TCDD (p < 0.01), and
those in the highest exposure group (344.7-3,890.2 ng/kg of blood fat) had a RR of 2.28
(95% CI = 1.14-4.59).
2.4.1.1.1.3.2.2. Study evaluation.
The Flesch-Janys et al. (1995) study used serum-based measures to determine cumulative
exposure to TCDD at the end of employment for all cohort members. They used the standard
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-45 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
one-compartment, first-order kinetic model and samples obtained from 190 male workers. This
quantitative measure of exposure permits an estimation of a dose-response relationship.
Confounding for other occupational exposures is unlikely to have biased the results. A
dose-response relationship persisted after excluding workers exposed to dimethyl sulfate. Other
potential exposures of interest included benzene and isomers of hexachlorocyclohexane.
Exposure to these agents, however, was highest in the hexachlorocyclohexane and lindane
department, where TCDD exposures were lower. As outlined earlier, the study findings are
unlikely to be biased for cigarette smoking as cigarette smoking in the cohort was similar to that
in the comparison population. Moreover, more recent analyses of serum-based TCDD exposure
measures found no correlation with smoking status in this cohort (Flesch-Janys et al., 1995)—a
necessary condition for confounding.
The authors used an exposure metric that described cumulative TCDD exposure of
workers at the time they were last exposed. As a result, the authors were unable to characterize
risks associated with this metric for different periods of latency despite a sufficient follow-up
period. Subsequent analyses constructed time-dependent measures of cumulative TCDD and
accounted for excretion of TCDD during follow-up.
In contrast to most risk assessments of TCDD exposure, this study modeled the
relationship between other dioxin-like compounds and the risk of cancer mortality using the
TOTTEQ metric.
2.4.1.1.1.3.2.3. Suitability of data for TCDD dose-response modeling.
The data used in this study satisfy most of the considerations developed for performing a
quantitative dose-response analysis. However, latency period was not examined in this study.
Dose-response analyses were, therefore, limited to a subsequent study of this cohort (Becher et
al., 1998), which did examine latency.
2.4.1.1.1.3.3. Flesch-Janys et al., 1998.
2.4.1.1.1.3.3.1. Study summary.
Flesch-Janys et al. (1998) undertook another analysis on this cohort that incorporated
additional sera data for 275 workers (39 females and 236 males). The follow-up period was the
same as that used in the 1995 analyses, with mortality follow-up extending until December 31,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-46 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
1992. Analyses were based on 1,189 males who were employed for at least 3 months from
January 1, 1952 onward. The authors continued this dose-response analysis to address
limitations in their previous work. One limitation was that the previous method did not account
for the elimination of TCDD while exposures were being accrued during follow-up. A second
limitation was that the amount of time workers spent in different departments was not
considered. In the 1998 study, the "area under the curve" approach was used because it accounts
for variations in concentrations over time and reflects cumulative exposure to TCDD. The
authors used a first-order kinetic model to link blood levels and working histories to derive
department-specific dose rates for TCDD. The TCDD background level of 3.4 ng/kg blood fat
for the German population was used (Papke et al., 1994). The dose rates were applied to
estimate the concentration of TCDD at every point in time for all cohort members. A cumulative
measure expressed as ng/kg blood fat multiplied by years was calculated and used in the SMR
analysis. SMRs were calculated using general population mortality rates for the German
population between 1952 and 1992. No lag period was incorporated into the derivation of the
SMRs. The SMRs were estimated for the entire cohort and for exposure groups based on
quartiles obtained from the area under the curve. Linear trend tests were also performed. The
overall SMR for cancer mortality in the cohort was 1.41 (95% CI = 1.17-1.68). This SMR value
was higher than the SMR of 1.21 reported for this same cohort with 3 fewer years of follow-up
(Manz et al., 1991). In terms of site-specific cancer mortality, excesses were found for
respiratory cancer (SMR = 1.71, 95% CI = 1.24-2.29) and rectal cancer (SMR = 2.30,
95% CI = 1.05-2.47). Increased risk for lymphatic and hematopoietic cancer (SMR = 2.16,
95% CI = 1.11-3.17) were also noted largely attributable (SMR = 3.73, 95% CI = 1.20-8.71) to
lymphosarcoma (i.e., non-Hodgkin's lymphoma). A dose-response relationship was observed
across quartiles of cumulative TCDD for all-cancer mortality (p < 0.01). The SMRs for these
quartiles were 1.24, 1.34, 1.34, and 1.73. Dose-response relationships were not observed for
lung cancer or hematopoietic cancers using this same metric. Dose-response relationships were
not observed with cumulative TEQ for any of the cancer sites examined (i.e., all cancers, lung
cancer, hematopoietic cancer).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-47 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.3.3.2. Study evaluation.
The approach used in the Flesch-Janys et al. (1998) study offers a distinct advantage over
earlier analyses involving the same cohort. Three more years of follow-up were available, and
the characterization of exposure using the "area under the curve" better captures changes in
cumulative exposure using a person-years approach rather than cumulative TCDD at the time of
last exposure. As noted previously, other occupational exposures or cigarette smoking are
unlikely to have biased the study findings. A sufficient length of follow-up had accrued, and
dose-response associations were evident. Dioxin-like compounds were evaluated in this study.
For TCDD, the mean concentration was 101.3 ng/kg at the time of measurement. For other
higher chlorinated congeners, the corresponding mean (without TCDD) was 89.3 ng/kg.
2.4.1.1.1.3.3.3. Suitability of data for TCDD dose-response modeling.
The data used in this study satisfy most of the considerations developed for performing a
quantitative dose-response analysis. However, latency was not examined in this study.
Dose-response analyses were, therefore, limited to a subsequent study of this cohort (Becher et
al., 1998) which did examine latency and supersedes the Flesch-Janys et al. (1998) study.
2.4.1.1.1.3.4. Becher et al., 1998.
2.4.1.1.1.3.4.1. Study summary.
The Becher et al. (1998) quantitative cancer risk assessment for the Hamburg cohort was
highlighted in the 2003 Reassessment as being appropriate for conducting dose-response
analysis. The integrated TCDD concentration over time, as estimated in the Flesch-Janys et al.
(1998) study, was used as the exposure variable. Estimates of the half-life of TCDD based on
the sample of 48 individuals with repeated measures were incorporated into the model that
back-calculated TCDD exposures to the end of the employment (Flesch-Janys et al., 1996). This
method took into account the age and body fat percentage of the workers. In Becher et al.
(1998), the analysis used the estimate of cumulative dose (integrated dose or area under the
curve) as a time-dependent variable.
Poisson and Cox regression models were used to characterize dose-response
relationships. Both models were applied to internal comparisons where a person-years offset
was used and to an external comparison where an offset of expected number of deaths was used.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
The person-years offset was used to account for varying person-time accrued by workers across
exposure categories. The use of the expected number of deaths as an offset allows risks to be
described in relation to that expected in the general population. Within each classification cell of
deaths and person-years, a continuous value TCDD and TEQ levels based on the geometric mean
were entered into the Poisson model. For the Cox model, accumulated dose was estimated based
on area under the curve for TCDD, TEQ, TEQ without TCDD, and P-hexachlorocyclohexane.
These other coexposure metrics were adjusted for in the Cox regression analyses. Other
covariates considered included in the models were year of entry, year of birth, and age at entry
into the cohort. A background level of 3.4 ng/kg blood fat for the German population was used
(Papke et al., 1994). A variety of latencies was evaluated (0, 5, 10, 15, and 20 years), and
attributable risk and absolute risk were estimated. The unexposed cohort of gas workers was
used for most internal analyses.
Internal and external comparisons using the Poisson model found positive associations
with TCDD exposure and mortality from all cancers combined. The slope associated with the
continuous measure of TCDD ((J,g/kg blood fat x years) for the internal comparison was 0.027
(p < 0.001), which decreased to 0.0156 (p = 0.07) after adjusting for age and calendar period.
The slope for the external comparison was 0.0163 (p = 0.055); this estimate was not adjusted for
other covariates. For TEQ, the slopes based on the internal comparisons were 0.0274 (p < 0.001)
in the univariate model and 0.0107 (p = 0.175) in the multivariate model after adjusting for age
and calendar period. The external estimate of slope for TEQ was 0.0109 (p = 0.164). Cox
regression of TCDD across six exposure categories, with a lag of 0 years, found a statistically
significant linear trend (p = 0.03) and those in the upper exposure group had a RR of 2.19
(95% CI = 0.76-6.29). These estimates were adjusted for year of entry, age at entry, and
duration of employment. A similar pattern was observed with the Cox regression analysis of
TEQ; the linear test for trend, however, was marginally statistically significant (p = 0.06).
Cox regression models that included both TCDD and TEQ (excluding TCDD) were
applied. In this model, the slope parameter for TCDD was marginally statistically significant
(P = 0.0089,p = 0.058), while the coefficient for TEQ (excluding TCDD) was not statistically
significant (P = -0,024, p = 0.70). This suggests that confounding by other dioxin-like
compounds was unlikely and the increased risk of cancer was due to TCDD exposure. For all
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-49 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
TEQs combined, the parameter estimate was generated from another Cox regression model and
was borderline statistically significant (P = 0.0078,/? = 0.066).
The authors used multiple Cox models to evaluate the effect of latency. The slope
estimates for both TCDD and TEQ increased dramatically with increasing latency. The slope
estimates for TCDD increased from 0.0096 to 0.0160 (p < 0.05) when latency was increased
from 0 to 20 years. Similar changes in the TEQ slopes were noted (0.0093 to 0.0157).
Evaluations of dose-response curves found that the best-fitting curve was concave in shape,
thereby yielding higher risk at low exposure. Differences between the fit of the class of models
considered [i.e., RR(x,P) = exp (P log(kx = 1))], however, were small.
Attributable risks were generated only for TCDD, as the data suggested no effects with
other TEQs. The additional lifetime risk of cancer assuming a daily intake of 1 pg TCDD/kg
body weight/day was estimated to range between 0.001 and 0.01.
2.4.1.1.1.3.4.2. Study evaluation.
The Becher et al. (1998) study represent perhaps the most detailed analyses performed on
any cohort to date. The findings were robust, as similar patterns were found with and without
using the gas supply worker cohort as the referent group. Exposures to other potential
confounding coexposures, such as dioxin-like compounds, were taken into account, and workers
with exposure to other carcinogens (e.g., lindane) were excluded. Furthermore, latency was
examined in this study, unlike earlier studies of this cohort.
2.4.1.1.1.3.4.3. Suitability of data for TCDD dose-response modeling.
This study was included in the quantitative dose-response modeling for the 2003
Reassessment (U.S. EPA, 2003). The data in the Becher et al. (1998) study are suitable for
conducting quantitative dose-response modeling. The exposure data capture cumulative
exposure to TCDD as well as exposures to other dioxin-like compounds. The length of the
follow-up is sufficient, and the study appears to not be subject to confounding or other types of
biases. Therefore, this study is utilized in quantitative dose-response analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-50 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.4. The Seveso cohort.
Several studies have evaluated the morbidity and mortality effects of residents exposed to
TCDD following a July 10, 1976, accidental release through an exhaust pipe at a chemical plant
in the town of Meda near Seveso, Italy. The released fluid mixture contained 2,4,5-T, sodium
trichlorophenate, ethylene glycol, and sodium hydroxide. Vegetation in the area showed
immediate signs of damage, and in the days following the accident, residents developed nausea,
headaches, eye irritation, and dermal lesions, particularly children.
This accident transported TCDD up to 6 km from the plant. Soil samples taken near the
plant revealed average levels of TCDD that ranged from 15.5 [j.g/m2to 580.4 (j,g/m2 in the most
contaminated area near the plant (referred to as Zone A) (Bertazzi et al., 2001). Zone A covered
87 hectares and extended 2,200 m south from the plant. Another, more distant contaminated
zone (Zone B) covering 270 hectares also had contaminated soil levels, but the TCDD
concentration range was much lower (1.7-4.3 (J,g/m3). A reference zone (Zone R), which
surrounded the two contaminated areas, had lower TCDD soil levels (range: 0.9-1.4 (J,g/m3) and
included approximately 30,000 residents. Following the accident, most residents in Zone A left
the area. Although residents in Zone B remained, they were under strict regulations to avoid
consuming homegrown products. In total, 736, 4,737, and 31,800 individuals lived in Zones A,
B, and R, respectively. Within days of the accident, 3,300 animals (mostly poultry and rabbits)
were found dead. Emergency slaughtering was undertaken to prevent TCDD from entering the
food chain, and within 2 years more than 80,000 animals had been slaughtered. Mechanisms
were put into place for long-term follow-up of these residents. Unlike the other studies based on
occupational cohorts, the follow-up of this population allows for risks to be characterized for
females.
The mortality studies from Seveso published to date have not incorporated serum TCDD
levels that were measured in individuals. Needham et al. (1997) describe the collection of serum
samples from a sample of the exposed population and control subjects in 1976. In 1988, human
exposure to TCDD was assessed by measuring small volumes of serum remaining from medical
examinations done in 1976. An examination of these data revealed some of the highest serum
TCDD levels ever reported, that the half-life of TCDD in this population was between 7 and
8 years, and that half-life varied between women and men. The half-life of TCDD in serum was
longer in women (~9 years) than in men (~7 years) (Needham et al., 1994). In this report, the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-51 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
findings of studies that characterized cancer risks in relation to exposure to TCDD from the 1976
accident are highlighted. These studies include comparisons of cancer mortality rates to the
general population based on zone of residence at the time of accident (Bertazzi et al., 2001;
Consonni et al., 2008). More recent work done by Warner et al. (2002) investigated the
relationship between serum-based measures of TCDD and breast cancer among participants in
the Seveso Women's Health Study (SWHS).
2.4.1.1.1.4.1. Bertazzi et al., 2001.
2.4.1.1.1.4.1.1. Study summary.
Several studies have reported on the mortality experience of Seveso residents. The more
recent publications having a longer follow-up of the cohort are evaluated here. In 2001, the
findings from a 20-year mortality study of Seveso residents was published (Bertazzi et al., 2001).
The Bertazzi et al. (2001) study was an extension of the 10- and 15-year follow-ups for mortality
(Bertazzi et al., 1989, 1997; Pesatori et al., 1998) and the 10-year follow-up for cancer incidence
(Bertazzi etal., 1993).
In this cohort, TCDD exposures were assigned to the population using a three-level
categorical variable representative of the individual's place of residence (Zones A, B, or R) at the
time of the accident or when the person first became a resident of the zone, if that was after
1976. An external comparison to the province of Lombardy was made by generating rate ratios
(RR) using Poisson regression techniques. Person-years of follow-up were tabulated across
strata defined by age, zone of residence, duration of residence, gender, calendar time, and
number of years that had elapsed since the time of exposure. Mortality rates during the
preaccident period also were compared to evaluate potential changes in rates due to the accident
and to evaluate whether patterns were consistent before and after the accident.
No overall excess in mortality rates from all cancer sites combined was observed in
Zones A or B (combined) when compared to the reference population of Lombardy
(n = 9 million residents) (RR = 1.0, 95% CI = 0.9-1.2). Analyses of site-specific cancer
mortality revealed statistically significant excesses among residents in Zones A or B (combined)
for cancer of the rectum (RR = 1.8, 95% CI = 1.0-3.3) and lymphatic and hematopoietic
malignancies (RR = 1.7, 95% CI = 1.2-2.5). Lymphatic and hematopoietic malignancies were
elevated in women (RR = 1.8, 95% CI = 1.1-3.2) and in men (RR = 1.7, 95% CI = 1.0-2.8).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-52 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Analyses stratified by the number of years since first exposure (i.e., 1976) revealed
higher risk among men with an increased number of years elapsed. Similar to other studies, the
RR for all cancers (combined) was 1.3 (95% CI = 1.0-1.7) among men 15-20 years after first
exposure. No such increase after 15 years postexposure, however, was noted in women
(RR = 0.8, 95% CI = 0.6-1.2).
2.4.1.1.1.4.1.2. Study evaluation.
Ascertainment of mortality appears to be excellent. Vital status was established using
similar methods for both the exposed and reference populations. No individual data were
collected and, therefore, the possibility that confounding by individual characteristics such as
cigarette smoking cannot be entirely dismissed. Bertazzi et al. (2001) do note that the
sociodemographic characteristics of residents in the three zones were similar based on
independently conducted surveys, and no differences in chronic respiratory disease were found
across the different zones. If excess mortality was attributable to cigarette smoking, such
excesses would be expected to be evident during the entire study period. Latency analyses
revealed elevated risks 15-20 years postaccident. Finally, no excesses were observed for other
smoking-related cancers of the larynx, esophagus, pancreas, and bladder. The observed excesses
in all cancer mortality do not appear to be attributed to differential smoking rates between the
two populations.
To examine potential for bias due to noncomparability in the two study populations, a
comparison of cancer mortality rates between the Seveso regions and the reference population of
Lombardy was conducted. Elevated rates for brain cancer mortality were noted in Seveso
relative to Lombardy, but the higher rates of leukemia mortality were found in Lombardy
relative to Seveso. That no excess was reported for all cancer sites combined lends credence to
the hypothesis that the exposure to TCDD from the accident increased rates of cancer after a
sufficient period of latency.
Stratified analyses were performed across several categorical variables including gender
and time since exposure. The numbers of cancer site-specific deaths are quite small in many of
the 5-year increments since first exposure. The study, therefore, has limited statistical power to
detect differences in mortality rates among the comparison groups for many cancer sites.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-53 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
Bertazzi et al. (2001) assigned exposures based on zone of residence. Soil sampling
within each zone revealed considerable variability in TCDD soil levels within each zone.
Moreover, some individuals would have left the area shortly after the accident, and determining
the extent to which individuals in Zone B who were subject to the recommendations near the
time of the accident adhered to them is difficult. As a result, exposure misclassification is
possible, and the use of individual measures of TCDD level in serum is preferred over zone of
residence for determining exposure. As noted by the authors, the study is better suited to "hazard
identification" than to quantitative dose-response analysis.
2.4.1.1.1.4.1.3. Suitability of data for TCDD dose-response modeling.
Given the variability in soil TCDD levels within each zone and the lack of individual
level, no effective dose can be estimated for quantitative dose-response analyses. Uncertainty in
identifying the critical exposure window for the Seveso cohort is a key limitation. The
evaluation of this study indicates that this study is not suitable for quantitative dose-response
analysis.
2.4.1.1.1.4.2. Pesatori et al., 2003.
2.4.1.1.1.4.2.1. Study summary.
Pesatori et al. (2003) published a review of the short- and long-term studies of morbidity
and mortality outcomes in the Seveso cohort in 2003. This paper presented external comparisons
of cancer incidence from 1977 to 1991 for Seveso males and females separately. As in the
Bertazzi et al. (2001) study, RRs were estimated using Poisson regression. For males who lived
in Zones A and B, the only statistically significantly elevated RR was for lymphatic and
hematopoietic cancers (RR = 1.9, 95% CI = 1.1-3.1). This excess was due primarily to
non-Hodgkin's lymphoma, which accounted for 8 of the 15 incidence cases (RR = 2.6,
95% CI = 1.3-5.3). Among females living in Zones A and B, higher rates were observed for
multiple myeloma (RR = 4.9, 95% CI = 1.5-16.1), cancer of the vagina (RR = 5.5,
95% CI = 1.3-23.8), and cancer of the biliary tract (RR = 3.0, 95% CI = 1.1-8.2).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-54 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
2.4.1.1.1.4.2.2. Study evaluation.
All RRs presented in the Pesatori et al. (2003) study were based on fewer than
five incident cases.
2.4.1.1.1.4.2.3. Suitability of data for TCDD dose-response modeling.
As with the studies of mortality among Seveso residents, the Pesatori et al. (2003) study
does not capture TCDD exposure on an individual basis, and soil TCDD levels considerably vary
within each zone. Therefore, the quality of the exposure data is insufficient for estimating the
effective dose needed for quantitative dose-response analysis.
2.4.1.1.1.4.3. Consonni et al, 2008.
2.4.1.1.1.4.3.1. Study summary.
Similar analytic methods were applied with 25 years of follow-up of the Seveso cohort
(Consonni et al., 2008). An important addition in this paper was the presentation of RRs for
Zone R, which had the lowest TCDD levels. Poisson regression models were used to calculate
RRs of mortality using Seregno as the reference population. Cancer deaths observed in Zones A
and B were 42 and 244, respectively.
No statistically significant differences in all cancer mortality relative to the reference
population were noted in any of the zones (Zone A: RR = 1.03, 95% CI = 0.76-1.39; Zone B:
RR = 0.92, 95% CI = 0.81-1.05; Zone R: RR = 0.97, 95% CI = 0.92-1.02). Statistically
significant excesses in mortality from non-Hodgkin's lymphoma (RR = 3.35,
95% CI = 1.07-10.46) and multiple myeloma (RR = 4.34, 95% CI = 1.07-17.52) were observed
in the area with the highest TCDD levels (Zone A). No other statistically significant increases in
cancer mortality relative to the reference population were apparent. The absence of elevated
breast cancer mortality among women in this study was noteworthy, as this finding differs from
the results of a study of Seveso women for which TCDD exposures were estimated using serum
samples (Warner et al., 2002). A more detailed description of this study is provided later in this
section.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-55 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
2.4.1.1.1.4.3.2. Study evaluation.
Although no individual-level data on smoking were available, the potential for
confounding is likely minimal. Independent smoking surveys found that the smoking prevalence
rates in Desio, one of cities affected by the accident, were similar to those in districts just outside
the study area (Cesana et al., 1995). As mentioned earlier, one would expect elevated RRs over
the entire study period if smoking had biased the study results, and not just after 15-20 years
since exposure to TCDD.
2.4.1.1.1.4.3.3. Suitability of data for TCDD dose-response modeling.
The lack of individual-level exposure data precludes quantitative dose-response modeling
using these data.
2.4.1.1.1.4.4. Baccarelli et al., 2006.
2.4.1.1.1.4.4.1. Study summary.
Given previous findings from Seveso, Baccarelli et al. (2006) examined t(14; 18)
translocations in the DNA of circulating lymphocytes of healthy dioxin-exposed individuals.
These translocations are associated with the development of cancer, namely follicular
lymphomas. The study included 211 healthy subjects of the Seveso area, and 101 who had
developed chloracne. The investigators analyzed data from 72 high-TCDD plasma level
individuals (>10 ppt) and 72 low-TCDD plasma levels (<10 ppt). A three-level categorical
variable was used to evaluate dose-response. This variable was developed by dividing those
with exposures >10 ppt into two groups: 10- <50 ppt, and 50-475.0 ppt. Trained interviewers
administered a questionnaire that collected data on demographic characteristics, diet, and
residential and occupational history.
The prevalence of t(14; 18) was estimated as those individuals having a t(14; 18) positive
blood sample divided by the t(14; 18) frequency (number of copies per million lymphocytes).
Baccarelli et al. (2006) found that the frequency of t(14; 18) was associated with plasma TCDD
levels, but no association between TCDD and the prevalence of t(14; 18) was detected.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-56 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2.4.1.1.1.4.4.2. Study evaluation.
Whether the frequency of t(14; 18) associated with plasma TCDD levels translates into an
increased risk of lymphoma is uncertain as prospective data of TCDD on those who developed
non-Hodgkin's lymphoma are lacking. Moreover, the t(14; 18) translocation could be an
important event in the pre-B stage cell that contributes to tumorigenicity, however subsequent
exposure to carcinogenic agents might be necessary for t(14; 18) cells to develop into a
malignancy (Hoglund et al., 2004).
2.4.1.1.1.4.4.3. Suitability of data for TCDD dose-response modeling.
Given that current TCDD plasma levels were measured for this study, it is unclear if the
effects of lymphocyte translocations may be due to initial high exposure or are a function of the
cumulative exposure for a longer exposure window. Additionally, whether the frequency of
t(14; 18) associated with plasma TCDD levels translates into an increased risk of lymphoma is
unknown. Dose-response analysis for this outcome, therefore, was not conducted.
2.4.1.1.1.4.5. Warner et al., 2002.
2.4.1.1.1.4.5.1. Study summary.
To date, Warner et al. (2002) is the only published investigation of the relationship
between serum-based measures of TCDD and cancer in Seveso. Eligible participants from the
SWHS (see Section 2.4.1.2.1.4 for details) were women who, at the time of the accident in 1976,
were 40 years of age or younger, had lived in one of the most highly contaminated zones (A or
B), and had adequate sera collected soon after the explosion. Enrollment in SWHS was begun in
March 1996 and lasted until July 1998. Of the total 1,271 eligible women, 981 agreed to
participate in the study. Cancer cases were identified during interview and confirmed through
review of medical records. Information on other risk factors including reproductive history and
cigarette smoking was obtained through interview.
Serum volumes greater than 0.5 mL collected between 1976 and 1981 volume were
analyzed. Most sera were collected in 1976/77 (n = 899); samples were collected in 1978-1981
for 54 women, and in 1996/97 for 28 women. For most samples collected after 1977, serum
TCDD levels were back-extrapolated using a first-order kinetic model with a 9-year half-life
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-57 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
(Pirkle et al. 1989). For 96 women with undetectable values, a serum level that was equal to
one-half the detection level was used.
Analyses were based only on women who provided serum samples; no extrapolation of
values to a larger population was done. Risks were therefore generated using data collected at an
individual level. Serum TCDD was analyzed as both a continuous variable and a categorical
variable. The distribution of serum TCDD levels of the 15 cases of breast cancer was examined
in relation to the distribution of all women in the SWHS. The median exposure was slightly
higher among with the 15 cases of breast cancer (71.8 ppt) compared to those without (55.1 ppt),
and the exposure distribution among breast cancer cases appeared to be shifted to the right (i.e.,
the exposures were higher but followed the same distribution); however, no formal test of
significance was conducted.
Warner et al. (2002) used Cox proportional hazards modeling techniques to evaluate risk
of breast cancer in relation to TCDD serum levels while controlling for a variety of potential risk
factors. In all, 21 women had been diagnosed with cancer, and of these, 15 cases were cancer of
the breast. The analysis revealed that for every 10-fold increase in TCDD log-serum levels (e.g.,
from 10 to 100 ppt) the risk of breast cancer increased by 2.1 (95% CI = 1.0-4.6). Risk
estimates also were generated across four categories (<20, 20.1-44, 44.1-100, >100 ppt), with
the lowest category used as the reference. The RRs estimated in the third and fourth highest
exposure categories were 4.5 (95% CI = 0.6-36.8) and 3.3 (95% CI = 0.4-28.0). Although
statistical significance was not achieved for either category, likely because of the small number
of cases, the greater than threefold risk evident in both categories is worth noting. Given that the
reference category had only one incident case underscores the limited inferences that can be
drawn from these analyses. The authors adjusted for numerous potential confounders, but
observed no differences between the crude and adjusted results; the authors, therefore, presented
unadjusted risks.
2.4.1.1.1.4.5.2. Study evaluation.
The findings from the Warner et al. (2002) study differ from reports in earlier studies in
which mortality outcomes noted the absence of an SMR association. The design of this study is
much stronger than earlier ones, given the improved characterization of exposure, the ability to
compare incidence rates within the cohort, the ability to control for potential confounding
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-58 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
variables at an individual level, and the availability of incident outcomes. The use of incident
cases (versus mortality data) should also help minimize potential bias due to disease survival.
Another important advantage was the ability to measure TCDD near the time of the accident,
thereby reducing the potential for exposure measurement error.
A potentially important limitation of the Warner et al. (2002) study was that information
was collected only from those who were alive as of March 1996. Therefore, TCDD and other
relevant risk factor data could not be collected for those who had previously died of breast
cancer. Thirty-three women could not participate because they were either too ill or had died.
Of these, three died of breast cancer. Given that there were only 15 breast cancer cases, the
exclusion of these 3 cases could have dramatically impacted the findings in either direction.
Another limitation was that, at the time of the follow-up, most women were still
premenopausal and therefore, most of the cohort had not yet attained the age of greater risk of
breast cancer. An ongoing follow-up of the cohort should be completed by 2010, which should
allow for increased number of incident breast cancers to be identified. Given that the current
analyses were based only on 15 incident cases, this will substantially improve the statistical
power of the study. A secondary benefit is that the increased follow-up will allow for an
investigation of possible differential effects according to the age the women were at the time of
exposure.
2.4.1.1.1.4.5.3. Suitability of data for TCDD dose-response modeling.
Several aspects of the Warner et al. (2002) study are weaknesses in the consideration of
this study for further dose-response modeling. Only 15 cases of breast cancer were available,
and no increases in risk were found with serum TCDD exposures between 20.1 and 44 ppt
(n = 2) when compared to those with <20 ppt (// = 1). The average age at the time of enrollment
was 40.8 years while the average age at diagnosis among the cases was 45.2 years. As most
women had not yet reached the age when breast cancer cases are typically diagnosed, additional
follow-up of the cohort would improve the quantitative dose-response analysis and strengthen
this study. A key strength of this study, however, is that Warner et al. (2002) includes an
investigation of the relationship between individual serum-based measures of TCDD and cancer
in Seveso. Despite the weaknesses, this study meets the evaluation considerations and criteria
for inclusion and will be analyzed for quantitative dose-response modeling.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-59 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.5. Chapaevsk study.
Industrial contamination of dioxin in the Chapaevsk region of Russia has been the focus
of research on the environmentally-induced cancer and other adverse health effects. The
Chapaevsk region is located in the Samara region of Russia and has a population of 83,000. The
region is home to a chemical plant that produced lindane and its derivatives between 1967 and
1987, which are believed to be responsible for local dioxin contamination. Soil sampling has
demonstrated a strong gradient of increased TCDD concentrations with decreased proximity to
the chemical plant (Revich et al., 2001).
2.4.1.1.1.5.1. Revich et al., 2001.
2.4.1.1.1.5.1.1. Study summary.
Revich et al. (2001) used a cross-sectional study to compare mortality rates of Chapaevsk
residents to two external populations of Russia and the region of Samara. Mortality rates for all
cancers combined among males in Chapaevsk were found to be 1.2 times higher when compared
to the Samara region as a whole and 1.3 times higher than Russia. Similar to other studies,
statistically significant excess was noted in men (SMR = 1.8, 95% CI = 1.6-1.9) but not in
women (SMR = 0.9, 95% CI = 0.8-1.1). Among men, the excess was highest for the
smoking-related cancers of the lung (SMR =3.1, 95% CI = 2.6-3.5) and larynx (SMR = 2.3,
95% CI = 1.2-3.8) and urinary organs (SMR = 2.6, 95% CI = 1.7-3.6). Among females, there
was no increased SMR for all cancer sites combined, but excesses for breast cancer (SMR = 2.1,
95% CI = 1.6-2.7) and cancer of the cervix (SMR= 1.5, 95% CI = 1.0-3.1) were statistically
significant.
Revich et al. (2001) also compared age-standardized cancer incidence rates in Chapaevsk
to those in Samara. Although statistical tests examining these differences were not reported,
higher incidence rates were observed for all cancers combined, cancer of the lip, cancer of the
oral cavity, and lung and bladder cancer among males in Chapaevsk. Considerably lower cancer
incidence rates also were observed for prostate cancer, cancer of the esophagus, and
leukemia/lymphoma among males from Chapaevsk. Among females, incidence rates were
higher in 1998 for all cancers in Chapaevsk when compared to Russia and the Samara region, an
observation that appears somewhat counter to the presented SMR of 0.9 for all cancer mortality
from 1995-1998. Like mortality, rates of breast cancer incidence among women in Chapaevsk
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-60 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
were higher than in Russia, as were rates of cervical cancer. Leukemia/lymphoma rates were
higher among women in Chapaevsk than in those who lived in the reference populations of
Samara and Russia. This finding is contrary to the finding for males who had lower rates of
leukemia/lymphoma in Chapaevsk.
2.4.1.1.1.5.1.2. Study evaluation.
Although the Revich et al. (2001) findings suggest TCDD exposures in Chapaevsk are
quite high relative to other parts of the world (Akhmedkhanov et al., 2002), evaluation of health
outcomes to date have been based on ecological data only. This analysis did not adjust for the
influence of other risk factors (e.g., smoking, reproductive characteristics) that could contribute
to increased cancer rates for lung cancer in men and breast cancer in women. Given that both the
SMRs and SIRs for cancer outcomes vary considerably between men and women, this suggests
the possibility that occupational exposures might be a contributing factor in these adverse health
outcomes.
Future research in Chapaevsk includes plans to conduct a breast cancer case-control
study. Women who were born from 1940 onward and who have been diagnosed with breast
cancer before the age of 55 were included in the study, although the plan to characterize TCDD
using serum is uncertain (Revich et al., 2005).
2.4.1.1.1.5.1.3. Suitability of data for TCDD dose-response modeling.
This study did not meet the considerations and criteria for inclusion in a quantitative
dose-response assessment. Given the lack of exposure data on an individual basis, no effective
dose can be estimated for this study population. As such, no dose-response modeling was
conducted.
2.4.1.1.1.6. The Air Force Health ("Ranch Hands "cohort) study.
Between 1962 and 1971, the U.S. military sprayed herbicides over Vietnam to destroy
crops that opposition forces depended upon, to clear vegetation from the perimeter of U.S. bases,
and to reduce the ability of opposition forces to hide. These herbicides were predominantly a
mixture of 2,4-D, 2,4,5-T, picloram, and cacodylic acid (Institute of Medicine, 2006). A main
chemical sprayed was Agent Orange, which was a 50% mixture of 2,4-D and 2,4,5-T. TCDD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-61 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
was produced as a contaminant of 2,4,5-T and had levels ranging from 0.05 to 50 ppm (Institute
of Medicine, 1994). A series of studies have investigated cancer outcomes among Vietnam
veterans. A review of military records to characterize exposure to Agent Orange led Stellman
and Stellman (1986) to conclude that assignment of herbicide levels should not be based solely
on self-reports or a crude measure such as military branch or area of service within Vietnam.
Investigations have been performed on the Ranch Hands cohort, which consisted of those who
were involved in the aerial spraying of Agent Orange between 1962 and 1971. More elaborate
methods were used to characterize exposures among these individuals, and these studies are
summarized below.
2.4.1.1.1.6.1. Akhtar et al., 2004.
2.4.1.1.1.6.1.1. Study summary.
Akhtar et al. (2004) investigated the incidence of cancer in the Ranch Hand cohort, which
was published after the release of the 2003 Reassessment document (U.S. EPA, 2003). The
Ranch Hand Unit was responsible for aerial spraying of herbicides, including Agent Orange, in
Vietnam from 1962 to 1971. Cancer incidence in the Ranch Hand cohort were compared to a
cohort that included other Air Force personnel who served in Southeast Asia during the same
period but were not involved in the spraying of pesticides. Health outcomes were identified
during the postservice period that extended from the time each veteran left Southeast Asia until
December 31, 1999. In contrast to previous analyses of this cohort, the Akhtar et al. (2004)
study took into account concerns that both the comparison and spraying cohorts had increased
risks of cancer, and addressed the possibility that workers with service in Vietnam or Southeast
Asia might have increased cancer risk. The authors addressed the latter concern by adjusting risk
estimates for the time spent in Southeast Asia and for the proportion of time spent in Vietnam.
The Ranch Hand cohort comprised 1,196 individuals, and the comparison cohort had
1,785 individuals. The comparison cohort was selected by matching date of birth, race, and
occupation (i.e., officer pilot, officer navigator, nonflying officer, enlisted flyer, or enlisted
ground personnel). TCDD levels were determined using serum levels collected from veterans
who completed a medical examination in 1987. For those who did not have a serum measure
taken in 1987, but provided one in subsequent years, TCDD levels were back-extrapolated to
1987 using a first-order kinetic model that assumed a half-life of 7.6 years. Those with
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-62 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
nonquantifiable levels were assigned a value of the limit of detection divided by the square root
of 2. A total of 1,009 and 1,429 individuals in the Ranch Hand and comparison cohorts,
respectively, provided serum measures that were used in the risk assessment. Veterans also were
categorized according to the time their tours ended. This date corresponded to changes in
herbicide use. These categories were before 1962 or after 1972 (no herbicides were used),
1962-1965 (before Agent Orange was used), 1966-1970 (when Agent Orange use was greatest),
and 1971-1972 (after Agent Orange was used). Information on incident cases of cancer in the
cohort was determined from physical examinations and medical records. Some malignancies
were discovered at death and coded from the underlying causes of death as detailed on the death
certificate. A total of 134 and 163 incident cases of cancer were identified in the Ranch Hand
and comparison cohort, respectively. Akhtar et al. (2004) describe case ascertainment verified
by record review as being complete.
External comparisons were made based on the expected cancer experience derived from
U.S. national rates using SIRs and the corresponding 95% confidence interval. Person-years and
events were tabulated by 5-year calendar and age intervals.
When compared to the general population, no statistically significant excesses in all
cancer incidence were observed for either the Ranch Hand (SIR = 1.09, 95% CI = 0.91-1.28) or
the comparison cohort (SIR = 0.94, 95% CI = 0.81-1.10). Statistically significant differences
were found for three site-specific cancers in the Ranch Hands cohort relative to the general
population. Excesses were noted for malignant melanoma (SIR = 2.33, 95% CI = 1.40-3.65)
and prostate cancer (SIR = 1.46, 95% CI = 1.04-2.00). In contrast, a reduced SIR was found for
cancers of the digestive system (SIR = 0.61, 95% CI = 0.36-0.96). The excess in prostate cancer
was also noted in the comparison cohort (SIR = 1.62, 95% CI = 1.23-2.10) relative to the
general population. External comparisons were repeated by restricting the cohorts to the period
when Agent Orange was used (1966-1970). Again, no statistically significant excesses in all
cancer incidence were noted in the Ranch Hand (SIR = 1.14, 95% CI = 0.95-1.37) or
comparison cohort (SIR = 0.94, 95% CI = 0.80-1.11). Statistically significant excesses
continued to be observed for malignant melanoma (SIR = 2.57, 95% CI = 1.52-4.09) and
prostate cancer (SIR = 1.68, 95% CI = 1.19-2.33) in the Ranch Hand component of the cohort.
No other statistically significant differences were found among Ranch Hands personnel.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-63 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
For internal cohort analyses, veterans were assigned to one of four exposure categories.
Those in the comparison cohort were assigned to the "comparison category." Ranch Hand
veterans that had TCDD serum levels <10 ppt were assigned to the "background" category.
Those with a TCDD levels >10 ppt had their TCDD level estimated at the end of their Vietnam
service with a first-order kinetic model that used a half-life of 7.6 years. These
back-extrapolated values that were less than 118.5 ppt were assigned to a "low" exposure group,
while those with values above 118.5 ppt were classified as "high" exposure. Akhtar et al. (2004)
used Cox regression models to describe risks across the exposure groups using the comparison
category as the reference. Risks were adjusted for age at tour, military occupation, smoking
history, skin reaction to sun exposure, and eye color. Internal cohort analyses were restricted to
those who spent no more than 2 years in Southeast Asia and Ranch Hand workers who served
exclusively in Vietnam, and the comparison cohort who served exclusively outside of Vietnam.
Statistically significant excesses of cancer incidence (all sites combined) were observed
in the highest two exposure groups. A statistically significant trend test (p = 0.04) was detected
based on the RRs for the background-, low-, and high- exposure groups: 1.44
(95% CI = 0.82-2.53); 2.23 (95% CI = 1.24-4.00), and 2.02 (95% CI = 1.03-3.95). For
malignant melanoma, the RRs across the three increasing exposure categories were 2.99, 7.42,
and 7.51. The corresponding risk estimates for prostate cancer were 1.50, 2.17, and 6.04.
2.4.1.1.1.6.1.2. Study evaluation.
An important strength of this study is the manner in which TCDD exposure was
estimated. Serum data were available for most veterans, and therefore, generalizing exposure
from a small sample of cohort members is not a concern as was the case with the NIOSH and
Hamburg cohorts. Back-extrapolating to derive past exposures was based on a methodology that
has been applied in many of the cohorts, thereby facilitating risk comparisons. An additional
strength of the study is the examination of incidence as a measure of disease occurrence rather
than mortality.
In contrast to the previous analysis (Ketchum et al., 1999) the analysis by Akhtar et al.
(2004) was restricted to individuals who spent no more than 2 years in Southeast Asia. Previous
research had demonstrated that increased time spent in Southeast Asia was associated with an
increased risk of cancer. Confounding might have been introduced given that the comparison
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-64 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
cohort spent much more time in Southeast Asia than the Ranch Hands. To illustrate, the median
number of days spent in Southeast Asia was 790 for comparison cohort members, and the
median days for the Ranch Hand cohort in the background, low, and high exposure groups were
426, 457, and 397, respectively. After restricting to those who spent at most 2 years, statistically
significant associations were observed for all cancer sites combined, prostate cancer, and
malignant melanoma using the internal cohort comparisons.
An important issue in the study is the high correlation between 2,4,5-T and 2,4-D, given
that both were used in equal concentrations in Agent Orange. As a result, distinguishing the
effects of each is impossible. This point is relevant, given that 2,4-D has been associated with
prostate cancer in several studies. As a result, the dose-response association with prostate cancer
might be due to 2,4-D exposure and not TCDD. This issue also has implications for the
interpretation of the dose-response pattern for all cancer sites combined, given that incident
prostate cancers accounted for 4 of the 12 incident cases in the high-exposure group.
2.4.1.1.1.6.1.3. Suitability of data for TCDD dose-response modeling.
The ascertainment of incident cases and characterization of exposure to TCDD based on
serum measures are strong features of the cohort. Confounding by 2,4-D is a major concern.
Since delineating the independent effects of other Agent Orange contaminants is not possible,
quantitative dose-response analysis was not conducted on this study.
2.4.1.1.1.6.2. Michalek andPavuk, 2008.
2.4.1.1.1.6.2.1. Study summary.
Michalek and Pavuk (2008) recently published an updated analysis of the incidence of
cancer and diabetes in the cohort of Ranch Hand veterans. As with the Akhtar et al. (2004)
analysis, the study included a comparison cohort of other Air Force veterans who served in
Southeast Asia at the same time but were not involved with the spraying of herbicides. This
study extended previous analyses (Henriksen et al., 1997; Ketchum et al., 1999) by addressing
the number of days of herbicide spraying, calendar period of service, and the time spent in
Southeast Asia. Veterans who attended at least one of five examinations were eligible for
inclusion. Incident cancer cases also were identified from medical records.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-65 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
The methods used to determine TCDD exposures were as described above in the review
of the Akhtar et al. (2004) study. Blood measures also were taken in 1992, 1997, and 2002 for
subjects with no quantifiable TCDD levels in 1987, those who refused in 1987, and those new to
the study. TCDD dose at the end of service in Vietnam was assigned to Ranch Hands that had
TCDD levels above background using a constant half-life of 7.6 years. Each veteran was then
assigned to one of four dose categories: comparison, background (i.e., Ranch Hands with 1987
TCDD <10 ppt), low (Ranch Hands with 1987 levels 10.1-91 ppt), and high (Ranch Hands with
1987 >91 ppt). Serum TCDD estimates are available for 1,597 veterans in the comparison
cohort, and 986 veterans in the Ranch Hand cohort. The comparison cohort was selected by
matching on date of birth, race, and occupation of the Ranch Hands.
Michalek and Pavuk (2008) used Cox regression to characterize risks of cancer incidence
across the three upper exposure categories using the comparison category as the referent group.
Risk estimates were adjusted for year of birth, race, smoking, body mass index at the qualifying
tour, military occupation, and skin reaction to sun exposure. Tests for trend for increased risk of
cancer were conducted by testing the continuous covariate logioTCDD.
Overall, no association between the TCDD exposure categories and RR of all-site cancer
was observed. Those in the highest exposure group had an RR of 0.9 (95% CI = 0.6-1.4).
Stratified analyses by calendar period of service showed more pronounced risk for those who
served before 1986 (when higher amounts of Agent Orange were used). A statistically
significant dose-response trend (p < 0.01) was observed for cancer risk and logioTCDD
exposure. The RRs for the background, low, and high groups used in these comparisons were
0.7 (95% CI = 0.4-1.3), 1.7 (95% CI = 1.0-2.9), and 1.5 (95% CI = 0.9-2.6). A statistically
significant increase, however, was noted when analyses were restricted to those who had sprayed
for at least 30 days before 1967 and spent time in Southeast Asia (RR = 2.2, 95% CI = 1.1-4.4).
2.4.1.1.1.6.2.2. Study evaluation.
Michalek and Pavuk (2008) used the same study population that Akhtar et al. (2004), and
so it has the same strengths and limitations as noted above. The follow-up, however, extends an
additional 5 years (until the end of 2004). The findings for the dose-response analyses were not
as compelling as the earlier Akhtar et al. (2004) findings.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-66 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.6.2.3. Suitability of data for TCDD dose-response modeling.
The key limitation precluding dose-response analysis for the Michalek and Pavuk (2008)
study is the possible confounding from the inability to control for 2,4-D and other agents used in
Agent Orange. As such, quantitative dose-response analysis was not conducted on this study.
2.4.1.1.1.7. Other studies of potential relevance to dose-response modeling.
2.4.1.1.1.7.1. t' Mannetje et al., 2005—New Zealand herbicide sprayers.
2.4.1.1.1.7.1.1. Study summary.
t'Mannetje et al. (2005) described the mortality experience of a cohort of New Zealand
workers who were employed in a plant located in New Plymouth. The plant produced phenoxy
herbicides and pentachlorophenol between 1950 and the mid-1980s. This study population also
was included in the international cohort of producers and sprayers of herbicides that was
analyzed by IARC (Saracci et al., 1991; Kogevinas et al., 1997). In this 2005 study, analyses
were restricted to those who had worked at least 1 month; clerical, kitchen, and field research
staff were excluded. The authors followed up 1,025 herbicide producers and 703 sprayers from
1969 and 1973, respectively, until the end of 2000.
The cohort consisted of two components: those involved with the production of
herbicides and those who were sprayers. For the herbicide producers, exposures were
determined by consulting occupational history records; no direct measures of exposure were
available. Each department of employment was assigned to one of 21 codes as in the IARC
international cohort (Saracci et al., 1991). Industrial hygienists and factory personnel with
knowledge of potential exposures in this workforce classified each job according to potential to
be exposed to TCDD, other chlorinated dioxins, and phenoxy herbicides. Exposure was defined
as a dichotomous variable (i.e., exposed and unexposed). Among producers, 813 were classified
as exposed, with the remaining 212 considered unexposed.
The "sprayer" component of the cohort includes those who were registered in the national
registry of applicators at any time from January 1973 until the end of 1984. For the sprayers,
detailed occupational information was lacking. Exposure was, therefore, based on an exposure
history questionnaire completed in a previous study of congenital malformations (Smith et al.,
1982). This questionnaire, administered to 548 applicators in 1980 and 232 applicators in 1982,
achieved a high response rate (89%). Participants were asked to provide information about
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-67 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2,4,5-T-containing product use on an annual basis from 1969 up to the year the survey was
completed. As the use of 2,4,5-T ceased in the mid-1980s, data on occupational exposure to
TCDD among these workers are fairly complete. Virtually all sprayers (699 of 703) were
exposed to TCDD, higher chlorinated dioxins, and phenoxy herbicides.
Deaths among workers were identified through record linkage to death registrations in the
New Zealand Health Information Service. Electoral rolls, drivers' licenses, and social security
records also were consulted to confirm identified deaths. External comparisons of mortality
were made to the New Zealand population using the SMR statistic. The mortality follow-up for
the producers began on January 1, 1969 and extended until December 31, 2000. For the
sprayers, the follow-up period extended from January 1, 1973 until December 31, 2000. A total
of 43 cancer deaths occurred in the producer group and 35 cancer deaths occurred in the sprayer
group in the cohort. Where possible, stratified analyses by duration of employment and
department were conducted. The departments examined for producers included synthesis,
formulation and lab, maintenance and waste, packing and transport, other, and unexposed.
SMRs were generated using the New Zealand population as an external referent. A linear test
for trend was applied to evaluate dose-response trends according to categories of duration of
employment. Stratified analyses also were also done for sprayers who started working before
1973, as TCDD levels in 2,4,5-T produced at the New Zealand plant dropped dramatically after
1973. Although an SMR was presented for female producers, given that only one cancer death
was observed, this study can provide no insight on differential risks between the sexes.
Among TCDD-exposed producers, for all cancers combined, no statistically significant
excess mortality was found when compared to the general population (SMR = 1.24,
95% CI = 0.90-1.67). No dose-response trend in the SMRs for all cancers was observed with
duration of employment (p = 0.44). No statistically significant elevated SMR was observed in
any of the duration of employment categories for any of the six specific departments examined.
A statistically significant positive linear trend, however, was noted among synthesis workers
(p = 0.04). There was some suggestion of reduced mortality in the upper exposure levels for
workers in the formulation and lab departments. For sprayers, the SMR for all cancer sites
combined was not elevated relative to the New Zealand general population (SMR = 0.82,
95% CI = 0.57-1.14), nor was a dose-response pattern observed with increasing duration of
employment (p = 0.86). Additionally, no statistically significant excess in cancer mortality for
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-68 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
all sites combined was evident in workers who were first employed either before 1973
(SMR = 0.75, 95% CI = 0.50-1.07) or from 1973 on(SMR= 1.81, 95% CI = 0.59-4.22). For
site-specific analyses of cancer mortality, an excess of multiple myeloma was observed among
production workers relative to the general population (SMR =5.51, 95% CI = 1.14-16.1). This
SMR was based on three deaths. No statistically significant excess (or deficit) of mortality was
found for any other cancer site examined in either the sprayers or the producers.
2.4.1.1.1.7.1.2. Study evaluation.
The physical activity demands of spraying contribute to a healthy worker effect that
manifests itself in a lower SMR based on both external comparisons to the general population as
a referent, and the SMR generated for the producers in the cohort. The analyses conducted using
a simple dichotomy of exposure and duration of employment are limited, as nearly all of the
sprayers were unexposed.
The dose-response pattern with duration of employment coupled with the observation
that higher levels of exposure to TCDD occurred among workers in the synthesis department is
an important finding. These workers were also exposed to several other contaminants, however,
that include processing chemicals, technical products, intermediates, and byproducts (Kauppinen
et al., 1993). These included phenoxy herbicides and dioxin-like compounds such as chlorinated
dioxins. Since the dichotomous exposure measure was based on exposure to TCDD, chlorinated
dioxins and phenoxy herbicides, the associated dose-response analyses presented in this study
should be interpreted cautiously in light of the inability to either characterize or control for these
potential confounders. As such, these co-exposures might have contributed to the dose-response
pattern observed with increased duration of employment in the synthesis workers.
2.4.1.1.1.7.1.3. Suitability of data for TCDD dose-response modeling.
Although the study authors completed a subsequent analysis of this cohort using
serum-derived TCDD (McBride et al., 2009a), the lack of individual-level TCDD exposures
precludes dose-response modeling.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-69 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.7.2. McBride et al., 2009b—New Zealand herbicide sprayers.
2.4.1.1.1.7.2.1. Study summary.
McBride et al. (2009b) published an updated analysis of the mortality of the New
Zealand cohort. The follow-up period was from January 1, 1969 to December 31, 2004
extending the previous study by an additional 4 years. In contrast to the previous study where
the cohort comprised individuals employed for at least 1 month prior to 1982 (or 1984)
(t'Mannetje et al., 2005), the cohort in this study consisted of all those who worked at least one
day between January 1, 1969 and October 1, 2003. This resulted in a cohort of 1,754 workers, of
which 247 died in the follow-up interval. Seventeen percent of the cohort members were lost to
follow-up, which could be a source of selection bias if loss to follow-up was related to both the
exposure metrics and the health outcome of interest. Previous data from this cohort (t'Mannetje
et al., 2005), however, showed fairly comparable loss to follow-up rates among the unexposed
(23%) and the exposed populations (17%).
Comparisons to the New Zealand general population were made using the SMR statistic.
Stratified analyses were conducted by duration of employment (<3 months, >3 months), sex,
latency (<15 years, >15 years), and period of hire (<1976, >1976). The authors defined latency
as the period between the day last worked and the earliest of date of death, date of emigration or
loss to follow-up, or December 31, 2004.
The overall SMR for mortality from all cancer sites combined relative to the New
Zealand population was 1.01 (95% CI = 0.85-1.10). Although not statistically significant there
was suggestion of an increased risk of rectal cancer (SMR = 2.03; 95%CI = 0.88-4.01) among
the employees. SMRs for lymphatic and hematopoietic cancers (overall SMR = 1.21,
95% CI = 0.52-2.39) included 3.12 (95% CI = 0.08-17.37) for Hodgkin's disease, 1.59
(95% CI = 0.43-4.07) for non-Hodgkin's lymphoma and 3.73, 95% CI = 1.20-8.71), and 1.66
(95%) CI = 0.20-5.99) for multiple myeloma. No statistically significant excess of cancer
mortality was noted among workers employed for <3 months (SMR = 1.19,
95% CI = 0.65-2.00), or for >3 months (SMR = 0.98, 95% CI = 0.75-1.26). A statistically
significant excess of digestive cancers was found for those who worked fewer than 3 months
relative to the New Zealand population (SMR = 2.52, 95% CI = 1.15-4.78). No excesses were
observed for any site-specific cancers when analyses were restricted to those who worked for 3
or more months. No statistically significant elevated SMRs were found for all cancers
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-70 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
(combined) either for a latency period of fewer than 15 years (SMR = 1.14, 95% CI = 0.72-1.71)
or a latency period of >15 years (SMR = 0.96, 95% CI = 0.72-1.26). Similarly, no statistically
significant excess in cancer mortality was observed for all cancer sites combined, or any
site-specific cancer when analyses were stratified by date of hire (<1976, >1976) or by sex. The
SMR among women who were employed at the site was 0.68 (95% CI = 0.45-1.00).
2.4.1.1.1.7.2.2. Study evaluation.
High rates of emigration in New Zealand (9% among workers in the cohort) contributed
to a fairly high loss to follow-up (22% among workers) during the study period. The loss to
follow-up would reduce the overall mortality estimates among the workers, which could
underestimate the SMRs if loss to follow-up (and health status) was not comparable in the
general population. For example, it is unclear if workers and the general population who
emigrated were sicker than those remaining in the cohort. Previous data from the cohort workers
suggests that loss to follow-up rates were slightly higher among the low and unexposed
populations (t'Mannetje et al., 2005; McBride et al., 2009a) worker population, so presumably
the highly exposed workers were not lost to follow-up more so than other workers.
2.4.1.1.1.7.2.3. Suitability of data for TCDD dose-response modeling.
This study extended the mortality follow-up and included stratified analyses to
investigate effect modification by period of latency, sex, and date of hire. A key limitation was
the lack of direct measures of exposure for study participants which precluded estimating
effective dose needed for dose-response modeling. This study did not meet the considerations
and criteria for inclusion in quantitative dose-response analysis.
2.4.1.1.1.7.3. McBride et al., 2009a—New Zealand herbicide sprayers.
2.4.1.1.1.7.3.1. Study summary.
McBride et al. (2009a) recently published the mortality experience of the New Zealand
cohort in relation to serum estimates of TCDD levels. This study included 1,599 workers who
were employed between 1969 and November 1, 1989, which was the date that 2,4,5-T was last
used. As in their study published earlier in the same year (McBride et al., 2009b), the follow-up
period extended from the first day of employment until December 31, 2004. Vital status was
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-71 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
ascertained through record linkage to the New Zealand Health Information Service Mortality
Collection and the Registrar General's Index to Deaths for deaths up to 1990.
All current and former workers who lived within 75 km of the plant were invited to
provide serum samples. A total of 346 of the eligible workers (68%) provided samples, which
represented 22% of the overall study population (346/1599). Based on the serum measures, 70%
(241/346) had been exposed to TCDD. This percentage is similar to the estimated 71% of
workers who were deemed to have been exposed based on a review of occupational records. The
mean serum TCDD value was 9.9 ppt. The highest exposures were observed for those employed
in the trichlorophenol operation (23.4 ppt). Values among unexposed workers averaged 4.9 ppt,
which is close to the background level of 3.9 ppt among individuals of similar age in the New
Zealand general population (Bates et al., 2004). Details on smoking histories of individuals were
also collected for the 346 individuals who provided serum, allowing for an examination of the
potential confounding role that smoking might have on derived risk estimates for TCDD.
Cumulative exposure to TCDD as a time-dependent metric was estimated for each
worker. A detailed description of the methods used to derive TCDD exposure was described in
Aylward et al. (2009). The qualitative TCDD scores available for those with serum measures
were used to estimate the cumulative exposures based on a half-life of approximately 7 years. A
time-dependent estimate of TCDD exposure was derived and the area under the curve was used
to obtain cumulative workplace TCDD exposure above background levels. Model performance
appears modest as the model explained only 30% of the variance (adjusted R2) when these
TCDD exposure estimates were compared with actual serum levels (Aylward et al., 2009).
As with previous analyses of the cohort (t' Mannetje et al., 2005; McBride et al., 2009b),
external comparisons to the New Zealand general population were made using the SMR statistic.
The SMR statistic also was used to compare mortality across four exposure groups relative to the
general population, as defined by the serum TCDD estimates: 0-68.3, 68.4-475.0,
475.1-2085.7, and >2085.8 ppt-month. The proportional hazards model also was used to
conduct internal cohort comparisons across these same four exposure groups. In these analyses,
age was used as the time variable, and the covariates of date of hire, sex, and birth year were
included in the proportional hazards model. The cut-points for these four exposure categories
were chosen so that approximately equal numbers of deaths were included in each category.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-72 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Consistent with earlier SMR analyses of the same cohort, no increased cancer mortality
was observed among "ever" exposed workers in this cohort when compared to the general
population (SMR =1.1, 95% CI = 0.9-1.4). No statistically significant excess was noted for any
of the site-specific cancers, although there was some suggestion of increased risk of soft tissue
sarcoma (SMR = 3.4, 95% CI = 0.1-19.5), multiple myeloma (SMR = 2.2, 95% CI = 0.2-8.1),
non-Hodgkin's lymphoma (SMR = 1.6, 95% CI = 0.3-4.7), and cancer of the rectum
(SMR = 2.0, 95% CI = 0.7-4.4). No statistically significant increases in cancer mortality (all
sites combined) was found in any of the four exposure categories as measured by the SMR
statistic, nor was a dose-response trend noted with increasing exposure categories. No
dose-response trend (based on SMR analyses) was noted for five site-specific cancers examined
(i.e., digestive organs, bronchus, trachea and lung, soft tissue sarcomas, lymphatic and
hematopoietic tissue, and non-Hodgkin's lymphoma), although SMRs for three of the
four exposure categories exceeded 2.0.
In contrast to the external cohort comparisons, the RRs generated with the proportional
hazards model supported a dose-response trend, as rate ratios increased across increasing TCDD
exposure categories. The RRs and their 95% confidence intervals relative to the lowest of the
four groups were 1.05 (95% CI = 0.48-2.26), 1.38 (95% CI = 0.64-2.97) and 1.58
(95%) CI = 0.71-3.52). Neither the linear (p = 0.29) or quadratic (p = 0.82) test for trend,
however, was statistically significant. An increased risk of lung cancer mortality was observed
in the highest TCDD exposure category relative to the lowest (RR = 5.75,
95%) CI = 0.76-42.24). The tests for trend for lung cancer, however, also were not statistically
significant.
A smoking survey was administered to a sample of surviving workers of this cohort, and
smoking prevalence was found to be slightly higher among those with higher cumulative
exposure (61%>) compared to lower exposures (51-56%>). These minor differences in smoking
prevalence unlikely was a strong enough confounder to explain the fivefold increase in risk of
lung cancer found in the highest exposure category. Although the smoking data assessment was
a strength of the study, it was limited to only sample of workers and was not available for those
who died of lung cancer.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-73 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.1.1.7.3.2. Study evaluation.
Given high rates of emigration, loss to follow-up (22%) was a potential concern in this
study. If comparable emigration rates did occur among the general population then the SMRs
would be underestimated. It is unclear to what extent emigration occurred among the general
population and whether emigration in both the worker and general populations was dependent on
health status. If emigration rates were comparable among these two populations, the associated
bias from the under-ascertainment of mortality in the lost to follow-up group would likely
attenuate a positive association between TCDD and cancer mortality. Among the worker
population, there was not much evidence of differential loss to follow-up with respect to
exposure as average exposures were lower (3.2 ppt) among those loss to follow up compared to
those with complete follow-up (5.7 ppt). Previous studies among this population also found
slightly higher loss to follow-up rates among the unexposed (23%) compared to the exposed
(17%>) workers (t'Mannetje et al., 2005).
McBride et al. (2009a) did not present results using a continuous measure of TCDD
exposure (lagged or untagged) as was done in most other occupational cohorts. Additionally, the
modeling did not consider the use of different periods of latency.
2.4.1.1.1.7.3.3. Suitability of data for TCDD dose-response modeling.
There is no evidence that the authors considered exposure metrics that are consistent with
environmental cancer-causing agents such as exposure modeling that takes latency into account.
Given that past occupational cohort studies of TCDD-exposed workers have consistently
demonstrated stronger association with lag interval of 15 years, such an approach should be
applied to this cohort. This precludes this study from consideration for quantitative
dose-response modeling.
2.4.1.1.1.7.4. Hooiveld et al., 1998—Netherlands workers.
2.4.1.1.1.7.4.1. Study summary.
Hooiveld et al. (1998) re-analyzed the mortality experience of a cohort of workers
employed in two chemical plants in the Netherlands using 6 additional years of follow-up from
an earlier study (Bueno de Mesquita et al., 1993). The cohort consisted of those employed
between 1955 and June 30, 1985, and vital status was ascertained until December 31, 1991 (i.e.,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-74 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
36 years of follow-up). These cohort members were involved in the synthesis and formulation of
phenoxy herbicides, of which the main product was 2,4,5-trichlorophenoxyacetic acid and
monochloroacetic acid. This cohort, with a shorter follow-up interval than the original study
(t'Mannetje et al., 2005), was included in the IARC international cohort. The cohort consisted of
1,167 workers, of which 906 were known to be alive at the end of the follow-up. The average
length of follow-up was 22.3 years, and only 10 individuals were lost to follow-up.
The authors used detailed occupational histories to assign exposures. Workers were
classified as exposed to phenoxy herbicides or chlorophenols and contaminants if they worked in
selected departments (i.e., synthesis, finishing, formulation, packing, maintenance/repair,
laboratory, chemical effluent waste, cleaning, shipping-transport, or plant supervision); were
exposed to the accident in 1963; or were exposed by proximity (i.e., if they entered an exposed
department at least once a week). The 1963 accident was the result of an uncontrolled reaction
in the autoclave in which 2,4,5-trichlorophenol was synthesized; an explosion resulted, with
subsequent release of PCDDs that included TCDD. Based on these methods of exposure
assignment, 562 workers were deemed to be exposed to phenoxy herbicides or chlorophenols,
and 567 were unexposed. Due to limited information, 27 workers were classified as having
unknown exposure.
TCDD exposures also were assigned using serum measured on a sample of workers who
were employed for at least 1 year and first started working before 1975. Dioxin-like compounds
including PCDDs were also measured in the serum samples, but not analyzed for this study. Of
the 144 subjects who were invited to provide samples, 94 agreed. TCDD levels were
back-extrapolated to the time of maximum exposure using a one-compartment, first-order kinetic
model that used a half-life estimate of 7.1 years. The mathematical model used was
ln(TCDDmax) = ln(TCDD) + lag x ln(2)/7.1. The lag was defined as the number of years since
last exposure for those exposed by virtue of their normal job duties. For those exposed as a
result of the accident in 1963, the lag was defined as the number of years since the accident
occurred.
The authors made external comparisons of cohort mortality to the Netherlands population
using the SMR statistics. Poisson regression was used to perform internal cohort comparisons
using unexposed workers as the referent. RRs (measured using rate ratios) generated from the
Poisson model also were used to compare mortality based on low, medium, and high TCDD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-75 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
serum-derived categories. The Poisson model included the following covariates as adjustment
factors: age, calendar period at end of follow-up, and time since first exposure.
When compared to the general population, workers had an excess mortality from cancer
(SMR = 1.5, 95% CI = 1.1-1.9), based on 51 cancer deaths. Generally, no excesses were
observed for site-specific cancers. The exception included eight deaths from cancers of the
urinary organs (SMR = 3.9, 95% CI = 1.7-7.6). Although not statistically significant, SMRs
comparable in magnitude to other studies were detected for non-Hodgkin's lymphoma
(SMR = 3.8, 95% CI = 0.8-11.0) and Hodgkin's disease (SMR = 3.2, 95% CI = 0.1-17.6). A
statistically significant excess of cancer mortality (n=20 deaths among occupational workers)
also was also observed relative to the general population when analyses were restricted to those
exposed as a result of the 1963 accident (SMR = 1.7, 95% CI = 1.1-2.7). Three deaths from
prostate cancer were also noted among these workers (SMR = 5.2, 95% CI = 1.1-15.3), but no
excess was observed with any other cancer site.
Internal cohort comparison also demonstrated an increased risk of all cancer mortality
among those exposed to phenoxy herbicides, chlorophenols, and contaminants relative to those
unexposed (RR = 4.1, 95% CI = 1.8-9.0). A statistically significant increased risk was also
noted for respiratory cancer mortality (RR = 7.5, 95% CI = 1.0-56.1). Analyses across
categories of TCDD exposure revealed excesses in cancer mortality for all cancer sites
combined; however, no dose-response trend was apparent.
2.4.1.1.1.7.4.2. Study evaluation.
Several other studies that have characterized cohorts by TCDD levels have used the area
under the curve approach and thus have derived an exposure metric that is time dependent.
Hooiveld et al. (1998) instead created an exposure metric to capture the maximum exposure
attained during the worker's employment. Characterizing risks using this metric assumes that
other TCDD exposures accrued during a workers' lifetime are not relevant predictors of cancer
risk.
2.4.1.1.1.7.4.3. Suitability of data for TCDD dose-response modeling.
One study limitation is that although dioxin-like compounds were measured in the serum
samples, Hooiveld et al. (1998) reported associations with mortality for TCDD only. There is
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-76 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
some utility to examining dose-response analyses using alternative exposure metrics as those
constructed in this cohort. However, the small number of identified cancer deaths, limitations in
terms of the exposure assignment (based on nonrepresentative sample, and maximum exposure
level) and concern over potential confounding by co-exposures preclude using these data for a
dose-response analysis.
2.4.1.1.2. Key characteristics of epidemiologic cancer studies
See Table 2-1 at the end of the chapter for a comparison of the length of follow-up,
latency period used, the half-life for TCDD used, and the fraction of TEQs accounted for by
TCDD (when applicable) for each study.
2.4.1.1.3. Feasibility of TCDD cancer dose-response modeling—summary discussion by
cohort
2.4.1.1.3.1. Using the NIOSH cohort in dose-response modeling.
It is important to evaluate the NIOSH cohort in cancer dose-response modeling of TCDD.
This cohort is the largest assembled to date, direct measures of TCDD based on sampling are
available, and the lengthy follow-up interval allows for latent effects to be taken into account.
Further, although this cohort consists mostly of male workers, these workers were occupationally
exposed to TCDD daily, as compared to the acute accidental exposures of other occupational
cohorts. Although the most recent analyses of a subset of the NIOSH cohort showed no
association between serum TCDD levels and cancer mortality, the study authors did not examine
latency effects (Collins et al., 2009). Incorporation of latency intervals is important in light of
the stronger dose-response relationships that consistently have been observed with a 15-20 year
latency interval in previous investigations of the NIOSH and other cohorts (Steenland et al.,
2001).
Most published studies of the NIOSH cohort did not evaluate exposures to dioxin-like
compounds. An exception is the analysis by Steenland et al. (2001). Although Steenland et al.
(2001) did not incorporate individual-level data on dioxin-like compounds, based on their
previous work (Piacitelli et al., 1992) they assumed that TEQ occupational exposures occurred as
a result of TCDD alone in this population. TCDD exposures provided a better fit to the data than
the TEQ-based metric, and 15-year latencies improved the fit for both metrics (relative to
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-77 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
unlagged exposures). The lifetime risk estimates for an increase in 10 TEQs (pg/kg of body
weight/day/sex) ranged from 0.05-0.18%. The value added for this measure is the incorporation
of the contribution of other dioxin-like compounds to the background rates.
Blue collar workers, such as those in the NIOSH cohort, typically have higher rates of
smoking than the general population (Lee et al., 2007; Bang and Kim, 2001). This potential
source of confounding would be expected to produce a higher SMR for lung cancer mortality,
and could contribute to the excess noted in the cohort with longer lag intervals. This bias,
however, likely is not large as no statistically significant excess of nonmalignant respiratory
mortality was found in these workers. Any associated bias from smoking would be expected to
be smaller for comparisons conducted within the cohort, as fellow workers would be expected to
be more homogeneous with respect to their risk factor profile than with an external general
population referent group. Stratified analyses using both internal and external comparison
groups also did not identify important differences in associations with TCDD exposure between
smoking and nonsmoking cancers. Thus, fatal cancer risk estimates reported for workers in the
NIOSH cohort appear to provide a reasonable estimate of the carcinogenic potency of TCDD.
Although the Steenland et al. (2001) study did not directly account for the possible
confounding effects of other occupational exposure, the authors did address this source of
potential bias. No known occupational exposures to carcinogens occurred, with the exception of
4-aminobiphenyl, which occurred at one plant. Two deaths from mesothelioma also occurred in
the cohort, so some exposure to asbestos might also have occurred in the cohort (Fingerhut et al.,
1991). The statistical analyses suggested that the inability to control for other occupational
exposures would not have unduly affected risk estimates generated from internal cohort
comparisons. For instance, the removal of one plant at a time from the analysis did not
materially change dose-response estimates generated from the Cox model (Cheng et al., 2006).
Moreover, adding a variable to represent plant in the Cox regression had little impact on the risk
estimates. Given that other occupational exposures varied by plant, a change in risk estimates
would be expected if such exposures were strong confounders.
The Cheng et al. (2006) analysis provides important information about the impact of
applying kinetic models to the data. The CADM TCDD kinetic model resulted in dramatic
decreases in the TCDD cancer mortality risk estimates when compared to the one-stage
compartmental model that had been applied. Although Cheng et al. (2006) suggested that the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-78 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
CADM model provides a better fit to the data than the typically used simple one-compartmental
model, statistical comparisons of model fit were not reported. Therefore, there is value in
presenting the range in risk estimates across different models when characterizing dose-response
relationships.
Finally, the half-life of TCDD is generally recognized to vary according to body fat
percentage, data that were not available for the NIOSH workers. The inability to account for
between-worker variability in body fat would introduce exposure measurement error. That body
fat percentage would not be expected to correlate with cumulative exposure to TCDD exposure,
however, would limit the potential for misclassification bias. The effect of any nondifferential
exposure measurement error likely would serve to attenuate the risk estimates of the study.
2.4.1.1.3.2. Using the BASF cohort in dose-response modeling.
The availability of blood lipid data for TCDD allows for characterization of cumulative
TCDD exposures in the BASF cohort. TCDD blood lipid data were collected for 90% of the
surviving members of the cohort (138 of 154) and these serum measures were used to generate
TCDD exposure estimates for all 254 cohort members. Therefore, the potential for
misclassification from extrapolating these exposures to the entire cohort may not be as likely as
for the NIOSH cohort where sera data were available for only a small fraction of workers. These
data were, however, collected long after the accident (36 years) and had to be back-extrapolated
to derive the initial exposures.
The data on this cohort included several risk factors such as cigarette smoking and body
mass index. One advantage is that cumulative TCDD levels by body mass index can be
estimates on an individual-level basis. As expected, the derived cumulative measures appear to
compare well with severity scores of chloracne. The finding that more pronounced risks are
found 15-20 years after first exposure are also consistent with findings from several other
cohorts (Fingerhut et al., 1991; Manz et al., 1991; Bertazzi et al., 2001).
One key limitation of the BASF cohort is its relatively small sample size (n = 243), which
limits the ability to evaluate dose-response relationships for site-specific cancers. Also, the
quality of the ascertainment of cancer incidence cannot be readily evaluated as the geographic
area of the cohort is not covered by a tumor registry. Ott and Zober (1996) state that nonfatal
cancers could have been more likely to be missed in early years, which could partially contribute
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-79 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
to the larger standardized incidence ratio found for cancer with longer latencies. Commenting on
risk differences derived from incident and decedent cancer outcomes is difficult. Among those
comprising the cohort, the ascertainment of incident outcomes was recognized to be less
complete in early years. Although the ascertainment of mortality outcomes was generally
regarded to be good among the 243 workers, some workers who died or moved likely were
missed when the cohort was constructed. These deaths would have been more likely to have
occurred several years before the second component of the cohort was assembled.
The use of the SMR statistic for this study population is associated with important
sources of uncertainties. Deaths were surely missed, particularly for the third component of the
cohort that accounts for approximately 38% (94/247) of the entire cohort; this factor would serve
to underestimate the overall SMR. As mentioned before, this component of the cohort was
assembled through the recruitment of workers known to be alive in 1986. Despite this limitation,
the characterization of exposure data and availability of other risk factor data at an individual
level allow the development of quantitative dose-response analyses.
2.4.1.1.3.3. Using the Hamburg cohort in dose-response modeling.
The Hamburg cohort lacked data on cigarette smoking, and, therefore, effect estimates
could not be adjusted for this covariate. Additional analyses that excluded lung cancers resulted
in an even stronger dose-response relationship between all cancer mortality and TCDD. Serum
levels of TCDD also were also not associated with smoking status in a subgroup of these workers
(Flesch-Janys et al., 1995) suggesting that smoking is not likely a confounder of the association
between all cancer mortality and TCDD.
An important limitation of the cohort is the reliance on blood and tissue measurements of
190 workers that likely represent a highly selective component of the cohort. This subset of
workers was identified at the end of the observation period, and therefore, excludes workers who
died or could not be traced. There are uncertainties in deriving department- and period-specific
estimates for a period that extends over three decades using this number of workers.
Additionally, the criteria applied to the reference population could have introduced some bias.
Workers were included only in the reference group if they had been employed for at least
10 years in a gas supply industry. The criteria were much different for the workers who were
exposed to TCDD (only 3 months of employment). As a result, the reference group likely would
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-80 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
be more susceptible to the healthy worker effect. Internal cohort comparisons, which should be
void of such bias, however, generally produced results similar to those based on the external
comparison population. Therefore, the Becher et al. (1998) study meets the criteria and
additional epidemiological considerations for dose-response modeling.
2.4.1.1.3.4. Using the Seveso cohort in dose-response modeling.
Unlike many of the occupational cohorts that were examined, data from the Seveso
cohort are representative of a residential population whose primary exposure was from a single
TCDD release. A notable exception is the BASF cohort where workers were exposed primarily
through two accidents that occurred in the plant. The Seveso data, therefore, might permit
cancer dose-response investigations in women and children.
Uncertainty in identifying the critical exposure window for most of the outcomes related
to the Seveso cohort is a key limitation. An important feature of the Seveso cohort, however, is
that TCDD levels were much lower among those in the highest exposure zones in Seveso
(medians range from 56-136 ng/kg) (Eskenazi et al., 2004) than those in the occupational
cohorts who had TCDD exposures that were sometimes more than 1,000 ng/kg. Given these
dramatic differences in exposures, the standardized mortality ratios (after incorporating a
15-20 year latency period) for all cancer sites combined are remarkably similar between the
Seveso and the occupational cohort analyses. Perhaps more importantly, the data from Seveso
might be more relevant for extrapolating to lower levels, given that exposures to TCDD are
two orders of magnitude higher than background levels (Smith and Lopipero, 2001).
The Warner et al. (2002) study found a positive association between serum levels of
TCDD and breast cancer. As noted previously, ascertainment of incident cases for all cancers
would allow for a dose-response relationship to be evaluated. Moreover, future breast cancer
analyses in this cohort should strengthen the quantitative dose response analyses of this specific
cancer site. The strengths of the Warner et al. (2002) study outlined earlier suggest that this
study should be considered for cancer dose-response modeling.
Earlier Seveso studies likely are unsuitable for conducting quantitative risk assessment.
These previous studies used an indirect measure of TCDD exposure, namely, zone of residence.
Soil concentrations of TCDD varied widely in these three zones (Zone A: 15.5-580.4 ppt;
Zone B: 1.7-4.3 ppt; and Zone R: 0.9-1.4 ppt), which could have resulted in considerable
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-81 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
exposure misclassification. The Warner et al. (2002) study greatly improved the characterization
of TCDD exposure using serum measures, and also allowed for control of salient risk factors that
may have resulted in bias due to confounding.
At this time it is unclear whether any study has examined the relationship between cancer
and serum estimates of TCDD among Seveso males exposed from the 1976 accident.
2.4.1.1.3.5. Using the Chapaevsk related data in dose-response modeling.
Currently, individual-level exposure data are lacking for residents of this area and there is
no established cohort for which cancer outcomes can be ascertained. These limitations,
therefore, preclude the inclusion of Chapaevsk data in a quantitative dose-response analysis.
2.4.1.1.3.6. Using the Ranch Hands cohort in dose-response modeling.
An important limitation of the Ranch Hands cohort for TCDD and cancer dose-response
modeling is an inability to isolate TCDD effects from the effects of other agents found in the
associated herbicides. Exposure to other dioxin-like compounds was not estimated in this study
and could confound the previously reported associations. As such, dose-response analyses on
this population were not conducted.
2.4.1.1.4. Discussion of general issues related to dose-response modeling
2.4.1.1.4.1. Ascertainment of exposures.
Several series of epidemiological data have used serum measures to estimate TCDD
levels. Serum data offer a distinct advantage in that they provide an objective means to
characterize TCDD exposure at the individual level. The serum measures in the occupational
cohorts, however, are limited in two important ways. First, these samples are generally collected
from small subsets of the larger cohorts; therefore, using these measures to extrapolate to the
remainder of the cohort could introduce bias due to exposure misclassification. The second
limitation is related to estimating the half-life of TCDD. As noted previously, exposures to
TCDD were back-extrapolated several decades from serum samples collected among surviving
members of several cohorts. This approach was used in the NIOSH, Ranch Hands, BASF, New
Zealand, and Hamburg cohorts. The reported half-life of TCDD among these populations was
reported between 7.1 to 9.0 years and shown to vary with several individual characteristics
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-82 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
including age, body fat composition, and smoking. The derivation of half-lives from a sample of
workers, and application of these estimates to retrospectively characterize exposure can
introduce uncertainty into the lifetime exposure estimates. It is important to note, however, that
sensitivity analyses results in several studies have been fairly consistent when evaluating the
impact of half-life of TCDD (Steenland et al., 2001; Flesch-Janys et al., 1995).
A unique advantage of the Seveso study is that serum measures were taken shortly after
the accident, and therefore characterization of TCDD exposure in this population does not
depend on assumptions needed to back-extrapolate exposures several decades.
2.4.1.1.4.2. Latency intervals.
Many of the epidemiological studies indicate stronger associations between TCDD and
cancer outcomes once a latency period has been considered. Generally, risks are higher when a
lag period of 15-20 years is included. As noted previously, this observation is consistent with
many other environmental carcinogens such as radon, radiation, and cigarette smoking. That
recent exposures do not contribute to increased cancer risk provides some support that the
initiation and promotion phases might occur many years before death making recent exposures
irrelevant for these analyses. The ability to discriminate between models of varying latency,
however, was limited in many studies. The application of biologically based modeling could
provide additional important insights on which phase(s) of carcinogenesis TCDD exerts an
influence. Such modeling, however, would necessitate having data on an individual-level basis.
Ideally, this modeling would use cancer incident data rather than mortality outcomes, given that
for many cancers, the median survival time exceeds 5 years.
2.4.1.1.4.3. Use of the SMR metric.
The occupational cohorts and the studies in Seveso and Chapaevsk have made inferences
regarding the effects of TCDD on mortality using the SMR. When compared to the general
population, the healthy worker effect may result in a downward bias in the SMR. This often can
manifest as SMRs less than 1 for several causes of mortality. The effect of this bias is, however,
generally lower for cancer outcomes. Cancer outcomes, whether incidence or death, typically
occur later in life and do not generally affect an individual's ability to work at earlier ages.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-83 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
There are several approaches that can be taken to minimize potential biases introduced by
the healthy worker effect, which would account for workers being healthier than the general
population. Comparisons of mortality (or cancer incidence) can be made to other cohorts of
similar workers. If done properly, this can allow for some control of characteristics such as
sociodemographic characteristics and smoking as the two populations can be matched by these
factors. However, it may be the case that other working populations are exposed to other
harmful exposures, thereby making it difficult to estimate risk associated with a specific agent
(such as TCDD) in the cohort of interest. A second and preferred approach to control for the
healthy worker effect, should it prove feasible, is to conduct comparisons of health outcomes in
relation to exposure within the cohort. These comparisons are less likely to be influenced by
other potential confounding variables such as smoking, socioeconomic status, and other
occupational exposures that are generally more homogeneous within the cohort relative to
external populations. Moreover, the mechanisms used to identify health outcomes and follow
individuals over time are generally applied in the same manner to all cohort members. Taken
together, where different comparisons have been made to generate risk estimates, those that have
been conducted using internal cohort comparisons are preferable.
In addition to potential bias from the health worker effect, the comparison of SMRs
between studies is not always straightforward and is not recommended by some (Rothman, 1986;
Myers and Thompson, 1998). The SMR is the ratio of the observed number of deaths to the
expected number of deaths and is often referred to as the method of indirect standardization. The
expected number of deaths is estimated by multiplying the number of person-years tabulated
across individuals in the cohort, stratified by age, by rates from a reference population that are
available for the same strata. Therefore, each population cohort will have an estimated number
of cases derived using a different underlying age structure. As outlined by Rothman (1986), the
mortality rates might not be directly comparable to each other, although the impact of such bias
will be much less if the age-distribution of the cohorts is similar. While it might be reasoned that
the TCDD exposed workers would have similar age distributions this is in fact not the case
(Becher et al., 1998; Ott et al., 1993; Thiess et al., 1982). This may be due to exposure occurring
both chronically, as well as from acute exposures due to accidental releases that happened at
various times at different plants. This is evident with the Hamburg and the BASF cohorts, as
most individuals comprising the BASF cohort were employed at the time of the accident
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-84 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
(1953/1954), while most of the Hamburg cohort (852/1048) was employed after 1954; the
follow-up of these cohorts ended at approximately the same time.
The method of direct standardization allows for a more meaningful comparison of
mortality rates to be made between cohorts. With this approach, weights (usually based on age
and sex) are drawn from a standard population and are, in turn, applied to disease rates for the
same strata observed in the cohort of interest. A comparison of weighted rates between different
cohorts would then be based on the same population standard.
Despite these limitations in comparing SMRs between studies, Armstrong (1995) argues
that the comparisons are valid if the underlying stratum specific rates in each exposure grouping
are in constant proportion to external rates. Comparisons of the SMRs between studies will be
biased only if there is an interaction between age and TCDD (i.e., the RR of disease due to
exposure differs by age). For cancer outcomes, the finding that associations become stronger
after a period of latency is incorporated into the analyses suggests that this assumption does not
hold true. That is, risk estimates would be lower among young workers. Similarly, for
noncancer outcomes, some of the data from the Seveso cohort suggests differential effects
according to the age at exposure.
The use of the SMR might also be biased in that workers exposed to TCDD could be
subject to more intensive follow-up than the general population, and as a result, differential
coding biases with cause of death might occur. Moreover, some cohorts (e.g., the BASF cohort)
have been assembled, in part, by actively seeking out survivors exposed to accidental releases of
dioxins. As such, they would not include persons who have died or who were lost to follow-up.
This would result in underascertainment of deaths and SMRs developed from these data. The
use of an internal cohort comparison offers distinct advantages to overcome potential sources of
selection bias. Given these uncertainty about comparability across the different studies,
conducting a meta-analysis of cancer outcomes for TCDD using the SMR statistic is not
warranted for this analysis.
2.4.1.1.4.4. All cancers versus site-specific.
An important consideration for quantitative dose-response modeling is the application of
models for all cancers combined, or for site-specific cancers. Consistency is often lacking for
site-specific cancers, which might be due in large part to the relatively small number of cases
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-85 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
identified for site-specific cancers in the cohorts. Although the risk estimates produced for all
cancer sites have important limitations and uncertainties, the data are far more consistent in
terms of the magnitude of an association and latency intervals. The IARC evaluation has put
forth the possibility of a pleuripotential mode of action between TCDD and the occurrence of
cancer. Despite the criticism of this assertion by some (Cole et al., 2003), the general
consistency of an increased risk for all-cancer mortality across the occupational cohorts when
latency intervals have been incorporated, provides adequate justification for dose-response
quantification of all cancer sites combined.
2.4.1.1.4.5. Summary of epidemiologic cancer study evaluations for dose-response modeling.
All epidemiologic cancer studies summarized above were evaluated for suitability of
quantitative dose-response assessment using the TCDD-specific considerations and study
inclusion criteria. The results of this evaluation are summarized in a matrix style array (see
Table 2-2) at the end of this section, and descriptively in Appendix B. Table 2-4 summarizes the
key epidemiologic cancer studies suitable for further TCDD dose-response analyses.
2.4.1.2. Noncancer
In this section, the available epidemiological data that could be used in a dose-response
analysis for noncancer endpoints are evaluated. Because many of the key studies also evaluated
cancer outcomes, the noncancer studies are presented in the same order as presented in
Section 2.4.1.1. Generally, the strengths and limitations of the cancer studies also apply to the
noncancer outcomes. In this section, key features of these studies that have direct relevance to
modeling of noncancer outcomes in particular are highlighted. To reduce redundancy, a detailed
overview of many of these studies is not provided here. Instead, the reader should refer to
Section 2.4.1.1.1.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-86 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2.4.1.2.1. Noncancer cohorts.
2.4.1.2.1.1. The NIOSH cohort.
2.4.1.2.1.1.1. Steenlandet al., 1999.
2.4.1.2.1.1.1.1. Study summary.
The 1999 published report of NIOSH workers exposed to TCDD also conducted external
cohort comparisons to the U.S. general population using SMRs for mortality outcomes other than
cancer (Steenland et al., 1999). Analyses are based on 3,538 workers employed at 8 plants from
1942 to 1984. SMRs were based on a mortality follow-up that was extended until the end of
1993. Cox regression analyses were used to compare mortality risk in relation to TCDD
exposure within the cohort.
2.4.1.2.1.1.1.2. Study evaluation.
Overall, no statistically significant differences in all-cause mortality (SMR= 1.03,
95% CI = 0.97-1.08) were observed. Mortality from ischemic heart disease (SMR = 1.09,
95% CI = 1.00-1.20) and accidents (SMR= 1.25, 95% CI = 1.03-1.50) was slightly elevated.
The dose-response relationship for ischemic heart disease observed with the SMRs calculated
across septile exposure categories, however, was not statistically significant (p = 0.14). Overall,
excess risk was not evident for diabetes, cerebrovascular disease, or nonmalignant respiratory
disease using the external population comparisons. Internal cohort comparisons using the Cox
regression model were performed using 0 and 15-year lag intervals. A dose-response trend was
observed for the derived ratios across the septiles for ischemic heart disease (p = 0.05) and
diabetes (p = 0.02). For ischemic heart disease mortality, those in the upper two septiles had rate
ratios of 1.57 (95% CI = 0.96-2.56) and 1.75 (95% CI = 1.07-2.87), respectively, relative to
those in the lowest septile. In contrast, an inverse dose-response relationship was observed for
diabetes mortality. The inverse association found for diabetes is inconsistent with the positive
association reported in the Ranch Hands study (Michalek and Pavuk, 2008). However, previous
reports have questioned the use of death certificates as the means to ascertain outcome as
diabetes may be under-reported especially among descendents with diabetes who die from cancer
(McEwen et al., 2006).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-87 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2.4.1.2.1.1.1.3. Suitability of data for TCDD dose-response modeling.
The inverse association with diabetes precludes dose-response analysis for this outcome.
Although a dose-response pattern was observed for ischemic heart disease mortality, it was
borderline statistically significant, and this association was not observed in other cohorts.
Furthermore, fatal outcomes are not a suitable basis for development of an RfD. For these
reasons, dose-response analysis for this outcome is precluded.
2.4.1.2.1.1.2. Collins et al., 2009.
2.4.1.2.1.1.2.1. Study summary.
Collins et al. (2009) recently described the mortality experience of Dow employees who
worked in Midland, Michigan. This plant produced 2,4,5-trichlorophenol between 1942 and
1979, and 2,4,5-T between 1948 and 1982. The cohort consisted of 1,615 workers exposed to
TCDD from as early as 1942; the follow-up of the cohort extended until 2003.
TCDD exposures were derived using serum samples obtained from 280 surviving
individuals. A simple one-compartment, first-order pharmacokinetic model was used to estimate
time-dependent TCDD measures. The area under the curve approach was then applied to
estimate cumulative TCDD exposure above background. A half-life of 7.2 years for TCDD
based on earlier work was incorporated into the exposure estimation (Flesch-Janys et al., 1996).
Collins et al. (2009) made an external comparison of the mortality rates of the cohort to
the U.S. general population using the SMR statistic. Noncancer causes of death included all
causes, diabetes, cerebrovascular disease, nonmalignant respiratory disease, cirrhosis of the liver,
and accidents. Overall, no statistically significant difference in all-cause mortality of these
workers was detected when compared to the general population (SMR = 0.9, 95% CI = 0.9-1.0).
Except for cirrhosis of the liver (SMR = 0.4, 95% CI = 0.1—0.8), no differences were found for
any of the noncancer causes of death relative to the general population.
Internal cohort analyses based on cumulative measures of TCDD were conducted for
mortality from diabetes, ischemic heart disease, and nonmalignant respiratory disease using the
Cox regression model. These models adjusted for possible confounders such as year of hire and
birth year. No statistically significant association was found between continuous measure of
TCDD and these causes of death.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-88 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.1.2.2. Study evaluation.
Given that the external comparisons may result in bias from the healthy worker effect,
results from the internal cohort comparisons using the Cox regression model are preferred.
These analyses were performed for diabetes, ischemic heart disease, and nonmalignant
respiratory disease. TCDD levels for these workers were estimated using a simple
one-compartment pharmacokinetic model (Aylward et al., 2007). The hazard ratios generated
from the Cox regression model were not statistically significant for any of the three noncancer
outcomes modeled.
2.4.1.2.1.1.2.3. Suitability of data for TCDD dose-response modeling.
No association of an increased risk for an adverse effect was observed with any of the
noncancer outcomes. Therefore, dose-response modeling based on this population was not
conducted.
2.4.1.2.1.2. The BASF cohort.
2.4.1.2.1.2.1. Ott and Zober, 1996.
2.4.1.2.1.2.1.1. Study summary.
In 1996, Ott and Zober published a report on the mortality experience of the cohort of
243 BASF male workers who were accidentally exposed to 2,3,7,8-TCDD in 1954 or in the clean
up that followed. The mortality follow-up of this cohort extended until the end of 1992.
External comparisons of mortality were made to the German population using the SMR statistic.
Internal cohort comparisons were also made by estimating cumulative TCDD for the cohort
using serum measures that were obtained from 138 workers. Ott et al. (1993) provided a detailed
account of the methodology to estimate TCDD. Briefly, a cumulative measure of TCDD
expressed in |ig/kg was derived, by first estimating the half-life of TCDD using individuals who
had repeated serum measures; the half-life was estimated to be 5.8 years. Individual-level data
on body fat were used to account for the influence of body fat on decay rates. Half-life estimates
of TCDD varied (range: 5.1-8.9 years) and were dependent on body fat composition (20% and
30%, respectively). This approach differed from previous analysis of this cohort that used a
constant 7-year half-life (Ott et al., 1993). TCDD levels at the time of serum sampling were then
estimated as the product of TCDD concentration in blood lipid and the total lipid weight for each
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-89 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
worker. Nonlinear models then were applied to estimate the contribution of duration of exposure
to TCDD dose extrapolated to the time of exposure.
External comparisons to the German population using the SMR statistic also were
examined across dose categories. The noncancer causes of death examined by Ott and Zober
(1996) included all-cause mortality, diseases of the circulatory system, ischemic heart disease,
diseases of the digestive system, external causes, suicide, and residual causes of death. Overall,
no statistically significant differences in the SMR with the general population for all-causes of
death (SMR = 0.9, 95% CI = 0.7-1.1) were found. No statistically significant differences were
noted for any of the other causes of death examined.
Ott and Zober (1996) performed internal cohort comparisons using the Cox regression
model. These analyses found no dose-response patterns when cause-specific mortality was
examined across increasing cumulative TCDD exposure categories. Although an inverse
association for diseases of the respiratory system (SMR = 0.1, 95% CI = 0.0-0.8) was detected,
it was based only on 1 reported case. Many of these comparisons are limited by small sample
sizes as 92 deaths occurred in the cohort, and of these, 31 were from cancer. Also, the third
component of the cohort was identified primarily from former employees who were alive in
1986. As a result, the SMR based on the general population might be underestimated by the
exclusion of deceased workers.
2.4.1.2.1.2.1.2. Study evaluation.
As noted previously, caution should be exercised in the interpretation of SMR values of
noncancer outcomes as they could be influenced by the healthy worker effect. Although the
mechanism of identifying vital status appears to be excellent and unbiased, SMRs might be
underestimated for the cohort due to the manner in which they were constructed. Specifically, a
large component of the cohort was assembled by actively seeking out former workers who were
known to be alive in 1986.
2.4.1.2.1.2.1.3. Suitability of data for TCDD dose-response modeling.
No dose-response patterns were observed between TCDD and the noncancer outcomes in
the Ott and Zober (1996) study. Therefore, dose-response modeling was not conducted.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-90 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.3. The Hamburg cohort.
2.4.1.2.1.3.1. Flesch-Janys et al., 1995.
2.4.1.2.1.3.1.1. Study summary.
Flesch-Janys et al. (1995) reported on the mortality experience of a cohort of individuals
employed by an herbicide-producing plant in Hamburg, Germany, covering the period 1952 to
1992. As described in more detail in Section 2.4.1.1.1.3, the authors developed a cumulative
measure of TCDD using serum measures from 190 workers. This study also examined the
relationship between total TEQ and mortality. In the study population, the mean TEQ without
TCDD was 155 ng/kg, and for the mean TEQ including TCDD was 296.5 ng/kg.
Risks relative to the unexposed referent group of gas workers were estimated using Cox
regression across six exposed TCDD groups (i.e., the first four quintiles, and the ninth and tenth
deciles). A linear dose-response relationship was found with all causes of mortality and
cardiovascular mortality (p < 0.01). The RR for all cardiovascular deaths in the upper exposure
category was 1.96 (95% CI = 1.15-3.34), although there was no evidence of a linear
dose-response trend (p = 0.27). The dose-response relationship was most marked for ischemic
heart disease, with a RR of 2.48 (95% CI = 1.32-4.66) in the highest exposure group. A
dose-response relationship was also observed across TEQ groupings for all cause mortality,
cardiovascular disease mortality, and ischemic heart disease mortality. The authors did not
perform joint modeling of TEQ (without TCDD) and TCDD, so determining the extent that
dioxin-like compounds contributed to an increased risk of mortality is not possible.
2.4.1.2.1.3.1.2. Study evaluation.
The Flesch-Janys et al. (1995) study lacks information on other potential risk factors for
cardiovascular disease, which could result in confounding if those risk factors are also related to
TCDD exposure. Dose-response patterns were strong, however, and persisted across numerous
TCDD (and TEQ) exposure categories based on the use of an external reference group (i.e., gas
workers) or based on the internal comparison. The findings based on the internal comparison are
noteworthy in that these groups should be more homogenous with respect to confounding
factors. As noted previously, the poor correlation between TCDD and smoking among workers
and similar smoking prevalence between the workers and the external gas company workers
suggest that smoking was not likely a confounder of the TCDD and cardiovascular disease
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-91 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
relationship. No other evaluation of noncancer mortality outcomes has been undertaken in this
cohort since 1995.
A strength of the Flesch-Janys et al. (1995) study was that it included the collection of
blood serum measures, which provided an objective measure of TCDD exposure. Blood serum
data, however, were obtained only for 16% of the cohort. The assumption of the first-order
kinetic elimination model is critical, given that measures were taken at the end of follow-up. The
model also assumed the half-life of TCDD was 6.9 years. If the kinetics are not first order, or if
the half-life estimate is inaccurate, estimates of TCDD levels during exposure would be biased,
particularly for workers having longer periods between exposure and PCDD and PCDF assays.
Sensitivity analyses completed by the authors suggest that such bias is not likely to present
because the results were unaffected when different model assumptions regarding kinetic and
half-lives were examined. The lack of an impact on RR estimates with varying half-life
estimates was similar to findings by Steenland et al. (2001).
2.4.1.2.1.3.1.3. Suitability of data for TCDD dose-response modeling.
Despite the aforementioned study strengths, the study focused on fatal outcomes such as
all cause mortality, cardiovascular disease mortality, and ischemic heart disease mortality. As
such, dose-response analysis was not conducted since these outcomes are not suitable for
development of an RfD.
2.4.1.2.1.4. The Seveso Women 'v Health Study (SWHS).
Eskenazi et al. (2000) presented an overview of the SWHS. The SWHS is the first
comprehensive epidemiologic study of the reproductive health of a female population exposed to
TCDD. The primary objective of the SWHS is to investigate the relationship of TCDD and
several reproductive endpoints, including endometriosis, menstrual cycle characteristics, birth
outcomes, infertility, and age at menopause. A second phase of follow-up that focuses on
osteoporosis, thyroid hormone, breast cancer, diabetes, and metabolic syndrome is expected to be
completed in 2010.
Women were eligible for participation in the SWHS if they resided in Zones A and B (the
most contaminated areas) at the time of the explosion, were 40 years of age or younger at the
time of the explosion in 1976, and samples of their blood were collected and stored between
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-92 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
1976 and 1980. The enrollment of women in the SWHS began in March 1996 and continued
until July 1998. Of the 1,271 eligible women, 17 could not be found, 21 had died, and 12 were
too ill to participate. Of the 96% of the remaining women, 80% (n = 981) participated in the
study. Participation in the SWHS included a blood draw and an interview by a trained nurse who
was blind to subjects' TCDD level and zones of residence at the time of the accident. The
interview included detailed information on potential confounders including occupational,
medical, and reproductive, and pregnancy history. Also, women who were premenopausal were
asked to undergo a vaginal ultrasound and pelvic exam and to complete a daily diary on
menstruation.
Depending on the health outcome under study, TCDD exposures were characterized for
the women at different times. For example, TCDD exposure levels were estimated at the time of
the accident for some studies and at the time of conception for others. The SWHS study
population has been used to investigate associations between maternal TCDD levels and the
following health outcomes: menstrual cycle characteristics (Eskenazi et al., 2002a);
endometriosis (Eskenazi et al., 2002b); birth outcomes (Eskenazi et al., 2003); age at menarche
(Warner et al., 2004); age at menopause (Eskenazi et al., 2005); uterine leiomyomas (Eskenazi et
al., 2007); and ovarian function (Warner et al., 2007). An evaluation of the studies in
chronological order is presented in this section.
2.4.1.2.1.4.1. Eskenazi et al., 2002a—Menstrual cycle characteristics.
2.4.1.2.1.4.1.1. Study summary.
Eskenazi et al. (2002a) evaluated serum TCDD exposures in relation to several menstrual
cycle characteristics in the SWHS. A total of 981 women who were 40 years of age or younger
at the time of the accident comprised the SWHS. The following exclusion criteria was applied
44 years of age or older, women with surgical or natural menopause, those with Turner's
syndrome, and those who in the past year had been pregnant, breastfed, or used an intrauterine
device or oral contraceptives.
A trained interviewer collected data on menstrual cycle characteristics using a
questionnaire. Women were asked to indicate how long their cycles were, whether the cycles
were regular (e.g., irregular cycle defined as length varied by more than 4 days), how many days
the menstrual flow lasted, and whether this flow was "scanty, moderate, or heavy." Information
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-93 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
was also collected on obstetric and gynecological conditions. TCDD exposures were derived
from serum samples collected in 1976-1985. The authors selected the earliest available serum
sample, and back-extrapolated to 1976 values using either the Filser model (Kreuzer et al., 1997)
for women aged 16 years or younger in 1976 (n = 20) or the first-order kinetic model (n = 6)
(Pirkle et al., 1989).
Serum TCDD levels were transformed using the loglO scale, and the relationships
between these levels and length of menstrual cycle and days of menstrual flow were examined
using linear regression. The authors applied logistic regression to characterize the risk between
logioTCDD and heaviness of flow or regularity of cycle. In these analyses, moderate or heavy
flow and regular cycle were used as the reference categories. Stratified analysis was performed
by menarcheal status at the time of the accident.
Overall, the association with TCDD exposure (per 10-fold increase) and length of
menstrual cycle was of marginal statistical significance for premenarcheal (P = 0.93,
95% CI = -0.01, 1.86) women, but not postmenarcheal women (P = -0.03, 95% CI = -0.61,
0.54). The corresponding estimates found for days of menstrual flow were P = 0.18
(95%) CI = -0.15, 0.51) and P = 0.16 (95% CI = -0.18, 0.50), respectively. Reduced flow was
not associated with TCDD when compared to moderate or heavy flow (odds ratio [OR] = 0.84,
95%) CI = 0.44, 1.61); effect modification by menarcheal status, however, was evident (p = 0.03).
Specifically, women exposed to TCDD who were premenarcheal had lower odds of reduced
flow, while those exposed to TCDD who were postmenarcheal did not. These findings counter
the hypothesis that TCDD exposure is related to ovarian dysfunction. Finally, statistically
significant ORs were found between serum TCDD levels (per 10-fold increase) and having an
irregular cycle (OR = 0.46, 95% CI = 0.23, 0.95). This inverse association was evident in both
premenarcheal women (OR = 0.50, 95% CI = 0.18, 1.38) and postmenarcheal women
(OR = 0.41, 95% CI = 0.15, 1.16).
2.4.1.2.1.4.1.2. Study evaluation.
Overall, the findings from the Eskenazi et al. (2002a) study suggest that exposures to
TCDD can affect menstrual cycle characteristics among women who were exposed before
menarche. Exposures to TCDD were well characterized using serum samples available on an
individual-level basis, and the design allowed for the influence of other risk factor data to be
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-94 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
controlled for in regression analyses. Analysis of TCDD levels and the length of menstrual cycle
in premenarcheal women produced associations that were of marginal statistical significance.
However, it is unclear whether the endpoints that were measured constitute adverse health
outcomes as they are not definitive markers of ovarian dysfunction. Another source of
uncertainty is measurement error due to the subjective nature of menstrual flow reporting. Any
resulting misclassification of the outcome should be nondifferential, as the measurement error is
unlikely to be dependent on TCDD exposure.
2.4.1.2.1.4.1.3. Suitability of data for TCDD dose-response modeling.
The lack of a clear adverse health outcome related to TCDD exposure is a weakness of
this study. Although it is difficult to define the critical window of exposure for quantitative
exposure calculations, it can be estimated for the women that were premenarcheal at the time of
the accident as 13 years. Therefore, this study is suitable for further consideration for
quantitative dose-response modeling.
2.4.1.2.1.4.2. Eskenazi et al, 2002b—Endometriosis.
2.4.1.2.1.4.2.1. Study summary.
The SWHS provided the opportunity to investigate the association between serum TCDD
levels and endometriosis (Eskenazi et al., 2002b). The rationale the authors provided for
undertaking this study was the experimental animal studies that suggested an association, the
high prevalence of endometriosis among infertile women where breast milk concentrations of
dioxin are high, and the unknown etiology of endometriosis. The study consisted of 601 women
who were younger than 30 years at the time of the Seveso accident. Stored sera that had been
collected between 1976 and 1980 were also available for these women.
Given that laparoscopy could not be performed on women unless clinically indicated, no
"gold" standard was available for endometriosis diagnosis. Based on the results of a validation
study they conducted in a clinical population, the researchers classified women as having
endometriosis based on symptom report, gynecologic exam results, and vaginal ultrasound.
TCDD was measured in sera in 1976 for 93% of the women. Values for women whose
serum TCDD levels were collected after 1977 and had values exceeding 10 ppt were
back-extrapolated to 1976 using either the Filser model (<16 years of age) (Kreuzer et al., 1997)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-95 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
or a first-order kinetic model (>16 years) (Pirkle et al., 1989). These estimates of TCDD were
then modeled as both continuous (on a log scale) and categorical (<20, 20.1-100, and >100 ppt)
exposures.
Polytomous logistic regression was applied within the cohort used to generate RRs. In
relation to women in the lowest exposure category, the RR for endometriosis among women in
the middle and upper categories was 1.2 (95% CI = 0.3-4.5) and 2.1 (95% CI = 0.5-8.0),
respectively. The trend tests were not statistically significant for either the categorical (p = 0.25)
and continuous measures of TCDD (p = 0.84).
2.4.1.2.1.4.2.2. Study evaluation.
It is important to note that disease misclassification could have led to an underestimate of
the true risk of endometriosis if this misclassification was not differential with respect to TCDD
exposure. Also, younger women were likely to be under-represented as those who had never
been sexually active could not be examined due to cultural reasons. Other dioxin-like
compounds (PCDD, PCDFs, or poly chlorinated biphenyls [PCBs]) were not considered because
of small serum volumes, but any potential TEQ exposures occurring in the population were
thought to be mostly attributable to TCDD in the exposed women.
2.4.1.2.1.4.2.3. Suitability of data for TCDD dose-response modeling.
Given that no statistically significant dose-response patterns were observed with either
log-transformed or across TCDD exposure categories, and that the elevated risks among those
with higher exposures had very wide confidence intervals (that included unity) quantitative
dose-response analyses were not recommended for this outcome.
2.4.1.2.1.4.3. Eskenazi et al., 2003—Adverse birth outcomes.
2.4.1.2.1.4.3.1. Study summary.
Eskenazi et al. (2003) examined the relationship between serum TCDD levels and birth
outcome measures. Analyses were based on 745 of the 981 women enrolled in the SWHS who
reported having been pregnant (n = 1,822). Most of these pregnancies (888 pregnancies among
510 women) occurred after the accident. Analysis of spontaneous abortions was restricted to
769 pregnancies among 476 women that did not end in abortion or in ectopic or molar
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-96 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
pregnancy. Congenital anomalies were evaluated for the 672 pregnancies that did not end in
spontaneous abortion. For the birth outcomes of fetal growth and gestational age, analysis was
performed using 608 singleton births from women without hypertensive pregnancy disorders.
TCDD exposures were based on serum measures, most of which were taken shortly after
the accident. Serum was collected in 1976-1977 for 413 women, between 1978 and 1981 for
12 women, and in 1996 for 19 women. TCDD exposures based on serum samples collected from
1977 onward were back-extrapolated to 1976.
Statistical analyses were performed on pregnancies that ended between 1976 and the time
of interview. A continuous measure of logioTCDD (base 10 scale) was used to investigate
associations with adverse birth outcomes. Logistic regression was used to characterize the
relationship between TCDD exposure spontaneous abortions, small for gestational age, and
preterm birth (<37 weeks gestation). Linear regression was used to describe the relationship
between TCDD and birth weight (in grams) and gestational age (in weeks).
The risk estimates were adjusted for a series of characteristics that included sex of infant,
history of low birthweight child, maternal height, maternal body mass index, maternal education,
maternal smoking during pregnancy, and parity. No association was evident between TCDD
serum levels and spontaneous abortion for pregnancies between 1976 and 1998 (OR = 0.8,
95% CI = 0.6-1.2), or those between 1976 and 1984 (OR = 1.0, 95% CI = 0.6-1.6). No
statistically significant associations were found for birth weight or small gestational age, though
the association with birth weight for pregnancies between 1976 and 1984 associated with a
10-fold increase in TCDD was fairly large and marginally statistically significant (P = -92,
95% CI = -204 to 0). No association was noted for preterm delivery in relation to logioTCCD
levels.
2.4.1.2.1.4.3.2. Study evaluation.
This study was well-designed with well characterized exposures. Statistically significant
associations were not evident, although the birth-weight findings should be pursued with further
follow-up of the cohort. As the authors point out, those who were most vulnerable at the time of
the accident (the youngest) had not yet completed their childbearing years. While the study
lacked exposure data for the fathers, the authors indicated that only a small proportion were
believed to have high exposures to TCDD. The key limitation of the study was a reliance on
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-97 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
self-reported measures of pregnancy history, which may lead to some misclassification of the
birth outcomes. The observation that a large proportion of Seveso women had a voluntary
abortion because of fears of possible birth defects due to exposures from the accident suggest an
awareness bias is possible as a result of differential reporting of birth outcomes according to
exposure status.
2.4.1.2.1.4.3.3. Suitability of data for TCDD dose-response modeling.
No statistically significant associations were found in the study; in addition, possible
awareness bias could have influenced the self-reported measures of birth outcomes. Therefore,
quantitative dose-response assessment was not considered for this study.
2.4.1.2.1.4.4. Warner et al., 2004—Age at menarche.
2.4.1.2.1.4.4.1. Study summary.
Warner et al. (2004) examined the relationship between TCDD and age at menarche in
the SWHS cohort. As described earlier in this report, the SWHS comprised 981 participants.
This study was restricted only to those who were premenarcheal at the time of the accident
(n = 282). The proportional hazards model was used to model TCDD exposures and age at
menarche. Age at menarche was determined by questionnaire administered by a trained
interviewer. Covariates examined as potential confounders included height, weight, body mass
index, athletic training at the time of interview, smoking, and alcohol consumption.
TCDD exposures were determined using serum samples collected from 257 of these
women between 1976 and 1977. For the remaining women, TCDD levels were quantified from
measures collected between 1978 and 1981 (n = 23) and in 1996 (n = 2). TCDD levels were
back-extrapolated to the time of the explosion in 1976. TCDD was modeled as both a
continuous variable (logioTCDD) and a categorical variable based on quartile values (<55.9,
56-140.2, 140.3-300, >300 ppt). The lowest group was further subdivided into those with levels
<20, and >20 ppt; this cut-point represented background levels found in a sample of women
living in an unexposed area.
No association was found between the continuous measure of TCDD and age at
menarche (hazard ration [HR] = 0.95, 95% CI = 0.83-1.09). Analyses restricted to those who
were younger than 8 in 1976 produced similar results (HR = 1.08, 95% CI = 0.89-1.30).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-98 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Additionally, no dose-response trend was observed with categorical measures of TCDD among
all women, as well as those under the age of 8. A marginally statistically significant association
with earlier menarche was found when analyses were limited to 84 women under the age of 5 at
the time of the accident (HR = 1.20, 95% CI = 0.98-1.60).
2.4.1.2.1.4.4.2. Study evaluation.
An important strength of the Warner et al. (2004) study is the ability to characterize
TCDD exposures using serum samples that were collected shortly after the accident occurred.
The outcome of interest, age at menarche, was determined by asking women "At what age did
you get your first menstrual period?" Recent work suggests that self-reported measures of age at
menarche decades later have modest agreement with responses provided during adolescence with
recall varying by education and by history of an adverse birth outcome (Cooper et al., 2005). In
the Seveso study, bias would be introduced if recall varied according to exposure levels.
2.4.1.2.1.4.4.3. Suitability of data for TCDD dose-response modeling.
Although the TCDD exposure characterization of study subjects was based on serum
data, and no major biases were introduced from the study design, the analyses produced largely
null associations. Therefore, quantitative dose-response assessment was not considered for this
study.
2.4.1.2.1.4.5. Eskenazi et al., 2005—Age at menopause.
2.4.1.2.1.4.5.1. Study summary.
Eskenazi et al. (2005) evaluated the relationship between age at onset of menopause and
serum levels of TCDD among women in the SWHS. Of the 981 women who agreed to
participate in SWHS, this analysis was restricted to those who had not reached natural
menopause before the time of the accident and who were at least 35 years of age at the time of
the interview. The recruitment and interview of women occurred approximately 20 to 22 years
after the accident (March 1996-July 1998).
The population was divided into quintiles of serum TCDD levels for the categorical
analysis. For most women (n = 564), TCDD levels were estimated from samples provided in
1976-1977. For the remaining women included in these analyses, TCDD levels were estimated
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-99 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
from samples collected between 1978 and 1982 (n = 28) and between 1996 and 1997 (n = 24).
As noted previously, exposure levels for women with post-1977 detectable levels of TCDD were
back-extrapolated to 1976 using either the first-order kinetic model (Pirkle et al., 1989)
(>16 years at time of accident) or the Filser model (<16 years at time of accident) (Kreuzer et al.,
1997). Women were classified as premenopausal if they were still menstruating or if they had
amenorrhea as a result of pregnancy or lactation (at the time of interview) with an indication of
subsequent menstruation based on maintained diaries or further examination. Subjects for which
amenorrhea had persisted for at least 1 year with no apparent medical explanation were classified
into a natural menopause category. The category, surgical menopause, pertained to women with
a medically confirmed hysterectomy or an oophorectomy. Finally, impending menopause was
defined for subjects in which menstruation had been absent for 2 months, but who provided
evidence of subsequent menstruation, or had a secretory endometrial lining, or indicated less
predictable cycles in the previous 2-5 years. If participants' menopausal status could not be
determined, they were grouped into the "other" category. This category included those for
whom status could not be determined due to current use of oral contraceptives, hormone
replacement therapy, or previous cancer chemotherapy.
Statistical analysis was based on both a continuous measure of log-transformed TCDD
exposures and categories based on quintiles (<20.4 ppt; 20.4-34.2 ppt; 34.3-54.1 ppt;
54.2-118.0 ppt; >118.0 ppt). The Cox model was used to generate hazard ratios as estimates of
relative risks and their 95% confidence intervals examining natural menopause as the outcome.
Several covariates previously identified as associated with menopausal status in the literature
were considered as potential confounders. These covariates included body mass index, physical
activity, premenopausal smoking, education, marital status, history of heart disease and other
medical conditions, and other reproductive characteristics.
The RRs were found to increase across the second through fourth quintiles (RRs =1.1,
1.4, and 1.6, respectively) of serum TCDD categories in relation to those in the lowest category,
but not in the upper quintile (RR = 1.0, 95% CI = 0.6-1.8). A statistically significant test of
trend was detected across the first four quartiles (p = 0.04) but not across all five quintiles
(p = 0.44). A statistically significant association with onset of menopause was not detected
(RR = 1.02, 95% CI = 0.8-1.3) based on the logTCDD continuous measure.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-100 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.4.5.2. Study evaluation.
The categorical exposure results from this study support a non-monotonic
dose-related-association for earlier menopause with increased serum TCDD levels up to
approximately 100-ppt TCDD serum, but not above. Eskenazi et al. (2005) speculated that the
inverse "U" shape of the dose-response relationship is explained by the mimicking of hormones
at lower doses of a chemical, while at higher levels the toxic effect of a chemical does not have
the capacity to either inhibit or stimulate hormonal effects.
A study limitation is the potential for residual confounding due to adjustment based on
current smoking status and not at the time of onset of menopause. It is unclear to what extent
smoking status may differ between these two time periods and whether smoking is related to
TCDD exposures in this cohort. Exposures to other dioxin-like compounds were not considered
in this study because of small serum volumes, but any potential TEQ exposures occurring in the
exposed population were thought to be mostly attributable to TCDD in the exposed women.
2.4.1.2.1.4.5.3. Suitability of data for TCDD dose-response modeling.
To date, this study is the only one that has examined the relationship between TCDD
levels and onset of menopause. Although the findings suggest the possibility of a nonlinear
dose-response function, the logioTCDD exposure metric was not statistically significant, nor
were any category-specific hazard ratios statistically significant relative to the lowest category.
Therefore, a quantitative dose-response analysis was not undertaken.
2.4.1.2.1.4.6. Warner et al., 2007—Ovarian function.
2.4.1.2.1.4.6.1. Study summary.
Warner et al. (2007) investigated the association between serum TCDD levels and
ovarian function in subjects in the SWHS who were younger than 40 in 1976 and for whom sera
collected after the accident had been stored. These women were recruited from March 1996 until
July 1998. Ovarian function analysis was limited to 363 women between 20 and 40 years of age
and who were not using oral contraceptives. Of these, 310 underwent transvaginal ultrasound
and were included in the functional ovarian cyst analysis. Ninety-six women were in the
preovulatory stage of their menstrual cycles and were included in the follicle analysis. For the
hormone analysis, 126 women who were in the last 2 weeks of their cycle were included.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-101 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
The authors used logistic regression to examine the relationship between TCDD and the
prevalence of ovarian follicles greater than 10 mm. Linear regression models examined the
continuous outcome variables: number of ovarian follicles >10 mm and diameter of dominant
ovarian follicle. Covariates considered for inclusion in the model were age at ultrasound, age at
accident, age at menarche, marital status, parity, gravidity, lactation history, current body mass
index, age at last birth, and smoking history. For the serum hormone analyses, estradiol and
progesterone were measured in blood at the time of interview. Ovulation status was defined as a
dichotomous variable (yes/no) based on a serum progesterone cut-point value of 3 ng/mL.
The adjusted ORs across categories of TCDD exhibited no dose-response trend for the
presence of follicles in relation to TCDD in the follicular phase; also, no statistically significant
differences were noted in any of the upper exposure categories relative to those in the lowest.
The adjusted OR for the continuous measure of logioTCDD was 0.99 (95% CI = 0.4-2.2). A
similar nonstatistically significant finding was found for logioTCDD in relation to ovulation in
both the luteal (OR = 0.99, 95% CI = 0.5-1.9) and mid-luteal phases (OR = 1.03,
95% CI = 0.4-2.7). Analyses of progesterone and estradiol also were not related to serum
TCDD levels for either the luteal or mid-luteal phases (p = 0.51 andp = 0.47).
2.4.1.2.1.4.6.2. Study evaluation.
The investigators found no relationship between serum TCDD levels and serum
progesterone and estradiol levels among women who were in the luteal phase at the time of
blood draw. No association with number of ovarian follicles detected from ultrasound.
Although no association was found, the authors suggested that the lack of significant results
could be because the women in SWHS were all exposed postnatally and the relevant and critical
time period for an effect might be in utero (animal studies support relevance of in utero
exposures).
2.4.1.2.1.4.6.3. Suitability of data for TCDD dose-response modeling.
One limitation of the study was the lack of examination of confounding by dioxin-like
compounds. The absence of associations between TCDD and adverse health effects in this study
precludes conducting quantitative dose-response analyses.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-102 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.4.7. Eskenazi et al., 2007—Uterine leiomyoma.
2.4.1.2.1.4.7.1. Study summary.
Associations between TCDD exposures and uterine leiomyoma (i.e., fibroids) were also
examined among 956 women in the SWHS (Eskenazi et al., 2007). The sample population was
based on the on the original 981 SWHS participants excluding 25 women diagnosed with
fibroids before the date of the accident (July 10, 1976). Women who previously had fibroids
were identified both through the administered questionnaire and the review of medical records.
Transvaginal ultrasounds were performed for 634 women to determine if they had fibroids at the
time of follow-up. Similar to other SWHS studies, exposure to TCDD was estimated using
serum collected from women shortly after the time of the accident, between 1978 and 1981 and
in 1996. TCDD levels were back-extrapolated to 1976 levels.
The study authors performed statistical analyses using two definitions of fibroids as
outcome measures. The first was fibroids detected before the study, and the second was fibroids
detected via ultrasound. A proportional odds method Dunson and Baird (2001) developed was
used to model the cumulative odds of onset of fibroids. This method combines historical and
current information of diagnoses of fibroids. Continuous and categorical measures of TCDD
were modeled. Regression models were adjusted for known or suspected risk factors of fibroids
including parity, family history of fibroids, age at menarche, body mass index, smoking, alcohol
use, and education.
2.4.1.2.1.4.7.2. Study evaluation.
Categorical measures of TCDD suggested an inverse dose-response relationship with the
onset of fibroids. Relative to those with TCDD levels less than 20 ppt, those having TCDD
exposures between 20.1 and 75.0 ppt and greater than 75.0 ppt had RRs of 0.58
(95% CI = 0.41-0.81), and 0.62 (95% CI = 0.44-0.89), respectively. The continuous measure of
logioTCDD produced a hazard ratio of 0.83 (95% CI = 0.65-1.07).
2.4.1.2.1.4.7.3. Suitability of data for TCDD dose-response modeling.
The inverse association between TCDD and uterine fibroids supports the possibility of an
anti-estrogenic effect of TCDD. The observed direction of the reported associations precludes
quantitative dose-response modeling.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-103 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2.4.1.2.1.5. Other Seveso noncancer studies.
2.4.1.2.1.5.1. Mocarelli et al., 2008—Semen quality.
2.4.1.2.1.5.1.1. Study summary.
Mocarelli et al. (2008) examined the relationship between TCDD and endocrine
disruption and semen quality in a cohort of Seveso men. A total of 397 subjects of the eligible
417 males (<26 years old in 1976) from Zone A and nearby contaminated areas were invited to
participate. Frozen serum samples were used to derive TCCD exposures. Also, 372 healthy
blood donors not living in the TCCD-contaminated area were invited to participate. The
researchers collected a health questionnaire and semen samples from participants. Analyses
were based on 257 individuals in the exposed group and 372 in the comparison group.
Semen samples were collected postmasturbatory at home. Ejaculate volume, sperm
motility, and sperm concentration were measured on these samples. Fasting blood samples also
were collected from the subjects for reproductive hormone analyses, including 17P-estradiol
(E2), follicle stimulating hormone (FSH), inhibin B, luteinizing hormone (LH), and testosterone.
The researchers estimated serum concentrations of TCDD from samples provided in
1976-1977, and also in 1997-1998 for individuals whose earlier samples had TCDD values that
exceeded 15 ppt. Serum concentrations for the comparison group were assumed to be less than
15 ppt in 1976 and 1977 and <6 ppt in 1998/2002 on the basis of serum results for residents in
uncontaminated areas. The exposed and comparison groups were divided into three groups
based on their age in 1976: 1-9, 10-17, and 18-26 years. Mocarelli et al. (2008) applied a
general linear model to the sperm and hormone data and included exposure status, age, smoking
status, body mass index, and occupational exposures as covariates.
Men exposed between the ages of 1 and 9 had reduced semen quality 22 years later.
Reduced sperm quality included decreases in sperm count (p = 0.025), progressive sperm
motility (p = 0.001), and total number of motile sperm (p = 0.01) relative to the comparison
group. The opposite pattern was observed for several indices of semen quality among those aged
10-17 at the time of the accident. For the hormone analyses, those in the exposed group had
lower serum E2 levels, and higher follicle stimulating hormone concentrations. Neither
testosterone levels nor inhibin B concentrations were associated with TCDD exposure.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-104 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.5.1.2. Study evaluation.
The findings of the Mocarelli et al. (2008) study support the hypothesis that exposure to
TCDD in infancy/prepuberty reduces sperm quality and could contribute to reported decrease in
sperm quality in young men in the industrialized world. Although most semen analysis studies
have low compliance rates in general population samples (20-40%) (Jorgenson et al., 2001;
Muller et al., 2004), the compliance rate in this study was much higher (60%). Given that the
compliance rates were similar between the exposed and comparison groups and the strong
differences detected across the two age groups, selection bias appears unlikely in this study.
2.4.1.2.1.5.1.3. Suitability of data for TCDD dose-response modeling.
Health outcomes are well defined in the Mocarelli et al. (2008) study, and exposures are
well characterized using serum data. Because the men exposed to elevated TCDD levels
between the ages of 1 and 9 had reduced semen quality 22 years later, it is difficult to identify the
relevant time interval over which TCDD dose should be considered. Specifically, it is difficult
to discern whether this effect is a consequence of the initial high exposure between 1 and 9 years
of age or a function of the cumulative exposure for this entire exposure window beginning at the
early age. However, the differences between these two dose estimates (the initial high exposure
versus the cumulative exposure for the 9 year window) are minimal (i.e. within an order of
magnitude). Despite the uncertainty in estimating the critical window of exposure,
dose-response analysis for this outcome was conducted.
2.4.1.2.1.5.2. Mocarelli et al., 1996, 2000—Sex ratio.
2.4.1.2.1.5.2.1. Study summary.
A letter to the editor was the first report of a possible change in the sex ratio from dioxin
among Seveso residents following the July 10, 1976 accident (Mocarelli et al., 1996). The
authors reported that 65% (n = 48) of the 74 total births that had occurred from April 1977 to
December 1984 were females. This male to female ratio of 26:48 (35%) is significantly different
from the worldwide birth ratio of 106 males to 100 females (51%) (James, 1995). Between 1985
and 1994, the Seveso male to female ratio leveled out at 60:64 (48%). The authors suggested
that the finding supported the hypothesis that dioxin might alter the sex ratio through several
possible mechanistic pathways.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-105 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Mocarelli et al. (2000) later reported on an investigation between serum-based TCDD
measures in parents and the sex ratio of offspring. In this study, serum samples were collected
from mothers and fathers who lived in the areas at the time of the explosion, were between the
ages of 3 and 45 at the time of the explosion, and produced offspring between April 1, 1977 and
December 31, 1996. The study population included 452 families and 674 offspring, and serum
measures were available for 296 mothers and 239 fathers. An estimate of TCDD at the time of
conception was also examined in relation to male to female birth ratios. TCDD exposure
estimates between the years of 1976 and 1996 were estimated using Filser's model (Kreuzer et
al., 1997).
Mocarelli et al. (2000) used chi-square test statistics to compare observed sex ratio to an
expected value of 0.51 in this Seveso population. Concentrations of TCDD were modeled as
categorical variables in several ways. First, a dichotomous variable was used whereby
unexposed parents were defined as those who lived outside Zones A, B, and R or had a serum
TCDD concentration of less than 15 ppt; parents with exposures of 15 ppt or higher were
considered exposed. Second, a trichotomous exposure variable was created that consisted of
parents who (1) lived outside Zones A, B, and R or had serum concentrations of less than 15 ppt,
(2) had serum concentrations of 15-80 ppt, and (3) had serum concentrations that exceeded
80 ppt. These cut-points were chosen as they represented tertiles based on the distribution of
TCDD among parents. Analyses were conducted separately for paternal and maternal TCDD
levels.
The overall proportion of 0.49 male births (based on male to female ratio of 328:346) was
not significantly different from the expected proportion of 0.51 (p> 0.05). Statistically
significant differences were found, however, if both parents had TCDD levels >15 ppt (sex
ratio = 0.44) or just the father had serum TCDD levels >15 ppt (sex ratio = 0.44). No
statistically significant differences were found when the fathers had TCDD levels less than
15 ppt, irrespective of the maternal levels. A dose-response pattern in the sex ratio was found
across the paternal exposure categories. That is, the sex ratio decreased with increased paternal
TCDD levels (linear test for trend, p = 0.008). In the unexposed group, the sex ratio (male to
female) was 0.56 (95% CI = 0.49-0.61), while in the highest exposure group
(281.0-26,400.0 ppt) the corresponding sex ratio was 0.38 (95% CI = 0.28-0.49).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-106 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Stratified analyses by age at paternal exposure revealed that the sex ratio was altered to a
greater degree among fathers who were younger than 19 at the time of the explosion. The male
to female ratio among the unexposed fathers was 0.56 (95% CI = 0.50-0.62), while it was 0.38
(95% CI = 0.30-0.47) for those younger than 19 when exposed and 0.47 (95% CI = 0.41-0.53)
for those exposed after 19. Regardless of the age at the time of exposure, however, fathers who
were exposed had a statistically significantly different birth ratio (they were more likely to father
girls) than those who were unexposed (p < 0.05).
Separate analysis of birth ratios based on paternal TCDD exposure estimated at the time
of conception did not show the same dose-response pattern but did show strong evidence of
consistently decreased male births relative to females. More specifically, the male to female
birth ratios among the four successive quartiles (first through fourth) were 0.41, 0.33, 0.33,
and 0.46.
2.4.1.2.1.5.2.2. Study evaluation.
Mocarelli et al. (2000) based the characterization of TCDD exposure on serum samples,
which is an objective method for characterizing dose. Unlike for the occupational cohorts, serum
measures for this study were taken close to the time of the accident, and therefore,
back-extrapolation of TCDD exposures is unnecessary. Exposure received before the age of 19
at the time of the explosion were more strongly associated with a reduced male to female ratio
than those received after the age of 19. The cut off age of 19 seems to be somewhat arbitrary,
resulting in a highly uncertain critical exposure window. TCDD levels at the time of conception
did not demonstrate a dose-response relationship, but paternal exposures resulted in consistently
reduced male to female birth ratios (range: 0.33-0.46).
The study findings are unlikely to be influenced by age at conception as these values
were found, on average, to be similar across calendar years. This suggests that age at conception
was not an important confounder and that the birth ratio findings may be related to paternal
exposures.
The methods used to identify births appear to be appropriate. Even if some
under-ascertainment of births occurred, there is no reason to believe that ascertainment would be
related to TCDD exposure and the sex of the baby. Therefore, no bias is suspected due to
incomplete birth ascertainment.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-107 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.5.2.3. Suitability of data for TCDD dose-response modeling.
TCDD exposures were well-characterized, and internal cohort analyses demonstrate
association between paternal TCDD levels at the time of the accident and birth ratio. However,
the change in sex ratio was only statistically significant when exposure occurred before 19 years
of age. It is impossible to identify the relevant time interval over which TCDD dose should be
considered for dose-response analysis; specifically, it is difficult to discern whether the different
sex ratio is a consequence of the initial peak exposure before 19 years of age or a function of the
average cumulative exposure over this entire exposure window. Assuming the initial high
exposure is the correct exposure window, using the initial exposures in a dose-response model
would yield LOAELs that are too high to be relevant to factor into the RfD calculation. The
differences between the two dose estimates are quite large. Dose-response analysis for this
outcome, therefore, was not conducted.
2.4.1.2.1.5.3. Baccarelli etal., 2008—Neonatal thyroidfunction.
2.4.1.2.1.5.3.1. Study summary.
Baccarelli et al. (2008) investigated the relationship between thyroid function and TCDD
among offspring of women of reproductive age who were exposed in the 1976 accident. This
health endpoint is relevant because thyroid function is important for energy metabolism and
nutrients and for stimulating growth and development of tissues. Neonatal thyroid function at
birth is evaluated through blood thyroid-stimulating hormone (b-TSH). Apart from iodine
deficiency, no other environmental exposure has been associated consistently with reduced
neonatal thyroid functioning.
The study population was drawn from 1,772 women who were identified as having lived
in the highly contaminated areas (Zones A or B) at the time of the accident or between July 10,
1976 and December 31, 1947; were of fertile age (born after 1947); and were alive as of
January 1, 1994. A random sample of 1,772 unexposed women who lived in the reference area
was selected using frequency matching by year of birth to the exposed women, and residency in
the reference area at the time of the accident. The reference area represents the noncontaminated
areas that surround the three zones of decreasing exposure (Zones A, B and R). In total,
55,576 women had lived in the reference area. Population registry offices (n = 472) were
contacted to detect children born to these women. Records could be traced for virtually all
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-108 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
subjects (1761/1772 exposed; 1762/1772 unexposed). Children born outside the Lombardy area
were excluded as b-TSH could not be obtained for them. This accounted for 156 of the
1,170 children identified. The analyses were based on the remaining 56, 425, and 533 singletons
born between January 1, 1994, and June 30, 2005 in Zone A, B, and from the reference area,
respectively.
Thyroid function is tested in all newborns by b-TSH measures in the region of Lombardy
where Seveso is located. These measures are obtained from blood samples taken 72 hours after
birth using a standardized protocol. The b-TSH levels were log transformed to approximate a
normal distribution. Linear regression analysis was used to conduct test for trends in mean
b-TSH levels across different covariates. Logistic regression was used to assess associations
between elevated b-TSH levels defined by the cutpoint of 5 [j,U/mL and residence in particular
zones of contamination. Generalized estimating equations were used to adjust the standard
errors of the ORs for correlation between siblings.
The mean levels of b-TSH were positively associated with average soil TCDD
concentrations in the three areas (Zone A: 1.66 [j,U/mL; Zone B: 1.35 |iU/mL; and Zone R:
0.98 (j.U/mL) (p < 0.001). Plasma TCDD levels also were shown to be much higher in a group of
51 newborns that had b-TSH levels >5 [j,U/mL. Compared to the reference population, adjusted
ORs were elevated for Zone B (OR = 1.90, 95% CI = 0.94-3.86) and Zone A (OR = 6.63,
95% CI = 2.36-18.6). These ORs were adjusted for gender, birth weight, birth order, maternal
age at delivery, hospital, and type of delivery. The adjusted ORs however differed only slightly
from those that were unadjusted (Zone B, OR = 1.79, 95% CI = 0.92-3.50; Zone A OR = 6.60,
95% CI = 2.45-17.8). Of the risk factors considered, both gender and birth weights were
associated with neonatal b-TSH.
The paper also included an analysis of children born to 109 women who were part of the
Seveso Chloracne Study (Baccarelli et al., 2005). A total of 51 children were born to 38 of these
women, of these 12 lived in Zone A, 10 in Zone B, 20 in Zone R, and 9 from the reference
population. Several congeners including TCDD were measured in maternal plasma. TCDD
levels were extrapolated to the date of delivery using a first-order pharmacokinetic model
(Michalek et al., 1996). The elimination rate used was 9.8 years based on the mean half-life
estimate from a previous study of women in the Seveso region (Michalek et al., 2002). TEQs
were calculated for a mixture of dioxin-like compounds by multiplying the concentration of each
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-109 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
congener by its toxicity equivalence factor. The maternal average TEQ was 44.8 ppt
(range: 11.6-330.4) among 51 mothers. The measurement of noncoplanar PCBs occurred only
later in the study (1996) and, therefore, total mean TEQs (i.e., including the sum of PCDDs,
PCDFs, coplanar PCBs, and noncoplanar PCBs) are available only on a subset (n = 37) of the
population. Dioxin-like congeners were examined in this study as several studies suggest
associations between the sum of PCBs, or individual congeners having decreased thyroxine (T4;
Longnecker et al., 2000; Sandau et al., 2002), and increased TSH (Alvarez-Pedrerol et al., 2008;
Chevrier et al., 2007).
The authors used a linear model to examine the association between maternal TCDD
levels and b-TSH. The standardized regression coefficient obtained from this model was 0.47
(p < 0.001). For the evaluation of TEQs, a similar association was noted for PCDDs, PCDFs,
and coplanar PCBs (n = 51, P = 0.45, p = 0.005) but not with non-coplanar PCBs (n = 37,
P = 0.16,p = 0.45). Multivariate regression models that were adjusted for several covariates
(i.e., gender, birth weight, birth order, maternal age at delivery, hospital, and type of delivery)
found statistically significant associations with plasma TCDD, PCDDs, PCDFs, and coplanar
PCBs, but not with non-coplanar PCBs. The sum of all total TEQs from the measured
compounds was not statistically significant (n = 37, P = 0.31 ,p = 0.14).
2.4.1.2.1.5.3.2. Study evaluation.
The Baccarelli et al. (2008) study satisfies the epidemiological considerations and criteria
for determining whether dose-response modeling should be pursued. The outcome is well
defined, and a dose-response pattern was observed. The study also contained a substudy that
characterized TCDD and exposures to other dioxin-like congeners and used serum measures for
a sample of mothers. Results were consistent among the zone of residence analysis and the
substudy based on serum measures.
2.4.1.2.1.5.3.3. Suitability of data for TCDD dose-response modeling.
Given the potential for exposure misclassification due to variability in TCDD soil levels
within each zone, modeling should rely on individual-level TCDD exposures derived from the
serum sampling substudy. The study data provide an opportunity for quantitative dose-response
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-110 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
analyses as the critical exposure window of 9 months can be used for exposure assessment
purposes.
2.4.1.2.1.5.4. Alaluusua et al., 2004—Oral hygiene.
2.4.1.2.1.5.4.1. Study summary.
Alaluusua et al. (2004) examined the relationship between TCDD and dental defects,
dental caries, and periodontal disease among Seveso residents who were children at the time of
the accident. Subjects were randomly selected from those individuals who had previously
provided serum samples in 1976, which was shortly after the accident. A total of 65 subjects
who were less than 9.5 years of age at the time of the accident, and who lived in Zones A, B, or
R were invited to participate. Recruitment was initiated 25 years after the time of the Seveso
accident. An additional 130 subjects from the surrounding area (outside Zones A, B, or R or
"non-ABR zone") having the same age restriction were recruited. Subjects were frequency
matched for age, sex, and education. Questionnaires were administered to these individuals to
collect detailed information on dental and medical histories, education, and smoking behaviors.
Ten subjects who had completed at least high school were randomly excluded from the non-ABR
zone to create groups with similar educational profiles. Participation rates for the ABR and
non-ABR zones were 74% and 58%, respectively.
One dentist who was blind to the patients' TCDD exposure levels assessed dental
aberrations. Dental caries was assessed using recommendations of the World Health
Organization. Periodontal status was described following a detailed evaluation of the surfaces of
the teeth. A radiographic examination was done to identify missing teeth, alveolar bone loss,
deformities in the roots, and jaw cysts.
Comparisons of the presence of dental enamel defects according to exposure status were
performed using logistic regression. Chi-square test statistics were applied to compare the
distributions in the prevalence of dental defects across several categorical covariates (i.e.,
education, age, and serum TCDD level). For those who were younger than 5 at the time of the
accident, dental defects were more prevalent among patients in zone ABR (42%) than those in
the non-ABR zone (26%) (p = 0.14). Zone ABR is characterized by higher levels of soil TCDD
levels relative to non-ABR. Serum levels permitted an improved characterization of risk as they
were available at an individual level, rather than using a zone of residence. Defect prevalence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-111 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
was highest among those in the upper serum TCDD category (700-26,000 ng/kg) with 60% of
subjects having dental defects. The continuous measure of serum TCDD was associated with
developmental defects (p = 0.007) and hypodontia (p = 0.05).
2.4.1.2.1.5.4.2. Study evaluation.
Although the subjects with serum measures were selected randomly, no direct measures
of TCDD were made in subjects from the unexposed area (i.e., non-ABR zones). That those who
resided in the non-ABR areas had lower TCDD exposures would be a reasonable assumption.
Alaluusua et al. (2004), however, provide few details about the sampling frame used to identify
these participants. Despite this fact, it is important to note that a dose-response pattern was
observed between TCDD exposure and presence of developmental defects in the ABR
population alone (p = 0.016). This finding is based on 27 subjects with developmental defects.
This positive association provides support for a quantitative dose-response modeling of dental
aberrations. The numbers of such subjects are small, however, with one, five, and nine subjects
having defects in the exposure groups of 31-226, 238-592, and 700-26,000 ng/kg TCDD,
respectively.
TCDD exposures were characterized using serum measures for those who resided in zone
ABR in 1976 (near the time of the accident). The authors could not account for additional
exposure to TCDD across subjects that might have occurred since the time of the accident, so
there is considerable uncertainty in delineating the critical exposure window for the reported
effects. In addition, the lack of exposure data for those in the non-ABR zone, however, makes
interpretation of the findings difficult. This difficulty is particularly evident, given that the
prevalence of dental defects was less among those in the low exposure category of zone ABR
(31-226 ng/kg TCDD) (10%) when compared to those in the non-ABR zone (26%).
2.4.1.2.1.5.4.3. Suitability of data for TCDD dose-response modeling.
Most of the considerations for conducting a dose-response analysis have been satisfied
with the study population, although, exposure assessment uncertainties are a limitation of this
study. For example, it is difficult to discern whether these health effects are a consequence of
the initial high exposure during childhood or a function of the cumulative exposure for this entire
exposure window beginning at the early age. If the latter is true, averaging exposure over the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-112 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
critical window would add considerable uncertainty to effective dose estimates given the large
difference between initial TCDD body burden and body burden at the end of the critical
exposure window. Despite the uncertainty in defining the critical window of exposure,
dose-response analysis was conducted for this outcome.
2.4.1.2.1.5.5. Bertazzi et al., 1989; Consonni et al., 2008—Mortality outcomes.
2.4.1.2.1.5.5.1. Study summary.
Several studies have evaluated the mortality of Seveso residents exposed to TCDD
following the 1976 accident. The earlier section of this report described the designs of these
studies and discussed their findings as they relate to cancer mortality. In this section, some of
the findings for other causes of death are described. A key feature of these studies is that
patterns of mortality among Seveso residents were investigated according to their zone of
residence at the time of explosion relative to general population rates.
A 10-year mortality follow-up of residents of Seveso was published in 1989 (Bertazzi et
al., 1989). Poisson regression was used to derive RRs for those who had lived in Zone A at the
time of explosion using a referent group consisting of inhabitants who had lived in the
uncontaminated study area. Between 1976 and 1986, no statistically significant difference was
observed in all-cause mortality relative to the general population among those who lived in the
most highly exposed area (Zone A) at the time of the accident. This finding was evident in both
males (RR = 0.86, 95% CI = 0.5-1.4) and females (RR = 1.14, 95% CI = 0.6-2.1). A
statistically significant excess in circulatory disease mortality was found among males relative to
those in the referent population (RR = 1.75, 95% CI = 1.0-3.2); this increased risk was more
pronounced when the follow-up period was restricted to the first 5 years after the accident
(1976-1981) (RR = 2.04, 95% CI = 1.04-4.2). Between 1982 and 1986, the RR decreased
substantially and was not statistically significant (RR = 1.19, 95% CI = 0.4-3.5). Among
females, a risk similar in magnitude was detected for circulatory disease mortality although it
was not statistically significant (RR = 1.89, 95% CI = 0.8-4.2). Contrary to the calendar
period-specific findings for males, the excess of circulatory mortality among females occurred
between 1982 and 1986 (RR = 2.91, 95% CI = 1.1-7.8) and not between 1976 and 1981
(RR = 1.12, 95% CI = 0.3-4.5). The number of deaths in this cohort with the 10 years of
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-113 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
follow-up was relatively small; in Zone A, 16 deaths were observed among males and 11 among
females.
The most recently published account of the mortality experience of Seveso residents
provides further information on follow-up of these residents until the end of 2001 (25 years after
the accident) (Consonni et al., 2008). Three exposure groups were considered: Zone A (very
high contamination), Zone B (high contamination), and Zone R (low contamination). The
reference population consisted of those residents who lived in unaffected surrounding areas, as
well as residents of five nearby towns. The authors used Poisson regression to compare
mortality rates for each zone relative to the reference population.
For all causes of death, no excess was found in Zone A, B, or R relative to the reference
population. Statistically significant excesses were noted for those who lived in Zone A relative
to the reference population for chronic rheumatic heart disease (RR = 5.74,
95% CI = 1.83-17.99) and chronic obstructive pulmonary disease (RR = 2.53,
95% CI = 1.20-5.32). These risks, however, were based on only 3 and 7 deaths, respectively.
For those in Zone A, no statistically significant excesses in mortality were noted for diabetes,
accidents, digestive diseases, ischemic heart disease, or stroke. Among Zone A residents,
stratified analysis by time since accident showed increased rates of circulatory disease 5-9 years
since the accident (RR = 1.84, 95% CI = 1.09-3.12). Increased mortality from diabetes relative
to the reference population was noted among females who lived in Zone B (RR = 1.78,
95% CI = 1.14-2.77).
2.4.1.2.1.5.5.2. Study evaluation.
The ascertainment of mortality in this cohort is nearly complete. Misclassification of
some health outcomes, such as diabetes, may occur due to use of death certificate data.
The characterization of exposure is based on zone of residence. Soil sampling indicated
considerable variability in TCDD soil levels, and therefore, the generation of risks based on zone
of residence likely does not accurately reflect individual exposure. Exposure misclassification
might also occur because residency in the areas does not necessarily reflect whether the
individual would have been present in the area at the time the accident occurred. Any exposure
misclassification would likely be nondifferential which would tend to bias the risk estimates
towards the null.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-114 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Although some excess of circulatory disease mortality was found, the finding was not
consistent between men and women. Moreover, excess circulatory disease mortality was more
pronounced among men within the first 5 years of exposure, while, for women, the excess was
more pronounced in years 5-10. Numerous other risk factors for circulatory disease were not
controlled for in these analyses and may be confounders if related to TCDD exposure. Taken
together, the possibility that TCDD increased circulatory disease mortality based on these data is
tenuous at best.
2.4.1.2.1.5.5.3. Suitability of data for TCDD dose-response modeling.
There is considerable uncertainty in these data due to the potential for outcome and
exposure misclassification. The lack of the individual-level TCDD levels and the examination of
fatal outcomes reported in this study are not a suitable basis for development of an RfD. For
these reasons, dose-response analysis for this outcome is not conducted.
2.4.1.2.1.5.6. Baccarelli et al., 2005—Chloracne.
2.4.1.2.1.5.6.1. Study summary.
Baccarelli et al. (2005) published findings from a case-control study of 110 chloracne
cases and 211 controls. The authors collected information on pigment characteristics and an
extensive list of diseases. This study was performed to yield information about the health status
of chloracne cases, TCDD-chloracne exposure response, and factors that could modify TCDD
toxicity. TCDD was measured from plasma. Following adjustment for confounding, TCDD was
associated with chloracne (OR = 3.7, 95% CI = 1.5-8.8), and the risk of chloracne was
considerably higher in subjects younger than 8 at the time of the accidents (OR = 7.4,
95% CI = 1.8-30.3). Among individuals with lighter hair, the association between TCDD and
chloracne was stronger than among those with darker hair.
2.4.1.2.1.5.6.2. Study evaluation.
Although a dose-response association was observed, chloracne is a rare health outcome
likely only to occur among those highly exposed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-115 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.5.6.3. Suitability of data for TCDD dose-response modeling.
Given the very high TCDD levels needed to cause chloracne, quantitative dose-response
modeling to characterize risks for the general population with much lower TCDD exposures
would be of little value. Therefore, quantitative dose-response assessment for the Baccarelli et
al. (2005) study was not conducted.
2.4.1.2.1.5.7. Baccarelli et al., 2002, 2004—Immunologic effects.
2.4.1.2.1.5.7.1. Study summary.
The relationship between TCDD and immunological effects was evaluated in a sample of
Seveso residents (Baccarelli et al., 2002, 2004). Both studies were based on findings from
62 individuals who were randomly selected from Zones A and B. An additional 59 subjects
were chosen from the surrounding noncontaminated areas. Residency was based on where
subjects lived at the time of the accident (July 10, 1976) (Landi et al., 1998). Frequency
matching ensured that the two groups of subjects were similar with respect to age, sex, and
cigarette smoking status.
TCDD levels were determined by mass spectrometric analysis of plasma samples.
TCDD levels at the time of sampling were obtained, and estimates of levels at the time of the
accident also were estimated by assuming an 8.2-year half-life (Landi et al., 1998). The plasma
was also used to characterize levels of the immunoglobulins (Ig) IgG and IgM and the
complement components C3 and C4. One subject was excluded due to lack of an immunological
evaluation. Analyses are, therefore, based on 58 subjects in the noncontaminated areas and
62 individuals from the contaminated areas.
Nonparametric tests were applied to test for differences between the two groups.
Multiple regression also was used to describe the relationship between the variables. Adjustment
was made for several potentially confounding variables that were collected via a questionnaire.
An inverse association was noted with increasing TCDD levels and plasma IgG levels;
this result remained statistically significant after adjusting for other potential confounding
variables in the regression models. Specifically, the slope coefficient and/>value for the
unadjusted model were -0.35 (p = 0.0002) and for the adjusted model thep-walue was 0.0004.
The authors did not present the slope coefficient for the adjusted model in either paper but noted
minimal differences between the adjusted and unadjusted results. In the 2004 analysis, the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-116 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
authors present IgG, IgM, IgA, C3, and C4 median and interquartile values across TCDD
exposure quintiles. Decreased levels of IgG were observed in the highest exposure groups.
Specifically, the median values across the five quintiles (for lowest to highest) were 1,526;
1,422; 1,363; 1,302; and 1,163. The Kruskal-Wallis test for differences across the TCDD
categories was statistically significant (p = 0.002), which is consistent with the findings for the
continuous measures of TCDD. This finding persisted after excluding those subjects with
inflammatory diseases and those who used antibiotics or nonsteroidal anti-inflammatory drugs.
For the other plasma measures, no dose-response relationship was apparent based on median
values for IgM, IgA, C3, or C4 across TCDD quintiles. The authors highlight the need for
additional research, particularly given the excess of lymphatic tumors noted in the area.
Exposure to other dioxin-like compounds for both the TCDD and nonexposed areas were
reported to be at background levels.
2.4.1.2.1.5.7.2. Study evaluation.
Both TCDD exposure and health outcome measures are well characterized. TCDD
exposures, in particular, are based on current serum measures and, therefore, are not dependent
on assumptions needed to back-extrapolate to earlier time periods of exposure.
A dose-response relationship between TCDD and IgG is well documented for the
unadjusted model, but no details are provided on the change in the slope coefficient when other
covariates were added to the model.
Interpreting the inverse association between TCDD exposure and IgG in terms of clinical
significance is not possible. The IgG values reported are much higher than those subjects with
antibody immunodeficiency disorders.
2.4.1.2.1.5.7.3. Suitability of data for TCDD dose-response modeling.
Although the data support an inverse dose-response association between IgG and TCDD,
because the relationship cannot be described in terms of clinical relevance with respect to a
specific adverse health outcome, these data were not suitable for qunatitative dose-response
modeling.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-117 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2.4.1.2.1.6. The Chayaevsk study.
2.4.1.2.1.6.1. Revich et al. (2001)—Mortality and reproductive health.
2.4.1.2.1.6.1.1. Study summary.
Revich et al. (2001) describe a series of investigations that have evaluated adverse health
outcomes among residents of Chapaevsk where ecological measures of TCDD have been noted
to be higher than expected. In the earlier cancer section of this report, the cross-sectional
comparisons of mortality that the authors carried out between Chapaevsk residents and a general
population reference were described. Although the general focus of this paper is on cancer, the
authors examined other adverse health outcomes.
For all-cause mortality, rates were found to be higher in Chapaevsk relative to the Samara
region and other nearby towns. The magnitude of this increase, however, was not quantified in
the review by Revich. Cardiovascular mortality accounted for nearly two-thirds of women's
deaths and almost half of those among men. The rates of cardiovascular mortality among
Chapaevsk men have been reported to be 1.14 times higher than those in Russia.
Revich et al. (2001) also reported on the occurrence of adverse reproductive events.
Although the authors indicated that official medical information was used to make comparisons
between regions, no details were provided about data quality, completeness, or surveillance
differences across areas. The presented rates for reproductive health outcomes should be
interpreted cautiously. A higher rate of spontaneous abortions (24.4 per 100 pregnancies
finished by delivery) was found in Chapaevsk women relative to rates that ranged between 10.6
and 15.2 found in five other areas. The frequency of preeclampsia also was found to be higher in
Chapaevsk women (44.1/100) relative to other towns, as was the proportion of low birth-weight
babies and preterm births. The percentage of newborns with low birth weight was slightly larger
in Chapaevsk (7.1%) when compared to other towns in Samara (5.1-6.2%); observed
differences, however, were not statistically significant. The authors also reported on the sex ratio
of newborns born between 1983 and 1997. These ratios (boys:girls) were highly variable and
ranged between 0.79 and 1.29. Given the annual variability of this ratio on a year-to-year basis,
it is unclear if this is largely due to natural fluctuations and to what extent this may result from
prior TCDD (or other contaminants) exposure TCDD and other contaminants.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-118 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.6.1.2. Study evaluation.
The review by Revich et al. (2001) highlights analyses that have been undertaken using
largely cross-sectional data. Although soil sampling measures appear to demonstrate decreasing
levels of TCDD in the soil with increasing distance from the plant, at this time, no
individual-level TCDD exposure data are available. Increased rates of mortality relative to the
Samara region in Russia were observed among Chapaevsk men for all cancer sites combined;
this excess risk however, was not observed among women. Although the authors provide
compelling evidence of increased adverse events among residents of Chapaevsk, the study lacks
a discussion about the validity of comparing health data across regions, and suffers from inherent
limitations from ecological studies such as exposure misclassification.
2.4.1.2.1.6.1.3. Suitability of data for TCDD dose-response modeling.
As with the cancer outcomes presented in this study, the data for noncancer outcomes are
limited by the absence of TCDD levels on an individual-level basis and information on other
potential confounding variables that could have biased the comparisons. Additional studies are
being undertaken to evaluate the relationship between TCDD and the sexual and physical
development of boys. The cross-sectional nature of the data that were presented does not
provide the necessary level of detail needed to estimate effective dose given the lack of
individual-level exposure data. Therefore, a quantitative dose-response analysis was not
conducted.
2.4.1.2.1.7. The Air Force Health ("Ranch Hands "cohort) study.
2.4.1.2.1.7.1. Michalek andPavuk (2008)—Diabetes.
2.4.1.2.1.7.1.1. Study summary.
Michalek and Pavuk (2008) examined both the incidence of cancer and the prevalence of
diabetes in the cohort of Ranch Hand workers exposed to TCDD. As noted previously, these
veterans were responsible for aerial spraying of Agent Orange in Vietnam between 1962 and
1971. Exposure to TCDD was estimated using serum collected from participants in 1987 and
assayed for TCDD. Exposure to TCDD was estimated using a first-order pharmacokinetic model
with a half-life of 7.6 years and provided an estimate of TCDD at the end of the tour of duty in
Vietnam. Veterans were grouped into four categories: comparison, background, low, and high.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-119 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Diabetes was identified from diagnoses during the post-Vietnam era from medical records.
Overall, no differences were shown in the RR of diabetes between the Ranch Hand unit and the
reference group (RR = 1.21 ,p = 0.16). Stratified analyses by days of spraying (<90 days,
>90 days), however, revealed a significant increase in risk of diabetes (RR = 1.32, p = 0.04)
among those who sprayed for at least 90 days. A dose-response relationship was also evident
when loglOTCDD was modeled in the combined cohort. Also, stratification by calendar period
showed a dose-response relationship for those whose last year of service was during or before
1969.
2.4.1.2.1.7.1.2. Study evaluation.
The Michalek and Pavuk (2008) study provides an opportunity to characterize risks of
diabetes as the study is not subject to some of the potential bias of case ascertainment based on
death certificates (D'Amico et al., 1999). The quality of the TCDD exposure estimates is high,
given that serum data were available at an individual-level basis for all Ranch Hand and
comparison veterans used in the cohort. Although disentangling the effects of 2,4-D and TCDD
is not possible because their concentrations in Agent Orange are equivalent, 2,4-D has not been
associated with diabetes.
2.4.1.2.1.7.1.3. Suitability of data for TCDD dose-response modeling.
The reported dose-response relationship between TCDD and diabetes is supported by
study strengths including the use of the individual-level level TCDD serum measures and the
identification of diabetes through medical records are important strengths of the Michalek and
Pavuk (2008) study. Nonetheless, the possible confounding from the inability to control for
2,4-D and other agents used in Agent Orange precludes a quantitative dose-response analysis.
2.4.1.2.1.8. Other noncattcer studies of TCDD.
2.4.1.2.1.8.1. McBride et al., 2009a—Noncancer mortality.
2.4.1.2.1.8.1.1. Study summary.
The McBride et al. (2009a) mortality study of New Zealand workers employed as
producer or sprayers with potential exposure to TCDD was described earlier in this report.
These individuals were employed at a plant that manufactured 2,4,-dichlorophenoxyacetic acid,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-120 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
and later 2,4,5-T and 4-chloro 2-methyphenoxyacetic acid. In 1987, the plant closed and 2,4,5-T
production ceased in 1988.
The cohort consisted of 1,754 individuals who were employed for at least one day at the
New Plymouth site between January 1, 1969, and October 1, 2003. Vital status was determined
until the end of 2004. Comparisons of mortality were made to the New Zealand general
population using the SMR statistic. Exposure was characterized by duration of employment.
Person-years of follow-up were tabulated across strata defined by age, calendar period, duration
of employment, sex, latency, and period of hire. Analyses were stratified to compare risks by
duration of employment (<3 or >3 months), latency (<15 or >15 years), and period of hire
(<1976, >1976).
Overall, no statistically significant differences in all-cause mortality relative to the
general population were found among those who worked for at least 3 months (SMR = 0.92,
95% CI = 0.80-1.06) or for less than 3 months (SMR = 1.23, 95% CI = 0.91-1.62). No
statistically significant excesses were found for mortality from diabetes, cerebrovascular disease,
heart diseases, or accidents. The incorporation of a latency period of 15 years revealed no
statistically significant excesses for these same causes of death. Similarly, no excesses for any
cause of death were noted among those who were hired either before or after 1976.
In subsequent analyses of the same cohort that used estimated TCDD levels from serum
samples, McBride et al. (2009b) found no excesses for all-cause mortality or mortality from
diabetes or heart disease.
2.4.1.2.1.8.1.2. Study evaluation.
For the McBride et al. (2009a) study, the size of the cohort is large enough to characterize
mortality risks relative to the general population for most common causes of deaths. An
important limitation of this study is the loss to follow-up of a substantial percentage of workers
(22%). This would have impacted statistical power by reducing the number of deaths among the
workers. If this incomplete ascertainment of mortality outcomes did not occur in a similar
fashion with the general population then the SMR may also be biased.
For noncancer causes of death, the use of the SMR statistic is more likely to be
influenced by the healthy-worker effect. Therefore, the findings obtained for these outcomes
should be interpreted with caution. Subsequent analyses published by the same authors
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-121 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
(McBride et al., 2009b) provide improved characterization of TCDD exposure using serum
samples.
2.4.1.2.1.8.1.3. Suitability of data for dose-response analysis.
Overall, no associations were evident between surrogate measures of TCDD (duration of
employment, year of hire) and noncancer mortality outcomes. Further, the use of mortality
endpoints is inconsistent with EPA RfD methodology. As such, these data do not support further
use in a quantitative dose-response analysis.
2.4.1.2.1.8.2. McBride et al., 2009b—Noncancer mortality.
2.4.1.2.1.8.2.1. Study summary.
McBride et al. (2009b) further analyzed the cohort of New Zealand workers to include
estimates of TCDD exposure based on serum samples. Current and former employees who were
still alive and living within 75 km of the site were asked to provide serum samples. Samples
were collected from 346 workers representing 22% (346/1599) of the entire study population.
These serum measures were used to estimate cumulative TCDD levels for all workers. The
exposure assessment approach by Flesch-Janys et al. (1996) was used to estimate time-dependent
exposures based on area under the curve models. This was based on a one-compartment
first-order kinetic model with a half-life of 7.2 years.
Comparisons of mortality were made to the general population using the SMR statistic.
The Cox proportional hazards model was used to conduct an internal cohort analysis across four
categories of cumulative TCDD levels for diabetes and ischemic heart disease mortality. The
RRs generated from these models were adjusted for sex, hire year, and birth year. No diabetes
deaths were observed among women, and therefore, analysis of this outcome was limited to men.
Relative to the general population, no difference in the all-cause mortality experience was
observed in exposed cohort members (SMR = 1.0, 95% CI = 0.9-1.2). Similarly, no excess in
these workers was observed for heart disease (SMR =1.1, 95% CI = 0.9-1.5); cerebrovascular
disease (SMR = 1.1, 95% CI = 0.6-1.9); diabetes (SMR =0.7, 95% CI = 0.2-2.2); or
nonmalignant respiratory disease (SMR = 0.8, 95% CI = 0.4-1.4). For the internal cohort
analysis, the RR associated with cumulative categorical TCDD measure was 1.0 for both
diabetes and ischemic heart disease.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-122 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.1.2.1.8.2.2. Study evaluation.
The McBride et al. (2009b) study extends the earlier work the same authors completed in
two ways. First, serum measures were used to estimate cumulative TCDD with methodology
that has been applied to several other cohorts of workers exposed to TCDD. Second, the authors
used regression analyses that examined individual-level TCDD exposures in relation to various
outcomes as part of the internal cohort comparisons. For noncancer outcomes, no dose-response
associations with TCDD were observed with the internal comparisons. Also, as found with
earlier analyses of this same cohort, no excess noncancer mortality relative to the New Zealand
general population was observed.
Associations between TCDD and diabetes have been found previously in TCDD-exposed
populations, most notably in the Ranch Hands cohort (Michalek and Pavuk, 2008). In this
cohort, only five deaths from diabetes were identified, and of these, only three occurred among
those who were exposed to TCDD. The study, therefore, has limited statistical power to
characterize associations between TCDD and mortality from diabetes. Further, the identification
of diabetes deaths is subject to misclassification errors due to under-reporting (McEwen et al.,
2006).
2.4.1.2.1.8.2.3. Suitability of data for TCDD dose-response modeling.
McBride et al. (2009b) found no statistically significant associations in any of the
noncancer causes of death. Furthermore, the use of mortality endpoints is inconsistent with EPA
RfD methodology. Therefore, the data were not suitable for quantitative dose-response analysis
for these outcomes.
2.4.1.2.1.8.3. Ryan et al., 2002—Sex ratio.
2.4.1.2.1.8.3.1. Study summary.
Ryan et al. (2002) conducted an investigation on the sex ratio in offspring of children of
pesticide workers who were involved with the production of trichlorophenol and the herbicide
2,4,5-T in Ufa, Bashkortostan, Russia. Ufa was the site of a state agrochemical plant that has
been in operation since the 1940s. Between 1961 and 1988, the plant employed more than
600 workers, most in their early 20s. Females, however, accounted for about 15% of the
workforce that produced 2,4,5-T and 30% for 2,4,5-trichlorophenol.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-123 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Serum samples previously taken in 1992 among 60 men, women, and children from the
factory and city of Ufa showed TCDD exposures that were approximately 30 times higher than
background levels (Ryan and Schecter, 2000). Blood data were subsequently measured on a
sample of 20 workers between 1997 and 2000, and on 23 2,4,5-trichlorophenol workers between
1997 and 2001. In all, 84 individuals who provided blood samples formed the basis of the
analysis in this study. Of these, 55 were exposed to 2,4,5-T and 29 were exposed to
2,4,5-trichlorophenol.
Ryan et al. (2002) reviewed company records for these workers to determine the number,
sex, and date of birth of any children; birth data were available for 198 workers. Awareness of
the study led other workers who had not provided serum to provide information on births that
occurred 9 months after the time of first employment in the factory.
The authors calculated descriptive statistics for the 198 workers and compared them to
values for the city of Ufa between 1959 and 1996. Tests of statistical significance were made
using the z-test, and the chi-square test. The observed proportion of male births (0.40) among
the factory workers was much lower than that for the city of Ufa (0.51) (p < 0.001). Stratified
analyses revealed that this lower ratio was observed only among those paternally exposed to
TCDD. Specifically, the proportion of male births among exposed fathers was 0.38 and among
exposed mothers was 0.51. This pattern was observed in both the workers exposed to 2,4,5-T
(proportion of male births = 0.40) and 2,4,5-trichlorophenol (proportion of male births = 0.35).
2.4.1.2.1.8.3.2. Study evaluation.
The Ryan et al. (2002) findings are consistent with earlier work completed for Seveso
residents (Mocarelli et al., 2000). Although serum measures were available for 84 individuals,
no dose-response of birth ratios was performed using exposure quantified at an individual-level
basis. This approach would have been preferred and consistent with that which Mocarelli et al.
(2000) used. All comparisons were made using an external comparison group, namely the sex
ratio observed in Ufa between 1959 and 1996.
Although serum measures were used to describe TCDD exposure for a sample of the
workers, individual-level dose estimates were not calculated for the study population.
Specifically, exposures were characterized many years after exposure, and no attempt was made
to back-extrapolate to the time of conception. The two groups of workers in the study also
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-124 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
reportedly had high exposure levels of 1,2,3,7,8-pentachlorodibenzo-p-dioxin. So, the group
level exposure classification (by plant) did not allow consideration of confounding due to other
dioxin-like compounds. Another limitation of the study is that the study population is likely
nonrepresentative of all workers employed at the plant. Participants included only those willing
to provide serum samples and those who volunteered to participate in the study after learning
about it in a public forum. If participation was dependent on TCDD exposures and the
reproductive health of these subjects, then bias may have occurred.
2.4.1.2.1.8.3.3. Suitability of data for TCDD dose-response modeling.
The findings are notable in their consistency with those found in Seveso residents by
Mocarelli et al. (2000). For the Ryan et al. (2002) study, serum data were quantified at an
individual-level basis. Risk estimates, however, were not derived in relation to these exposures
but instead in two separate subgroups (2,4,5-T and 2,4,5-trichlorophenol workers). This
important limitation precludes the use of these data for quantitative dose-response modeling.
2.4.1.2.2. Feasibility of dose-response modeling for noncancer.
Relatively few study populations permit quantitative dose-response modeling to be
performed for noncancer outcomes. The serum collected among Seveso men and women
provide an opportunity to characterize risks for several health conditions in relation to TCDD
exposure. The collection of these serum samples, shortly after the accident does not require the
back-extrapolation of TCDD levels as in the occupational cohorts, which should reduce the
exposure assessment uncertainty and minimize the potential for exposure misclassification.
An added feature of the SWHS is the detailed collection of other risk factor data from
trained interviewers. These data allow for risk estimates to be adjusted for potential confounding
variables. For the evaluations of reproductive health outcomes, this adjustment is critical given
there are various documented risk factors for the different outcomes that were examined. For
some health outcomes, continued follow-up of the cohort is needed, given that several of the
Seveso studies suggest that those exposed at a very young age might be more susceptible to
subsequent adverse health effects.
The findings of positive associations and dose-response relationships with serum-based
measures of TCDD suggest several noncancer health outcomes could be associated with TCDD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-125 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
exposure. These health outcomes include neonatal thyroid function, sex ratio, diabetes, and
semen quality. Although findings have suggested an association between TCDD and age at
menopause, they were not statistically significant and no dose-response trend was observed.
Weak or nonstatistically significant associations have been noted for endometriosis and
menstrual cycle characteristics and do not support quantitative dose-response analyses.
Associations between TCDD exposure and cardiovascular disease have been noted in
some, but not all, of the occupational cohorts, and also shortly after the accident among Seveso
residents. Findings from the cohort studies based on external comparisons using the SMR
statistic should be interpreted cautiously due to potential bias from the healthy worker effect.
Because the magnitude of the healthy worker bias is recognized to be larger for cardiovascular
diseases than for cancer outcomes, risk estimates in some occupational cohorts might be
underestimated for cardiovascular outcomes. Information on cardiovascular risk factors
generally was not captured in these studies, and sensitivity analyses were generally designed to
examine risk estimates generated for cancer outcomes.
2.4.1.2.3. Summary of epidemiologic noncancer study evaluations for dose-response
modeling.
All epidemiologic noncancer studies summarized above were evaluated for suitability of
quantitative dose-response assessment using the TCDD-specific considerations and study
inclusion criteria. The results of this evaluation are summarized in a matrix style array (see
Table 2-3) at the end of the chapter, and descriptively in Appendix B. The key epidemiologic
noncancer studies suitable for further TCDD dose-response assessment are presented in
Table 2-5.
2.4.2. Summary of Animal Bioassay Studies
This section summarizes studies that have met the in vivo animal bioassay TCDD study
inclusion criteria (see Section 2.3.2) and are considered in the dose-response modeling conducted
later in this document (see Sections 4 and 5). The sections that follow summarize the
experimental protocol, the results, and the NOAELs and LOAELs identified in reproductive
studies, developmental studies, and general toxicity studies (subdivided by duration).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-126 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
To evaluate and discuss studies consistently, doses were converted to nanograms per
kilogram body weight per day (ng/kg-day) and were also adjusted for continuous exposure.
Some doses were adjusted based on daily dietary intake and body weight. For these studies,
EPA uses 10% of an animal's body weight as the daily feed rate. More commonly, doses were
adjusted from 5 days/week to a 7 days/week standard adjustment, in which case administered
doses were multiplied by 5 and divided by 7 to obtain continuous doses. To adjust for weekly
dosing, the weekly administered doses were multiplied by the administration frequency per week
(in days) and divided by 7 to give continuous doses.
Other exposure protocols used a single loading dose followed by weekly maintenance
doses. To adjust these doses, the loading dose was added to the maintenance doses multiplied by
the administration frequency, and this sum was divided by the exposure duration to give a
continuous dosing rate. The doses administered in single dose studies were not averaged over
the observation period.
2.4.2.1. Reproductive Studies
2.4.2.1.1. Bowman et al, 1989a, b (and related Schantz and Bowman, 1989; Schantz et al.,
1986).
Female rhesus monkeys (6 to 10 years old; 8 per treatment) were exposed to 0 or 5 ppt
(for 3.5 years), or 25 ppt (for 4 years) TCDD (purity not specified) (Bowman et al., 1989a, b;
Schantz and Bowman, 1989; Schantz et al., 1986). Female monkeys were mated to unexposed
males after 7 months (Cohort I) and 27 months (Cohort II) of exposure, then again 10 months
post-exposure (Cohort III). The average daily doses to mothers were equivalent to 0, 0.15, and
0.67 ng/kg-day. The 0.67 ng/kg-day dose group had reduced reproductive rates in both Cohorts I
(p < 0.001) and II (p < 0.025; Bowman et al., 1989a). The mean number of days of offspring
survival (p < 0.023) also decreased. No effects on birth weight or growth, or physical evidence
of toxicity (Bowman et al., 1989b) were observed. Behavioral effects were observed in the
offspring (Cohort I: 7, 6, and 0 offspring, respectively; Cohort II: 3, 5, and 0 offspring,
respectively; Cohort III: 6, 7, and 3, respectively). In the 0.67 ng/kg-day dose group, the number
of offspring was insufficient to form a group in either Cohorts I or II. Offspring in the
0.15 ng/kg-day dose group had alterations in social behavior of the mother-infant pairs (mothers
had increased care giving, which appeared to be an effect of the infants and not due to the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-127 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
treatment of the mother) and peer group of the offspring after weaning (Cohort I offspring were
more dominant or aggressive and exhibited more self-directed behavior; Bowman et al., 1989b).
The performance of learning tasks was inversely related to the level of TCDD in the body fat.
Schantz and Bowman (1989) examined effects using discrimination-reversal learning (RL) and
delayed spatial alteration (DSA). RL detected effects in the 0.15 ng/kg-day group as measured
by retarded learning of the shape reversal (p < 0.05), but DSA did not. Schantz et al. (1986)
combined the cohorts and looked at 5, 5, and 3 mother-infant pairs in the 0, 0.15, and
0.67 ng/kg-day groups, respectively. They found that TCDD-exposed mother-infant pairs spent
more time in close, social contact compared to the controls (mutual ventral contact,/? < 0.025;
nipple contact,/? < 0.01) and infants had reduced locomotor activity (p < 0.05), but the
dose-effect was complex. Of note is that the control groups contained fewer males than did the
TCDD-exposed groups.
In a follow-up study, Rier et al. (2001) examined the DLC levels of sera collected from
some monkeys in this study. They reported that animals in this study had elevated serum PCB77
and PCB126 levels and an increased serum TEQ. In fact, the fractional contribution of serum
TCDD levels to total serum TEQ was 30% in treated animals. In this study, it is not possible to
determine the contribution of TCDD alone to the developmental effect due to the background
contamination; thus, EPA has not developed a TCDD LOAEL from the study.
2.4.2.1.2. Hochstein et al., 2001.
Adult female mink (12/treatment group) were administered dietary concentrations of
0.0006 (control), 0.016, 0.053, 0.180, or 1.40 ppb TCDD (purity >99.8%) for 132 days
(Hochstein et al., 2001). This dose is estimated to be equivalent to 0.03 (control), 0.8, 2.65, 9,
and 70 ng/kg-day assuming a food consumption of 5% of body weight per day. Females were
mated with unexposed males beginning on treatment day 35. Females were allowed to mate
every fourth day during a 29-day mating period or until a confirmed mating. Mated females
were presented with a second male either the day after initial mating or 8 days later. In the
70 ng/kg-day group, the treated animals were lethargic after 4 to 5 weeks, with several having
bloody (tarry) stools near the end of the trial. Two animals in the 70 ng/kg-day dose group died
prior to study termination. These animals had lost a large percentage of their body weight
(24-43%)), and had pale yellow livers and intestinal hemorrhages. Histopathology from both
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-128 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
mink indicated marked diffuse hepatocellular vacuolation. The mean body weight decreased in
all treatment groups including the control (losing an average of 3.29% of initial body weight),
compared to a dose-dependent loss of up to 26% in the 70 ng/kg-day group. Mating and
reproduction were considered subnormal in all groups. The number of females that gave birth in
the 0.03 (control), 0.8, 2.65, 9, and 70 ng/kg-day dose groups were 5/12, 0/12, 3/12, 8/12, and
0/11, respectively. The study authors speculated that the subnormal breeding and reproductive
performances in the control females likely were due to the indoor environment in which the mink
were housed. In the three groups that gave birth, there was a dose-dependent decrease in kit
body weight at birth, which was significant (p < 0.05) in the 9 mg/kg-day group compared to the
controls. The body weight in the kits was not significantly different at 3 or 6 weeks after birth.
Three-week survival rates of 71, 47, and 11% were recorded for kits in the 0.03 (control), 2.65,
and 9 ng/kg-day dose groups, respectively. Six-week kit survival rates were 62, 29, and 11% in
the 0.03 (control), 2.65, and 9 ng/kg-day dose groups, respectively.
In the adult females, clinical signs of toxicity were noted in the 70 ng/kg-day group near
the end of the study and included alopecia and notably thickened, deformed, and elongated
toenails. There was a dose-dependent decrease in plasma total solids, total protein, and
osmolality that reached statistical significance (p < 0.05) in the two highest exposure groups.
Anion gap was significantly decreased (p < 0.05) and alanine aminotranferase was significantly
increased in the 70 ng/kg-day group compared to the controls. At terminal sacrifice, there was a
dose-related decrease in body weight. There was a dose-related increase in liver weight that
reached statistical significance (p < 0.05) in the 70 ng/kg-day dose group. The brains of 42% of
the animals in the 70 ng/kg-day dose group had localized accumulation of lymphatic cells within
the meninges with mild extension into the adjacent neuropil and mild gliosis. Of the 10 mink
surviving to study termination in the 70 ng/kg-day group, 3 had periportal hepatocellular
vacuolation. These same brain and liver lesions were not observed in the control mink.
As there were no litters produced in the low-dose group and pregnancy outcomes were
not dose related, the 0.8 ng/kg-day exposure level does not inform the choice of NOAEL or
LOAEL. Thus, the LOAEL for this study is 2.65 ng/kg-day (132-day maternal exposure
duration) based on reduced kit survival (47% of control at 6 weeks). A NOAEL cannot be
determined for this study.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-129 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2.4.2.1.3. Ikeda et al., 2005.
Ikeda et al. (2005) studied the effect of repeated TCDD exposure to F0 dams on the male
gonads of F1 generation and sex ratio in the F2 generation. Twelve female Holtzman rats were
treated with a single dose of 400 ng/kg TCDD (>98% purity) orally, via gavage, followed by
weekly treatment doses of 80 ng/kg TCDD (16.5 ng/kg-day adjusted for continuous exposure of
10 weeks; specified 2 weeks premating, assumed 1 week for successful mating, 3 weeks of
gestation, and specified 4 weeks to weaning) during mating, pregnancy, and lactational periods
(total exposure duration approximately 10 weeks). Corn oil served as the control in another
group of 12 dams. Four dams were sacrificed on gestation day (GD) 20 to evaluate the in utero
toxicity of TCDD. Litter sizes from the remaining eight dams were examined on postnatal day
(PND) 2, and some of the F1 offspring were sacrificed to estimate TCDD tissue concentrations.
The remaining offspring were weaned on PND 28. Some of the F1 (number not specified)
offspring were mated with untreated females on PND 98, following which, litter size, sex ratio,
weight, and anogenital distance of F2 pups were examined on PND 2. Mated and unmated F1
males were sacrificed and the testes, epididymis, seminal vesicle, and the ventral prostate were
weighed; the cauda epididymis was weighed and examined for sperm count.
All fetuses in the control and TCDD group as a result of in utero exposure in the F0
generation survived. Litter size, sex ratio, and anogenital distance in the F1 generation on
PND 2 were not altered as a result of in utero TCDD exposure. Pup weight was significantly
(p < 0.05) lower in the TCDD-treated group than in controls. TCDD concentration in the
adipose tissue of the F0 dams on GD 20 was significantly (p < 0.05) higher than in the liver.
Adipose TCDD was significantly (p < 0.01) reduced at weaning, however, compared to
concentrations on GD 20. F1 pup liver TCDD concentration increased significantly (p < 0.01)
and was higher on PND 28 than PND2. The liver weight in F1 males increased by 14-fold at
PND 28 compared to PND 2, implying a transfer of approximately 850 pg of TCDD from the
dam to the F1 pup livers during lactation. TCDD also was detected in pup adipose tissue on
PND 28. Body weight of TCDD-exposed F1 males was significantly (p < 0.001) lower than
control males at weaning (PND 28). No significant differences in testis and cauda epididymis
weights were observed between the control and treated groups. Ventral prostate weight in the F1
males exposed to TCDD, however, was approximately 60% lower than controls. No change in
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-130 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
weight of the body, brain, testes, cauda epididymis, or seminal vesicle was observed at PND 120.
Ventral prostate weight, however, was 16% lower than that of the control group (p < 0.001).
Sperm count in the cauda epididymis of the F1 males was not affected by TCDD exposure.
Examination of F2 generation litters indicated no significant differences in litter size, pup
body weight, and anogenital distance between TCDD-treated or vehicle control groups. The
percentage of male F2 pups born to maternally and lactationally TCDD-exposed males was
significantly (p < 0.05) lower (38%) than those sired by control group males (52%). Every
female mated with maternally TCDD-exposed F1 males delivered more female than male pups.
A LOAEL for TCDD of 16.5 ng/kg-day for an estimated 10 week exposure duration in
F0 rat dams is identified in this study for decreased development of the ventral prostate in the F1
generation (60% lower than controls) and for significantly (p < 0.05) altered sex ratio (decreased
percentage of males) in the F2 generation. A NOAEL cannot be determined for this study.
2.4.2.1.4. Ishihara et ah, 2007.
Ishihara et al. (2007) examined the effect of repeated TCDD exposure of F0 males on the
sex ratio of F1 offspring. Seven-week-old male ICR mice (n = 127) were divided into three
groups and treated via gastric intubation with an initial loading dose of either 2 or 2,000 ng
TCDD/kg BW or an equivalent volume of sesame oil (vehicle) as control, followed by a weekly
maintenance doses of 0, 0.4, or 400 ng/kg until the animals were 12 weeks old. One week after
the last exposure, the animals were mated with untreated female mice. On the day a vaginal plug
was identified, F0 male mice were sacrificed and major organs including testes, epididymis, and
liver were removed and weighed. Organ tissues also were examined for histopathological and
immunohistochemical changes. Treatment levels, averaged over the 6 week period from start of
treatment to mating (five maintenance doses), were 0, 0.095, and 950 ng/kg-day for the control,
low dose and high dose groups, respectively.
All TCDD-treated males successfully impregnated untreated females and yielded viable
offspring. Mortality, pup weights, and mating and fertility indices were not affected by TCDD
exposure. There were no significant differences in body weights or in relative weights of testes,
epididymis, or livers in the TCDD-treated F0 males compared to the control group. The livers of
some animals (number not specified) in the high-dose group, however, were larger and heavier
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-131 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
than in the controls or the low-dose group. Hence, tissues from the high-dose animals were
selected for detailed immunohistochemical examination.
General histopathological findings in the TCDD-treated groups showed no changes in
cell morphology in germ, Sertoli, and Leydig cells of the testes. Arrangement of the germ cells
was normal and there was no difference in the epididymis spermatozoon number in either of the
TCDD-treated groups compared to controls. Livers of some of the animals in the high-dose
group however, showed enlarged and vacuolated areas in the centrilobular area when compared
to the low-dose group and the control group. Immunohistochemical and quantitative
immunohistological findings showed a marked increase in staining intensity for cytochrome
P450 (CYP)l A1 in the cytoplasm of the hepatocytes in the centrilobular area of the high-dose
TCDD group compared to the cells in the low-dose and the control groups. In addition,
proportions of immunoreactive CYP1A1 areas in the liver sections of the high-dose group were
higher than in the low-dose and control groups. The proportions of immunoreactive CYP1A1
also varied across animals (n = 33) in the high-dose group.
In addition to the above findings, there was a dose-related decrease in the male/female
sex ratio. The proportion of male offspring of the high-dose group was significantly lower
(p < 0.05) than that observed in controls (46.2% versus 53.1%, respectively). Hepatic
immunoreactive CYP1A1 staining levels in individual F0 males were strongly correlated with
the sex ratio of their offspring.
A LOAEL for TCDD of 950 ng/kg-day for a 6 week exposure duration of F0 male mice
is identified for significantly (p < 0.05) decreased male/female sex ratio (i.e., higher proportion
of female offspring) in the F1 generation. The NOAEL is 0.095 ng/kg-day.
2.4.2.1.5. Latchoumycandane and Mathur, 2002 (and related: Latchoumycandane et al.,
2002a, b, 2003).
Latchoumycandane and Mathur (2002) conducted a study to determine whether treatment
with vitamin E protected rat testes from TCDD-induced oxidative stress. Groups of albino male
Wistar rats (n = 6) were administered an oral dose of 0 (vehicle alone) 1, 10, or 100 ng
TCDD/kg-day for 45 days, while another group of animals (n = 6) was co-administered TCDD at
the same doses, along with vitamin E at a therapeutic dose of 20 mg/kg-day for 45 days. At
study termination, animals were fasted overnight, weighed, and sacrificed. Testis, epididymis,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-132 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
seminal vesicles, and ventral prostate were removed, weighed, and preserved for further
examination. The left testis was used to determine daily sperm production, while the right testis
was used for biochemical studies. Superoxide dismutase, catalase, glutathione reductase, and
glutathione peroxidase activity were measured in the testes, along with production of hydrogen
peroxide and lipid peroxidation.
Body weights of TCDD-treated rats did not differ significantly from the control group.
Testis, epididymis, seminal vesicle, and ventral prostate weights in the TCDD-treated groups,
however, decreased significantly (p < 0.05) when compared to controls. None of these changes
were observed in the TCDD-exposed groups receiving vitamin E. There was a dose-related
decrease in daily sperm production (p < 0.05) in all three TCDD-treated groups when compared
to the control group. In contrast, the TCDD treatment groups that also received vitamin E did
not show any significant changes in daily sperm production compared to the controls. The
TCDD-treated groups also showed significantly (p < 0.05) lower activities of the antioxidant
enzymes (superoxide dismutase, catalase, glutathione reductase, and glutathione peroxidase) than
the control group. Levels of hydrogen peroxide and lipid peroxidation increased significantly
(p < 0.05) in the testes of the rats treated with TCDD compared to the corresponding controls.
The TCDD-treated groups that had been co-administered vitamin E show no difference in
antioxidant enzyme activities or in reactive oxygen species production when compared with
controls.
A LOAEL for TCDD of 1.0 ng/kg-day for a 45-day exposure duration in rats is identified
in this study for significantly (p < 0.05) reduced sperm production and significantly (p < 0.05)
decreased reproductive organ weights. A NOAEL cannot be determined for this study.
2.4.2.1.6. Murray et al, 1979.
Male (10-16 per treatment) and female (20-32 per treatment) Sprague-Dawley rats were
administered diets containing TCDD (purity >99%) to achieve daily concentrations of 1, 10, or
100 ng/kg-day through three generations. After 90 days of treatment, F0 rats were mated to
produce Fla offspring. Thirty-three days after weaning of the last Fla litter, the F0 rats were
mated again to produce Fib offspring. Some F0 rats were mated a third time for a cross-mating
study. The Fib and F2 rats were mated at about 130 days of age to produce the F2 and F3
generations. No clinical signs of toxicity or changes in body weight and food consumption were
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-133 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
observed in F0 rats during the 90 days of treatment before mating. The 100 ng/kg-day group was
discontinued due to the lack of offspring. In the three surviving offspring (all males), no changes
in appearance, body weight, or food consumption occurred. A dose of 10 ng/kg-day caused a
consistent decreased body weight in both sexes of F1 and F2 rats, which was associated with
decreased food consumption. A significant (p < 0.05) decrease in fertility in F1 and F2 rats
occurred, but not in F0 rats, administered 10 ng/kg-day. The number of live pups and gestational
survival index were significantly (p < 0.05) decreased in the 100 ng/kg-day F0 rats and in the
10 ng/kg-day F1 and F2 rats. The gestational survival index also was significantly (p < 0.05)
decreased in F2 rats administered 1 ng/kg-day. Postnatal survival was significantly (p < 0.05)
reduced only in F2 rats administered 10 ng/kg-day. Growth (as measured by body weight) was
affected at 10 ng/kg-day only in the third generation. In the 10 ng/kg-day group, a significant
(p < 0.05) decrease in relative thymus weight and increase in liver weight also occurred in F3 rats
(weights were not measured in F2 rats). Additionally, mating 100 ng/kg-day TCDD-treated
females with untreated males increased the percent of implants resorbed as assessed by uterine
histopathology.
The reproductive LOAEL is 10 ng/kg-day, based on a significant (p < 0.05) decrease in
fertility (33-37% lower than controls); decrease in the number of live pups (18-27% lower than
controls); decrease in gestational survival (10-11% lower than controls); decrease in postnatal
survival (32% lower than controls); and decreased postnatal body weight (14—19% lower than
controls at weaning) in one or more generations. The reproductive NOAEL is 1 ng/kg-day.
2.4.2.1.7. Rier et al., 1993,1995.
Reir et al. (1993, 1995) examined the impact of chronic TCDD exposure on
endometriosis in monkeys. Female rhesus monkeys (8 animals per treatment group) were
exposed to 0, 5, or 25 ppt TCDD (purity not specified) in feed for 4 years. Previously, Bowman
et al. (1989b) determined that these dietary concentrations were equivalent to 0, 0.15, and
0.67 ng/kg-day, respectively. Ten years after termination of TCDD treatment, the presence of
endometriosis was determined via laparoscopic surgical procedure, and the severity of the
disease was assessed. The study authors reported that three monkeys in the 0.67 ng/kg-day
exposure group died at 7, 9, and 10 years after termination of TCDD treatment. Autopsy results
attributed the deaths to widespread and severe peritoneal endometriosis (all three monkeys)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-134 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
along with obstruction of the colon (one monkey) and blockage of the jejunum (one monkey).
Other deaths also occurred in the control group (1 death from birthing complications and another
from an unknown cause); in the 0.15 ng/kg-day dose group (1 death due to natural causes with
no endometriosis), and in the 0.67 ng/kg-day dose group (1 death due to a breeding fight with no
incidence of endometriosis). At study termination, 17 live animals plus the 3 that had previously
died of endometriosis were evaluated (total n = 20).
Incidence of endometriosis was significantly (p < 0.05) higher than in the control group
with 71 and 86 % incidence rates in the 0.15 and 0.67 ng/kg-day dose groups, respectively,
compared to 33% in the control group. Severity of endometriosis was also significantly
(p < 0.001) correlated with TCDD dose. Staging by rAFS indicated that untreated control
animals had either minimal or no incidence of endometriosis. In comparison, endometriosis was
absent in 2 of the 7 monkeys in the 0.15 ng/kg-day dose group, while only 1 of the 7 animals in
the high dose group was disease free. Moderate-to-severe disease was observed in 3 of the
7 animals in the 0.15 ng/kg-day dose group and 5 of the 7 animals in the 0.67 ng/kg-day dose
group. Moderate-to-severe disease was not observed in the control group. The authors also
compared the incidence and severity of endometriosis in TCDD-exposed animals with
304 normal, non-neutered females with no dioxin exposure and reported that the disease was not
present in monkeys that were less than 13 years of age, while the disease rate was 30% among
animals 13 years of age or older. The study authors report that these findings are in agreement
with human and rhesus studies demonstrating that the prevalence of detectable endometriosis can
increase with advanced age.
As noted previously, in a follow-up study, Rier et al. (2001) examined the DLC levels of
sera collected from some monkeys in this study. They reported that animals in this study had
elevated serum PCB77 and PCB126 levels and an increased serum TEQ; the fractional
contribution of serum TCDD levels to total serum TEQ was 30% in treated animals. They also
reported that the severity of the endometriosis corresponded to the serum PCB77 concentrations
rather than total TCDD. In this study, it is not possible to determine the contribution of TCDD
alone to the edometriosis due to the background contamination; thus, EPA has not developed a
TCDD LOAEL from the study.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-135 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.2.1.8. Shi et al., 2007.
Pregnant Sprague-Dawley rat dams (3 per treatment group) were administered 0, 1,5, 50,
or 200 ng/kg TCDD (purity >99%) in corn oil by gavage on GD 14 and GD 21 and on PND 7
and PND 14 for lactational exposure to pups (Shi et al., 2007). Ten female pups per treatment
were selected and administered TCDD weekly at the same dose levels through their reproductive
lifespan (approximately 11 months). The corresponding equivalent daily TCDD doses are 0,
0.14, 0.71, 7.14, and 28.6 ng/kg-day. Vaginal opening was slightly but significantly (p < 0.05)
delayed in 28.6 ng/kg-day females. Vaginal opening was also delayed, but not significantly, in
the 0.14 and 7.14 ng/kg-day groups. Reproductive senescence with normal cyclicity was
significantly (p < 0.05) accelerated beginning at 9 months in 7.14 and 28.6 ng/kg-day females.
Serum estradiol concentrations were decreased at all time points across the estrous cycle in a
dose-dependent manner with a statistically significant decrease (p < 0.05) in all but the lowest
dose group. TCDD exposure, however, did not affect the number or size distribution of ovarian
follicles; responsiveness of the pituitary gland to gonadotropin-releasing hormone, or serum
profiles of FSH, LH, or progesterone.
A LOAEL for TCDD of 0.71 ng/kg-day for an 11-month exposure duration was
identified in this study based on significantly (p < 0.05) decreased estradiol levels in offspring.
The NOAEL for this study is 0.14 ng/kg-day.
2.4.2.1.9. Yang et al., 2000.
Yang et al. (2000) studied the impact of TCDD exposure on the incidence and severity of
endometriosis in female rhesus monkeys. Groups of 7- to 10-year old nulliparous cynomolgus
monkeys were treated with 0 (n = 5), 1, 5, or 25 (n = 6 per group) ng/kg BW TCDD 5 days per
week via gelatin capsules for 12 months. Because the monkeys received one capsule 5 days per
week, the doses adjusted for continuous exposure were 0, 0.71, 3.57, and 17.86 ng/kg-day. Prior
to TCDD administration, all animals had endometriosis induced during days 12-14 of the
menstrual cycle by auto-transplantation of endometrial-strips in multiple abdominal sites. All
TCDD-treated and control groups were laparoscopically examined during months 1, 3, and 6 to
monitor the survival of endometrial implantations and to obtain peritoneal fluid to determine the
concentration and immunotype of endometrial growth regulator cytokines interleukin-6 (IL-6)
and interleukin-6 soluble receptor (IL-6sR). Because insufficient peritoneal fluids were present
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-136 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
in the treated and control monkeys, however, the study authors collected blood samples at 6 and
12 months during laparoscopy for routine hematology and to assess the circulating levels of IL-6
and IL-6sR. All animals were sacrificed at 12 months, and circulating levels of gonadal steroids
also were measured at the time of necropsy.
No changes were observed among treatment levels in general toxicological endpoints
such as body weight changes, food consumption, hematological endpoints, general activity
levels, and caretaker interaction. In addition, TCDD did not impact circulating levels of gonadal
steroids measured during necropsy. Similarly, there were no differences in the number of
menstrual cycles, the length of the menstrual cycle, and bleeding intervals. Endometrial implants
were found in at least one site in all TCDD-treated and control monkeys during the first
laparoscopic examination. Follow-up laparoscopies revealed that there was a continuous loss of
endometrial implants over time in each dose group. At the 1-, 3-, and 6-month examination, the
number of endometrial losses was not significantly different among different dose groups. At
the 12-month examination, however, a significantly (p < 0.05) higher rate of survival of
endometrial implants was observed in the 3.57 and 17.86 ng/kg-day dose groups compared to the
control group. The highest rate of endometrial implant survival was observed in the ovaries
regardless of the dose group. In contrast, all lesions disappeared from the left broad ligament,
whereas two on the right broad ligament and one on the uterine fundus survived. There was a
dose-dependent divergence in the growth response of endometrial implants following TCDD
exposure. Both the maximum and minimum implant diameters in the 17.86 ng/kg-day dose
group were significantly (p < 0.05) larger compared to controls. In contrast, the maximum and
minimum implant diameters in the 0.71 ng/kg-day dose group were significantly (p < 0.05)
smaller compared to controls. TCDD did not impact implant diameters in the 3.57 ng/kg-day
dose group when compared to controls. Histological examinations revealed that endometrial
glands and stromal cells were present in all surviving implants. Sections examined in the
17.86 ng/kg-day of TCDD possessed cystic endometrial glands that were more frequently
observed in this dose group compared to other groups including controls. In addition, circulating
levels of IL-6 were significantly (p < 0.05) lower in monkeys exposed to 17.86 ng/kg-day TCDD
both at 6 and 12 months compared to the control group. In contrast, circulating levels of IL-6sR
were significantly (p < 0.05) higher in animals treated with 3.57 and 17.86 ng/kg-day TCDD at
6 months, while the levels were higher only in the 17.86 ng/kg-day TCDD group at 12 months.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-137 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
A LOAEL for TCDD of 17.86 ng/kg-day for a 1 year exposure duration was identified in
this study for significantly (p < 0.05) increased endometriosis induced by endometrial implant
survival, significantly (p < 0.05) increased maximum and minimum implant diameters, and
growth regulatory cytokine dysregulation (as assessed by significantly decreased IL-6 levels,
p < 0.05). ANOAEL of 3.57 ng/kg-day is identified in this study.
2.4.2.2. Developmental Studies
2.4.2.2.1. Amin et al., 2000.
Amin et al. (2000) studied the impact of in-utero TCDD exposure on the reproductive
behavior in male pups. Groups of pregnant Harlan Sprague-Dawley rats (n = 108 divided into
4 cohorts; number of animals in the TCDD treatment group is ~3 per dose group) were dosed via
gavage with 0, 25, or 100 ng/kg-day TCDD (purity >98%) in corn oil on GDs 10-16. On the
day of birth (PND 0), pups were examined for gross abnormalities and the number of live pups,
their weights, and sex were recorded from each litter. Litters consisting of more than eight pups
were reduced to eight, comprised of four males and four females when possible. Litters
consisting of fewer than five pups were excluded from the study to minimize between-litter
differences in growth rate, maternal behavior, and lactational exposure. After this exclusion,
approximately 10 to 11 litters per exposure group remained. All pups were weaned on day 21
and one male and one female were retained to assess reproductive development, play behavior,
reproductive behavior, and saccharin preference behavior. Both male and female pups were
tested for saccharin preference between 189 and 234 days of age. A saccharin preference test
was conducted for 8 days. For the first 4 days, rats were provided bottles containing tap water,
and on days 5 and 6 the animals were provided a bottle containing water and a bottle containing
0.25% saccharin solution. On days 7 and 8, the animals were provided water and a bottle
containing 0.50% of saccharin solution. A 0.50% saccharin solution was used because previous
studies have reported that male rats exhibited a greater reduction in preference for this saccharin
concentration compared to females, hence the sex difference in preference is more marked at this
saccharine dose.
None of the treated dams exhibited any signs of toxicity as a result of exposure to TCDD.
Gestational body weight, liver weight, litter size and percent live births were all comparable to
the corresponding control group. Birth rate and weaning weight of the pups also were not
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-138 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
affected by TCDD exposure. Sex-related water consumption, however, was significantly
(p < 0.001) affected during the first 4 days with female pups drinking more water per 100 g of
body weight compared to the respective male counterparts. Saccharin consumption was
significantly (p < 0.001) affected, with females consuming greater amounts of saccharin solution
per 100 g body weight compared to the corresponding males. Additionally, both male and
female pups drank significantly (p < 0.001) more of the 0.25% saccharin solution compared to
the 0.50% saccharin solution. Females of all exposure groups consumed less of both the 0.25
and 0.50%) saccharin solution compared to the same-sex control group. Comparisons of each
exposure group to the control group indicated that only the high TCDD exposure group
(100 ng/kg-day) different significantly (p < 0.05) compared to control in the consumption of
0.25%o saccharin solution. In contrast, for the 0.50% saccharin solution, both the low and high
TCDD dose groups differed significantly (p < 0.05 and p < 0.01, respectively) compared to the
control group. The saccharin preference of TCDD-exposed male rats did not differ from that of
the male control group. The TCDD-exposed females' preference for saccharin solution,
however, was significantly reduced in both the 25 (p < 0.05) and the 100 ng/kg-day (p < 0.005)
dose group compared to that of the female controls. The study authors state that the reduction in
saccharin consumption and preference in females could be due to the anti-estrogenic action of
TCDD and that recent research reports suggest that TCDD can decrease the level of estrogen
receptor (ER) mRNA by blocking the ability of ER to transactivate from the estrogen response
element.
A LOAEL for TCDD of 25 ng/kg-day for 7 days of gestational exposure is identified for
significantly (p < 0.05) decreased preference in the consumption of 0.25% saccharin solution. A
NOAEL cannot be determined for this study.
2.4.2.2.2. Bell et al., 2007a.
Bell et al. (2007a) examined the reproductive effects of TCDD in rats exposed during
development. Female CRL:WI (Han) rats were treated with TCDD (99% purity; dissolved in
acetone) in the diet at concentrations of 0 (acetone alone; n = 75), 28, 93, or 530
(n = 65/group) ng TCDD/kg diet, which provided average doses of 0, 2.4, 8, or 46 ng/kg-day,
respectively. Rats were exposed to TCDD 12 weeks prior to mating, during mating, and through
pregnancy. Dams were switched to the control diet after parturition. Litters from pregnant dams
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-139 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
were reduced to a maximum size of eight on PND 4 and to five males (if possible) on PND 21.
These males were left untreated until sacrificed (25/group, one/litter) on PND 70, while all
remaining animals were sacrificed on PND 120. All sacrificed animals were necropsied and
received a seminology examination. Prior to sacrifice, during weeks 12 and 13, 20 animals from
each dose group were tested for learning ability and motor activity, and were also administered a
functional observation battery. During postnatal week 16, groups of 20 male F1 rats from each
treatment group were paired with untreated virgin females for 7 days, and mated females were
killed on GD 16 and examined for terminal body weights, pregnancy status, number of corpora
lutea, and number of intrauterine implantations.
The study authors found no evidence of direct maternal toxicity from exposure to TCDD.
In the high-dose groups, 8 of 27 dams suffered complete litter loss compared to 3 dams in the
control group, but the difference was not statistically significant. Pup survival at PND 4 was also
lower in the high-dose group, but the difference again was not statistically significant.
A dose-related decrease in mean pup body weight was observed on PND 1, and this trend
continued throughout the lactation period. High-dose male pups had lower body weights when
compared to controls at PND 21, with this trend continuing over the course of the study.
Balanopreputial separation (BPS) was significantly (p < 0.05) delayed compared to controls in
all three treatment groups by 1.8, 1.9, and 4.4 days in the low-, medium-, and high-dose groups,
respectively. The study authors reported that adjustment for lower body weights observed at
PND 21 and PND 42 did not affect the estimate of delay in BPS. No adverse effects from
maternal treatment were observed on learning or in functional observational battery performance.
Offspring in the high-dose group exhibited less activity when compared to controls (p < 0.05)
when they were subjected to a test of motor activity for 30 minutes.
The median precoital time was 2-3 days for all 20 F1 males that were mated during
postnatal week 16. The uterine and implantation data were similar in all dose groups and there
were no significant differences in the proportion of male offspring between groups. Epididymal
sperm counts and sperm motility did not differ significantly between dose groups in animals
sacrificed during postnatal week 10. The mean number of spermatids was significantly lower
(14%; p < 0.05) and the proportion of abnormal sperm was significantly (p < 0.05) higher in the
high-dose group when compared to controls on PND 70. These effects, however, were not seen
in animals sacrificed on PND 120.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-140 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Terminal body weights were significantly (p < 0.05) decreased in the high-dose group
(6.9 %) compared to controls on PND 120, while the depression in body weight in the
medium-dose group (5.5%) was not statistically significant. At PND 70, the relative and
absolute testis weight of the high-dose group was less than the controls (12 and 18%,
respectively). Absolute spleen weight in the high-dose group was significantly higher (8%) on
PND 70, and increased significantly (p < 0.05) by 1-3% on PND 120 in all dose groups
compared to controls. Kidney weight in the low and medium-dose groups was significantly
(p < 0.05) greater than in controls (-2%) at PND 120. In addition to these organs, ventral
prostate (9.4%) and relative liver (—4.5%) weights were significantly (p < 0.05) higher than
controls on PND 120 in the medium- and low- and high-dose groups, respectively. On
PND 120, absolute brain weight was significantly (p < 0.05) less than the control in the
medium-dose group, while relative brain weight was significantly (p < 0.05) higher than the
control in the low- and high-dose group. Histological examination revealed no unusual findings.
A LOAEL for TCDD of 2.4 ng/kg-day following an estimated 17 week exposure duration
of dams was identified in this study for significantly (p < 0.05) delayed BPS. A NOAEL was not
identified in this study.
2.4.2.2.3. Franczak et al., 2006.
Franczak et al. (2006) examined the impact of chronic TCDD exposure on the onset of
reproductive senescence in female rats. Pregnant Sprague-Dawley rats (// = 2-3/dose group)
were fed 50 or 200 ng/kg TCDD (>99% purity) or corn oil vehicle (4 mL/kg) orally on GD 14
and 21 and PND 7 and 14 to provide in utero and lactational exposure to TCDD. On PND 21,
female pups (n = 7/dose group) were weaned and were subsequently given weekly doses of 50 or
200 ng/kg-week TCDD by gavage (7.14 or 28.6 ng/kg-day adjusted for continuous exposure;
administered doses divided by 7) or corn oil vehicle. Exposure continued for up to 8 months,
and animals were observed for changes in estrus cycle at 4, 6, and 8 months. Rats were
sacrificed at 8 months of age when the TCDD-treated animals had entered the transition to
reproductive senescence. Following sacrifice, diestrus concentrations of serum LH, FSH,
progesterone, and estradiol were measured, and the ovaries were collected for examination.
Estrus cycles at 4 months exhibited normal cyclicity in both TCDD-exposed groups and
did not differ significantly from the control group. At 6 months, however, there was a tendency
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-141 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
(p < 0.1) toward loss of normal estrus cyclicity in animals treated with TCDD. At the 8 month
observation, estrus cyclicity was significantly (p < 0.05) different in both dioxin-exposed groups
compared to controls (cumulative TCDD exposure is reported as 1.7 and 8 |ig/kg for the 50 and
200 ng/kg dose groups, respectively). The study authors noted that although the low-dose
animals showed an increased prevalence of prolonged cycles, persistent estrus or diestrus was
observed in only 10% of the rats. Conversely, approximately 50% of the rats exhibited loss of
cyclicity in the high-dose group. There were no changes in the number and size distribution of
ovarian follicles or the number of corpora lutea at either dose. Progesterone levels at 8 months
tended to be higher (p < 0.08) in animals receiving either 7.14 or 28.6 ng/kg-day TCDD
compared to controls, while serum estradiol concentrations were significantly (p < 0.03) lower at
diestrus. Serum LH levels in TCDD-treated animals were comparable to those in the control
group, while FSH levels were elevated in rats receiving 7.14 ng/kg-day TCDD, but not in the
28.6 ng/kg-day dose group.
A LOAEL for TCDD of 7.14 ng/kg-day for an 8-month exposure duration was identified
for significantly (p < 0.03) decreased serum estradiol levels. A NOAEL cannot be determined
for this study.
2.4.2.2.4. Hojo et al, 2002 (and related: Zareba et al., 2002).
Hojo et al. (2002) studied the impact of prenatal exposure to TCDD on sexually
dimorphic behavior in rats. Thirty-six pregnant Sprague-Dawley rats were assigned according to
a randomized block design to groups receiving 0, 20, 60, or 180 ng/kg TCDD (98% purity) on
GD 8. Litters from pregnant dams were culled to 5 females and 5 males on PND 4 and allowed
to wean normally, at which time 5, 5, 6, and 5 litters from the 0, 20, 60, and 180 ng/kg TCDD
treatment groups, respectively, were maintained for examination of behavioral response.
Offspring were exposed to TCDD (from a single maternal exposure) for about 35 days through
gestation and lactation. After weaning at PND 21, offspring were fed ad libitum until PND 80, at
which time a fixed amount of food was supplied daily to maintain constant body weights. At
90 days old, the rats in these treatment groups were trained to press a lever to obtain food pellets
using two operant behavior procedures. Initially, each lever press was reinforced. The
fixed-ratio (FR) requirement was then increased every fourth session from the initial setting of 1
to values between 6 and 71. The responses for 30 days were studied under a multiple schedule
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-142 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
combining FR 11 and another schedule requiring a pause of at least 10 sec between responses
(differential reinforcement of low rate, or DRL 10-sec)
Pup and dam body weights were not affected by TCDD exposure, and all pups were
successfully trained in the lever-press response within 3-4 days. Analyses of the FR procedure
data indicated that the male pups responded at a lower rate at all TCDD doses when compared to
the control group. In case of female pups, all TCDD-treated groups responded at a higher rate
than controls. None of these results were, by themselves, however, statistically significant.
Examination of the FR 11 and DRL 10-second data indicated that when considering the FR
component of this multiple procedure, males from all three treatment groups responded at lower
rates when compared to the controls. Conversely, all female pups responded at a higher rate than
controls. In addition, the treatment-by-sex interaction was significant (p = 0.036), with the
60 ng/kg female pups responding at a higher rate than the 60-ng/kg male pups. Examination of
the delayed response component in the multiple FR 11 and DRL 10-sec procedures indicated that
almost all TCDD treatment groups were affected. Like the FR component, male pups at all
TCDD dose groups responded at a lower rate compared to controls, while female pups at all dose
groups responded at a higher rate than controls. There was also a significant (p = 0.001)
sex-by-treatment interaction for the DRL 10-sec similar to the FR component. Following
behavioral testing, the animals were sacrificed and cortical depth measurements were taken in
selected right and left brain regions. Reduced cortical thickness and altered brain morphometry
were observed in both male and female offspring in the 180-ng/kg exposure group when
compared to controls (reported in a separate article; Zareba et al., 2002).
A nominal LOAEL for TCDD of 20 ng/kg for a single exposure on GD 8 is established
for this study based on abrogation of sexually dimorphic neurobehavioral responses. A NOAEL
cannot be derived for this study.
2.4.2.2.5. Kattainen et al., 2001.
Pregnant Line A, B, and C rats derived from Han/Wistar and Long-Evans rats
(4-8 pregnant dams/strain/treatment group) were administered a single gavage dose of 0, 30,
100, 300, or 1,000 ng/kg TCDD (purity >99%) in corn oil on GD 15 (Kattainen et al., 2001). On
PND 1, the litters were culled to three males and three females. Offspring were weaned on
PND 28. Female pups were sacrificed on PND 35 and male pups were sacrificed on PND 70.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-143 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
TCDD treatment did not affect body weight or cause clinical signs of toxicity in the dams. In
Line B offspring, body weights in the 1,000 ng/kg group were slightly decreased during
PND 1-7, while Line C offspring had slightly decreased body weights throughout the study
period (data were not provided). The development of the third molar was affected the most in
Line C offspring. In 5 of 10 Line C females and 6 of 10 Line C males treated with 1,000 ng/kg
TCDD, the lower third molar did not develop. In comparison, 1 of 19 Line A females and 1 of
18 Line B females administered 1,000 ng/kg TCDD lacked the third molar at sacrifice. Third
molars were present in all the controls and all male Line A and B offspring administered
1,000 ng/kg. Due to the lack of eruption of the third molar in the majority of Line B and C
control females (only 30% erupted), however, the effects of TCDD on third molar eruption could
only be evaluated in Line A female offspring (with 94% eruption). There was a dose-dependent
decrease in the eruption of the lower third molar in Line A female offspring with a significant
(p < 0.05) decrease observed in the 300 and 1,000 ng/kg dose groups. In the male offspring, any
third molar that developed erupted by PND 70. The mesiodistal length of the existing lower
third molar was reduced in a dose-dependent manner in both genders of all three rat lines. In
Line A and C females, the decrease was significant (p < 0.05) at all doses. The size of the
second molars was also significantly decreased with 1,000 ng/kg (p < 0.05) in all but Line C
males.
A developmental LOAEL for TCDD of 30 ng/kg for maternal exposure on GD 15 is
established for this study, based on impaired tooth development (significantly reduced
mesiodistal length of the lower third molar by approximately 12% to 38% \p < 0.05]). A
NOAEL could not be determined.
2.4.2.2.6. Keller et al., 2007, 2008a, b.
Keller et al. (2007, 2008a, b) conducted three separate experiments to assess the impact
of TCDD on molar tooth development using different mouse strains. In Experiment 1, Keller et
al. (2007) used six inbred mouse strains (C57BL/6J, BALB/cByJ, A/J, CBA/J, C3H/HeJ, and
C57BL/10J) known to possess high affinity ligand-binding aryl hydrocarbon receptor alleles (b),
two with bl alleles (C57BL/6J and CBA/J), and four with b2 alleles (BALB/cByJ, A/J,
C3H/HeJ, and CBA/J). Females (number not specified) from each strain were mated with males
of the same strain. On GD 13, each pregnant female was assigned to one of the four dose groups
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-144 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
and treated with 0, 10, 100, or 1,000 ng TCDD/kg BW via oral gavage. The control group
received corn oil. GD 13 was chosen for dosing because the first morphological signs of tooth
development occur on GD 11. The first visible signs of the Ml (molar) occur on GDs 13-14
followed by final cuspal morphology, which is determined on GD 15. The F1 offspring of
females from each strain were weaned and separated by sex at PND 28 and were euthanized at
PND 70. Each F1 mouse was examined for the presence or absence of both maxillary (M3) and
mandibular third molars (M3) on both the left and right sides. In addition, all mice were scored
as either normal or variant in Mi morphology for both molar rows.
In Experiment 2 (Keller et al., 2008a), dams from six inbred mouse strains (C57BL/6J,
BALB/cByJ, A/J, CBA/J, C3H/HeJ, and C57BL/10J) were orally dosed on GD 13 with 0, 10,
100, or 1,000 ng TCDD/kg BW in corn oil. GD 13 was used as the dosing day because it
coincided with the formation of Meckel's cartilage (a major signal center) in the mouse mandible
that is followed shortly by intramembranous bone formation on GD 15. The A/J mouse strain
was abandoned because the authors had difficulty rearing the offspring from this strain. All
offspring (n = 4 or 5 per treatment group) from the remaining strains were euthanized at 70 days
of age. Mandible size and shape from all selected offspring were examined using geometric
morphometric methods to assess the impact of TCDD exposure.
In Experiment 3 (Keller et al., 2008b), dams from six inbred mouse strains (C57BL/6J,
BALB/cByJ, A/J, C3H/HeJ, CBA/J, and C57BL/10J) were treated with a single oral dose of 0,
10, 100, or 1,000 ng TCDD/kg-BW in corn oil. GD 13 was chosen as the dosing day because the
first visible signs of the first molar (Ml) occurs on GDs 13-14 and the final cuspal morphology
(the pattern of projections on the chewing surface of the tooth) is not determined until after
GD 15. Similar to Experiment 2, the A/J mouse strain was abandoned due to difficulty in rearing
offspring. All offspring (n = 107-110 in each of the five strains for all treatment groups) were
euthanized at 70 days of age and their molar size, shape, and asymmetry traits were examined
using geometric morphometric methods.
In Experiment 1, all four M3s were present in all dose groups in mice from C57BL/6J,
BALB/cByJ, and C57BL/10J strains. A similar response was observed in the A/J strain mice
with only 3 of 51 F1 mice exhibiting missing third molars. Approximately one-third of the mice
from the CBA/J and C3H/HeJ strains, however, were missing at least one M3 or M3 molar. The
numbers of CBA/J mice missing one or both M3 or M3 molars were 0/29, 2/21, 6/29, and 30/30
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-145 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
in the 0, 10, 100, and 1,000 ng/kg groups, respectively. In the C3H/Hej animals, the numbers
missing one or both molars were 1/24, 3/28, 1/26, and 30/36, respectively.
Maternal TCDD exposure was also found to affect the frequency of Mi variants, but only
in the C57BL/10J strain, and the dose-response relationship was nonmonotonic. The proportions
of variants observed in the 0, 10, 100, and 1,000 ng/kg dose groups were 33, 68, 59, and 58%,
respectively.
A LOAEL for TCDD of 10 ng/kg maternal exposure on GD 13 is identified for this study
for increased incidence (33%) of the Mi variant in the C57BL/10J mouse strain. A NOAEL
cannot be determined in this study.
In Experiment 2 TCDD exposure of dams did not affect offspring survival or 10-week
body weight in any of the inbred mouse strains used. Analysis of variance (ANOVA) indicated
that although mandible size in both male and female offspring varied significantly (p < 0.0001)
among strains, it was not affected by TCDD exposure. In contrast, analysis of covariance
indicated that TCDD exposure significantly (p = 0.0033) decreased the mandible size in male
offspring in the C3H/HeJ strain at all treatment groups. The mean mandible size was similar
across all treatment groups in both sexes in all strains with male offspring exhibiting larger
mandibles compared to females. Males in the C3H/HeJ strain exhibited a significant (level not
reported) downward trend in mandible size throughout all treatment groups. Females in the C3H
strain also showed a similar trend in mandible size, but the trend was not significant. ANOVA
on mandible shape indicated that males had significantly (p < 0.0001) different mandible shape
in strain x treatment groups. In contrast, in female offspring, although the mandible shape was
significantly (p < 0.0001) different due to strains, treatment groups, and litter, the strain
x treatment interaction was not significant. Male offspring from the C3H/HeJ and C57BL/6J
mouse strains appear to be more sensitive to TCDD than BALB/cByJ or CBA/J mice, with the
C57BL/10J strain exhibiting intermediate sensitivity. In addition to these analyses, Procrustes
distance analysis also indicated that C3H/HeJ mice had the greatest response to the highest dose
of TCDD, followed by the C57BL/6J strain. Female offspring in the C3H/HeJ and C57BL/6J
strains also exhibited the largest change in Procrustes distance with TCDD exposure. This trend,
however, was not statistically significant (p = 0.29).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-146 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
A LOAEL for TCDD of 10 ng/kg maternal exposure on GD 13 was identified for this
study for significantly (p = 0.0033) decreased mandible shape and size in male C3H/HeJ mice.
A NOAEL cannot be determined in this study.
In Experiment 3, effect of TCDD exposure on offspring survival or body weight was not
reported. Three-way ANOVA results showed significant (p < 0.0001) differences in molar size
among strains, sexes, and litters, but not between treatment groups. Molar size difference in sex
x strain interaction was significant (p = 0.03), whereas differences in sex x treatment and sex
x strain x treatment were not significant. Additionally, molar size in treatment x strain
interaction also was not statistically significant. Based on these results, the authors reported that
molar size varied significantly (p < 0.0001) among all five strains tested, with all strains
exhibiting similar trends in all four treatment groups. Strain differences in molar size were more
apparent in male offspring. A hormesis-like trend in molar size was observed in all strains
(except in BALBc/ByJ) and sexes with an increase at the 100 ng/kg dose and a decrease in the
1,000 ng/kg dose. In addition to lack of difference in molar size for all treatment groups in all
strains, fluctuating asymmetry in molar size also did not increase with increasing doses of
TCDD.
In contrast to these results on molar size, the Procrustes ANOVA indicated that molar
shape was significantly (p < 0.0001) affected by strain, sex, treatment, and litter size. Molar
shape in sex x strain and sex x strain x treatment interactions was also highly significant
(p < 0.0001). Based on these results, the authors concluded that differences between males and
females varied based on the strain, and that the effect of TCDD exposure on each strain also
differed for male and female offspring. Because molar shape in treatment x strain interaction
was significant (p < 0.0001), differences in molar shape between the three treatment groups and
the control group were analyzed for each strain using nonorthogonal contrasts. In male
offspring, contrasts between the control group and 1,000 ng/kg were statistically significant only
in the C3H/HeJ (p < 0.0001) and CBA/J (p < 0.03) strains. These results suggest that these two
strains are most susceptible to TCDD effect on molar shape, and similar results were observed in
female offspring of these two strains. The contrast in molar shape between the control and the
100 ng/kg treatment group for the female C57BL/6J mice also was statistically significant
(p = 0.0096). On the whole, when considering Procrustes distance results for molar shape, the
C3H/HeJ male offspring had the largest response at the low and high doses, while the female
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-147 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
offspring had the largest response at low and mid doses. This observation in male C3H/HeJ
mice is consistent with that of TCDD-induced changes in mandible size from Keller et al.
(2008a).
A LOAEL for TCDD of 10 ng/kg maternal exposure on GD 13 is identified for this study
for significant (p < 0.0001) differences in molar shape in male C3H/HeJ mice. A NOAEL
cannot be determined in this study.
2.4.2.2.7. Kuchiiwa et al, 2002.
Kuchiiwa et al. (2002) studied the impact of in utero and lactational TCDD exposure on
serotonin-immunoreactive neurons in raphae nuclei on F1 male mouse offspring. Twenty-one
adult female ddY mice (seven per treatment group) were administered TCDD (99.1% purity) by
oral gavage once a week for 8 weeks at doses of 0, 4.9, or 490 ng/kg (0, 0.7, or 70 ng/kg-day
average daily dose; administered doses divided by 7) or an equivalent volume of olive oil vehicle
(6.7 mL/kg) by gavage. Immediately following the final treatment, the mice were housed with
untreated male mice for mating. At approximately 20-21 days after mating, 3 female mice from
each dose group, including the control group gave birth to 10-12 offspring. One day after birth,
each litter was culled to 10 offspring to accommodate similar lactational TCDD exposure. On
PND 28, the offspring were weaned, and three offspring from each TCDD exposed group and
the control group were selected for an immunocytochemical examination at 42 days of age.
Following sacrifice of these offspring, the brain of each animal was removed and every second
serial section of the brain was processed for immunocytochemistry. In addition to the serial
sections of the brain, cells from 18 offspring (6 males per treatment group) were used to assess
the number of cells in the dorsal and median raphe nucleus, the supralemniscal area, and the
Nucleus raphe magnus.
Examination of external morphology, birth, and postnatal body weights indicated that
there were no differences between the male TCDD-exposed offspring and the control male
offspring. TCDD-exposed males, however, were aggressive toward other normal mice and were
also hypersensitive to soft touch.
Serotonin-immunoreactive neurons were found to be distributed throughout the entire
brainstem in 42-day-old males, and the general pattern in the TCDD-exposed animals was
consistent with those observed in control male offspring. Serotonergic neurons were identified
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-148 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
and counted in the caudal linear nucleus, the median and dorsal raphe nucleus, Nucleus raphe
pontis, interpeduncular nucleus, supralemniscal area, pedunculopontine segmental nuclei, deep
mensencephalic nucleus, Nucleus raphe magnus, pallidus, and obscurus, dorsal and medial to the
facial nucleus and the ventrolateral medulla. Results from computerized cell counts (n = 6)
showed an average of 1,573.3 immunoreactive neurons in the raphe nuclei from the control
group versus 716.3 and 419.8 neurons in the low- and high-dose offspring, respectively. The
numbers of immunoreactive neurons in the individual raphe nuclei (dorsalis, medianus, magnus,
and B9) from the TCDD-exposed offspring were significantly (p < 0.01) lower than control
values, with the degree of reduction being dose-related.
In the absence of other relevant neurotoxicity endpoints, reduced serotonin is not an
adverse endpoint of toxicological significance in and of itself, thus, neither a NOAEL nor a
LOAEL can be established for this study. A lowest-observed-effect level (LOEL) of
0.7 ng/kg-day for an 8-week exposure duration is identified in this study for a significantly
(p < 0.01) lower number of serotonin-immunoreactive neurons in the raphe nuclei of male
offspring. A no-observed-effect level (NOEL) cannot be determined for this study.
2.4.2.2.8. Li et ah, 2006.
Pregnant and pseudopregnant (obtained by mating normal estrous female mice with
vasectomized male mice) NIH mice (10 per treatment group) were exposed to 0, 2, 50, or
100 ng/kg-day of TCDD (purity 99%) during early gestation (GDs 1-8), preimplantation
(GDs 1-3), or peri-implantation to postimplantation (GDs 4-8) (Li et al., 2006). On GD 9,
animals were evaluated. The two highest TCDD doses (50 and 100 ng/kg-day) caused
significant (p < 0.05) early embryo loss independent of gestational exposure time. At
100 ng/kg-day, however, the embryo loss was greater when administered during GDs 1-8 or
GDs 1-3 compared to GDs 4-8 (p < 0.01). Uterine weight was significantly decreased in the
pseudopregnant mice when administered 50 or 100 ng/kg-day TCDD during GDs 1-8
(p < 0.001) or 1-3 (p < 0.01), but was only decreased at 100- ng/kg-day in pseudopregnant mice
when administered during GDs 4-8 (p < 0.01). Estradiol levels were increased at all TCDD
treatment levels (100% at the lowest dose), but statistical significance was not indicated. All
doses at all treatment times resulted in a significant reduction (p < 0.01) in serum progesterone
levels, with a 45% decrease at the lowest dose. Because the hormone effects were observed
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-149 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
following 4 days of treatment, the nominal doses were averaged over the entire test period of
8 days prior to measurement. The resulting average daily doses of TCDD were 0, 1, 25, and
50 ng/kg-day.
A LOAEL of 2 ng/kg-day administered for 4 to 8 days is established in this study for a
significant (p < 0.01) decrease in progesterone (45% above control) and an approximate 2-fold
increase in estradiol levels (significance not indicated). A NOAEL cannot be determined.
2.4.2.2.9. Markowski et al., 2001.
Pregnant Holtzman rats (4-7 per treatment group) were administered a single gavage
dose of 0, 20, 60, or 180 ng/kg TCDD (purity not specified) in olive oil on GD 18 (Markowski et
al., 2001). One female rat from each liter (4-7 per treatment group) was assigned to training on
a wheel apparatus to respond on a lever for brief opportunities to run. Once animals responded
to an FR1 schedule of reinforcement, the requirement for lever pressing was increased to FR2,
FR5, FR10, FR20, and FR30 schedules. After each training session, the estrous cycle stage was
determined. Maternal body weight, length of gestation, number of pups per litter, and sex
distribution within litters were unaffected by treatment. For each of the FR schedules, there was
a significant dose-related (p = 0.0001) decrease in the number of earned run opportunities, lever
response rate, and total number of revolutions in the wheel in the adult female offspring. There
was no correlation between estrous cycle and responding for access to wheel running.
The developmental LOAEL for this study is a single dose of 20 ng/kg administered on
GD 18 for neurobehavioral effects. A NOAEL cannot be determined for this study.
2.4.2.2.10. Miettinen et al., 2006
Miettinen et al. (2006) administered a single oral dose of 0, 30, 100, 300, or 1,000 ng/kg
TCDD (purity >99%) in corn oil on GD 15 to pregnant Line C rats. The offspring (24-32 per
treatment group) were assigned to a sugar-rich cariogenic diet (via feed and drinking water) and
were orally inoculated three separate times with fresh cultures of Streptococcus mutans. Three
control groups varied with regard to TCDD exposure and administration of a cariogenic diet.
Two of the control groups received no TCDD, and the offspring were either maintained on a
normal diet without inoculation with S. mutans (CI;n = 48) or were given the cariogenic diet
with S. mutans inoculation (C2; n = 42). The final control group was maternally exposed to
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-150 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
1,000 ng/kg TCDD with offspring fed a normal diet without S. mutans inoculation (C3; n = 12).
TCDD did not affect the maternal or offspring body weight. Survival of the offspring was
reduced in the 1,000 ng/kg dose group (50-58% survival compared to 83-95% in CI and C2,
respectively). All offspring administered 1,000 ng/kg were missing all lower third molars.
Two animals (8%) in the 100 ng/kg group were missing one of their lower third molars. All
doses, except the 100 ng/kg dose, caused a significant (p < 0.05) increase in the number of caries
lesions compared to group C2 (60, 79, 76, 83, and 91% in the C2, 30, 100, 300, and 1,000 ng/kg
groups, respectively). Group C3 (1,000 ng/kg TCDD exposure, normal diet) animals also had
increased caries lesions compared to CI (8% versus 0%, respectively). There were no changes
in tooth mineral composition that could explain the increase in caries susceptibility.
The developmental LOAEL from this study is a single dose of 30 ng/kg administered on
GD 15 based on the significant (p < 0.05) increase in dental caries in pups (30% above control).
A NOAEL cannot be determined from this study.
2.4.2.2.11. Nohara et al, 2000.
Pregnant Holtzman rats were administered 0, 12.5, 50, 200, or 800 ng/kg TCDD in corn
oil by gavage on GD 15 (Nohara et al., 2000). On PND 2, five males were randomly selected
from each litter and dose group. TCDD was detected in the thymus, spleen, and bone marrow of
the male pups on PND 21 and PND 49. TCDD was still detected in the thymus and spleen on
PND 120 but the levels decreased over time. The TCDD concentration was highest in the
thymus at all time points. There were no changes in the body, thymus, or spleen weights of the
male offspring on PND 5, PND 21, PND 49, or PND 120. On PND 5, there was a 200-fold
increase in CYP1 Al in the thymus of the high-dose male pups. CYP1 Al was only slightly
increased in the spleen. This induction decreased through PND 49. There was a slight (not
statistically significant) dose-dependent decrease in thymus cellularity in the male offspring at
PND 120. Spleen cellularity at PND 49 decreased in a dose-dependent manner (15—50% of the
control), with a statistically significant (p < 0.05) decrease observed in the high-dose group. A
slight but not significant reduction in spleen cellularity was noted in the high-dose group at
PND 21. The same effect was not observed at PND 120, nor was there any change in the percent
of B or T cells in the spleen. No changes in cytokine levels were observed in the 800-ng/kg
group.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-151 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Although a change in spleen cellularity on PND 49 (puberty) was observed, this effect
was transient and there were no coexisting changes in the percentage of splenic lymphocytes,
spleen weight, and cytokine levels. Therefore, a developmental NOAEL of a single dose of
800 ng/kg administered on GD 15 is identified for this study. A LOAEL is not established.
2.4.2.2.12. Ohsako et al, 2001.
Pregnant Holtzman rats (6 per treatment group) were administered 0, 12.5, 50, 200, or
800 ng/kg TCDD (purity >99.5%) in corn oil by gavage on GD 15 (Ohsako et al., 2001). On
PND 2, five males were randomly selected from each litter. Two male offspring from each litter
were sacrificed on PND 49 and PND 120. Neither maternal nor male offspring body weight was
affected by TCDD treatment. TCDD was detected in both fat and testes at all dose levels
(including controls) with highest levels found in fat. There were no apparent treatment-related
effects on testicular weight, epididymal weight, daily sperm production, cauda epididymal sperm
reserves, luteinizing hormone, follicle stimulating hormone, or testosterone levels. There was,
however, a clear dose-dependent decrease in urogenital complex weight and ventral prostate
weight at both PND 49 and PND 120. For male offspring, statistically-significant (p < 0.05)
decreases were noted in urogenital complex weight at PND 120 in the 200 and 800 ng/kg groups,
in ventral prostate weight at PND 49 in 800 ng/kg group, and at PND 120 in the 200 and
800 ng/kg groups. There was also a dose-dependent decrease in anogenital distance (the length
between the base of the genital tubercle and the anterior edge of the anus); the decrease was not
statistically significant at PND 49. At PND 120, however, male offspring in all but the lowest
dose group had significantly (p < 0.05) reduced anogenital distance compared to the control
animals. There was also a dose-dependent increase in 5aR-II mRNA expression in the ventral
prostate on PND 49 with significant increases (p < 0.05) in the 200 and 800 ng/kg animals.
There was a significant (p < 0.01) decrease in the androgen receptor mRNA in the ventral
prostate on PND 49 at all doses tested. Similar effects were not observed on PND 120 or in the
caput epididymis on PND 49.
The developmental LOAEL for this study is a single dose of 50 ng/kg administered on
GD 15 for significantly (p < 0.01) reduced anogenital distance in male offspring (approximately
14%). The NOAEL for this study is 12.5 ng/kg.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-152 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.2.2.13. Schantz et ah, 1996.
Schantz et al. (1996) studied the impact of in utero TCDD exposure on spatial learning in
male and female pups. Groups of pregnant Harlan Sprague-Dawley rats (n = 108, divided into
4 cohorts; number of animals in each TCDD group approximately 4 per treatment group) were
dosed via gavage with 0, 25, or 100 ng/kg-day TCDD (purity >98%) in corn oil on GDs 10-16.
On the day of birth (post natal day [PND] 0), the pups were examined for gross abnormalities
and the number of live pups, weight, and sex were recorded for each litter. On PND 2, litters
were culled to eight animals and were balanced to include four males and four females whenever
possible. To minimize litter-size effects, litters with fewer than five pups were excluded from
the study. The exclusion of these litters resulted in 10-11 litters per treatment group. Pups were
weaned on PND 21 and one male and one female pup from each litter were maintained for the
learning tests. Pups were tested 5 days per week for spatial learning and memory in a radial arm
maze and a T-maze. A radial arm maze working memory test and a T-maze DSA task were used
a part of the testing process.
TCDD treatment did not affect dam gestational weight gain, dam liver weight, gestation
length, litter size, percentage of live births, birth weight, or postnatal growth of the pups
observed during the course of the study. Exposed pups, however, exhibited some signs of
toxicity in all exposure groups. Thymus weight was decreased and liver weight was increased in
the 100 ng/kg-day TCDD dose group. Also, liver microsomal 7-ethoxyresorufin-O-deethylase
(EROD) activity was markedly induced in pups from both the 25 and 100 ng/kg-day dose
groups. In the radial maze test, rats from all TCDD exposure groups displayed a significant
(p < 0.01) learning behavior as shown by progressively fewer errors from the first block of
sessions through the fourth session. The treatment by sex and treatment by session block
interactions were not significant. Comparisons between the average number of errors per session
block in the TCDD-exposed and control group indicated that both the 25 and the 100 ng/kg-day
dose groups made significantly (p < 0.05 andp < 0.001, respectively) fewer errors compared to
the control group. TCDD did not significantly affect adjacent arm selection behavior as
measured by C statistic; hence the reduction in errors observed did not appear to be accounted
for by an increased tendency to run into adjacent arms. Female pups had a significant (p < 0.05)
shorter radial arm maze latency, however, compared to the male pups. In the T-maze test,
TCDD did not significantly affect the percent of correct performance. All exposure groups
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-153 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
performed best at the shortest delay, which showed a decline as the length of the inter-trial delay
interval was increased. Additionally, all treated groups improved their performance over a
three-block session period. This finding indicated that animals in all groups could learn the task.
These observations were confirmed by a highly significant main effect of delay (p < 0.001) and
highly significant main effect of session blocks (p < 0.001). At the shortest 15-second delay,
average percent correct performance increased from 75 to 92%, while at the longest 40-second
delay, the average percent correct performance increased from 62 to 82%. A significant
(p < 0.05) main effect of exposure was evident in latency to respond in the T-maze.
Comparisons of the exposed group to control group, however, indicated that none of the
individual exposure groups differed significantly from the controls. Because no clear pattern
was observed in the various exposure groups, differences in latency to respond had no impact on
learning of the task.
Based on these results, the study authors state that the fact TCDD seems to have a
facilitatory effect on radial arm maze learning in rats should be interpreted with caution and
needs further evaluation using different and more varied learning tasks. No toxicologically
adverse endpoints were concurrently examined. Thus, a LOAEL and a NOAEL cannot be
determined for this study.
2.4.2.2.14. Seo et al., 1995.
To study developmental effects of TCDD on thyroid hormone levels, time-mated female
Sprague-Dawley rat dams (// = 10-14/treatment group) were administered 25 or 100 ng/kg-day
of TCDD (>98%) pure) in corn oil via gavage from GDs 10-16. Vehicle controls received
equivalent amounts of corn oil. The study also investigated PCB treatment outcomes. At birth,
pups were weighed and grossly examined for abnormalities. At 2 days of age, litters with fewer
than 5 pups were excluded from the analysis and the remaining litters were culled to 4 males and
4 females. Each treatment group contained 10 or 11 litters. Pups remained with the dams until
weaning. At weaning, 4-6 pups were retained for neurobehavioral tests (which were not
reported as part of this study). The remaining offspring were sacrificed, which provided
5-9 litters per treatment group. Data were collected from one male and one female where
possible. No signs of toxicity were evident in the dams; measurements on dams included
gestational weight gain, liver weight, litter size, and live births. Pup birth weight and weaning
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-154 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
weight were unaffected by treatment. In pups sacrificed at weaning (21 days old), a significant
(p < 0.05) decrease occurred in thymus weight for the high-dose group, but not in thyroid, liver,
or brain weight. A significant (p < 0.05) decrease (20.4%) was observed in T4 in high-dose
females. Thyroid stimulating hormone and T3 were unaffected by treatment. Uridine
diphosphate (UDP)-glucuronosyl transferase activity towards 4-nitrophenol significantly
(p < 0.05) increased in both treatment groups over control values, and the increase in the
high-dose group was significantly (p < 0.05) greater than in the low-dose group. Liver
microsomal EROD activity was significantly (p < 0.05) increased in both treatment groups, but
is considered to be an adaptive response and not adverse.
A LOAEL of 100 ng/kg-day for decreased thymus weights and decreased thyroxine is
identified for this study. A NOAEL of 25 ng/kg-day is established.
2.4.2.2.15. Simanainett et al, 2004.
Simanainen et al. (2004) studied the impact of in utero and lactational TCDD exposure
on the male reproductive system in three rat lines that are differentially sensitive to TCDD.
Groups of 5 to 8 pregnant Line A, B, and C C57BL/6N CYP1A2 dams were given a single dose
of 0, 30, 100, 300, or 1,000 ng/kg of TCDD (purity >99%) in corn oil on GD 15 via oral gavage.
Control animals were similarly dosed with a corn oil vehicle. One day after birth, litters were
randomly culled to include three males and three females to allow uniform postnatal exposure.
Offspring were weaned on PND 28. Dam and pup viabilities were monitored throughout the
study. Pup body weights were determined on PNDs 1, 4, 7, 14, and 28. Anogenital distance and
crown-rump length were measured on PNDs 1 and 4. On day 70, pups were sacrificed and trunk
blood was collected. Serum was collected for testosterone analysis. The testes, cauda of the
right epididymis, ventral prostrate, seminal vesicles, and thymus was dissected and weighed.
Absolute and relative organ weights were determined, and cauda epididymis and testes were also
preserved for sperm count analysis.
TCDD caused no mortality or overt signs of toxicity to the dams. Pup survival from
implantation to the day after birth also was not affected by TCDD exposure. Survival from the
day of implantation to the day after birth, however, was uncharacteristically lower in control
Line B rats (41%), resulting in a significant difference compared with the two lowest doses (30
and 100 ng/mg TCDD). The average survival percentage in the controls for Line A, B, and C
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-155 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
rats was 85% (range 80-86%); 64% (41—86%); and 74% (63—85%); respectively. Percentage of
male pup survival in each line between PND 1 and PND 28 was 99% except for Line B males
exposed to 30 ng/kg TCDD and Line C males exposed to 30 or 100 ng/kg, where male survival
rate averaged 81% (range 81—83%). On PND 70, a significant (p < 0.05) reduction in body
weight was observed only in Line B and C rats at 1,000 ng/kg. In pups exposed to 1,000 ng/kg
TCDD, both absolute and relative weight of the ventral, anterior, and dorsolateral prostrate
decreased in all three lines at most postnatal time points measured. The change was most
consistent and significant (p < 0.05) in the ventral lobe. Animals exposed to 1,000 ng/kg TCDD
had an average decrease in absolute weight of the anterior prostrate of 37, 32, and 34% in
Lines A, B and C, respectively. Additionally, the average dorsolateral prostrate weight was also
decreased by 34, 28, and 39% in Lines A, B, and C, respectively. The effect on the ventral
prostrate was reversible with the only significant (p < 0.05) decrease in weight observed in
Line B rats at PND 70 in the 1,000 ng/kg TCDD dose group. The authors reported that TCDD
had no consistent effects on the weight of seminal vesicles. The absolute weights of the testis
and epididymis showed a significant (p < 0.05) increase on PNDs 28-49, but the relative testis,
epididymis, and cauda epididymis weights remained unchanged. In pups exposed to 1,000 ng/kg
TCDD, severe malformation, including small caput and cauda and degeneration of corpus
epididymis, was observed. Malformations in the epididymis were observed in 6 of 44 Line C
male rat offspring and 3 of 47 Line A male rat offspring. In Line A, B, and C rats at PND 70in
the 1,000 ng/kg TCDD dose group, daily sperm production was reduced by 9, 25, and 36% and
cauda epididymal sperm reserves were reduced by 18, 42, and 49%, respectively. Daily sperm
reduction (17%) was significant (p < 0.05) in Line C rats at a TCDD dose of 300 ng/kg and in
Line B and C rats at 1,000 ng/kg. A reduction in cauda epididymal sperm reserves (25%) was
significant (p < 0.05) in Line C rats at 300 and 1,000 ng/kg TCDD.
A LOAEL for TCDD of 300 ng/kg is identified for reduction in daily sperm production
and cauda epididymal sperm reserves in Line C rats. A NOAEL of 100 ng/kg is identified for
this study.
2.4.2.2.16. Sugita-Konishi et al., 2003.
Sugita-Konishi et al. (2003) examined the immunotoxic effects of lactational exposure to
TCDD in newborn mice. Eight pregnant female C57BL/6NCji mice were administered 0, 1.8, or
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-156 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
18 ng/L of TCDD via drinking water from parturition to weaning of the offspring (for a total of
17 days). Based on an average water intake of 14-16 mL/day, the average daily intake of TCDD
for the dams was 1.14 and 11.3 ng/kg-day in the low- and high-dose groups, respectively. In
male offspring sacrificed at weaning (21 days after birth), there was a statistically-significant
(p < 0.05) decrease in relative spleen weight and a statistically-significant (p < 0.005) increase in
thymic CD4+ cells in the high-dose group. The changes in relative spleen weight and thymic
CD4+ cells were dose related, but effects in the low-dose group did not achieve statistical
significance. Changes in spleen weight and CD4+ cell numbers were not observed in the female
offspring. In a separate experiment, offspring infected with Listeria monocytogenes following
lactational TCDD exposure exhibited a statistically significant increase in serum tumor necrosis
factor alpha (TNF-a) 2 days after infection in both sexes in the low- (p < 0.05) and high-dose
(p < 0.005) groups. There was also a statistically significant increase in serum interferon gamma
in Listeria-infected high-dose females (p < 0.05). The number of bacteria in the spleen was also
significantly increased (p < 0.05) 2 days after infection in the high-dose females compared to the
controls, but not in males. Listeria levels in the spleen returned to control levels by 4 days after
infection in both sexes.
Based on these results, a LOAEL for TCDD of 11.3 ng/kg-day following a 17 day
exposure to dams was identified for significantly (p < 0.05) decreased spleen weight (in male
pups), a significant (p < 0.005) increase in thymic CD4+ cells (in male pups), and for increased
susceptibility to Listeria monocytogenes (in male and female pups). The NOAEL for this study
is 1.14 ng/kg-day.
2.4.2.3. Acute Studies
2.4.2.3.1. Burleson et al., 1996.
Burleson et al. (1996) studied the impact of TCDD exposure on mice that were
challenged with the influenza virus 7 days after treatment with TCDD. Groups of 8-week-old
female B6C3F1 mice (n = 20, 2 replicate groups) were treated one time with 0, 1,5, 10, 50, 100,
or 6,000 ng/kg TCDD (purity >99%, dissolved in corn oil) via oral gavage. In addition to the
treated groups, randomly selected animals were assigned as a sentinel group and screened for
numerous pathogens. Results of all tests performed on this sentinel group were negative. Seven
days after TCDD treatment, all animals were lightly anesthetized and infected intranasally with a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-157 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
highly lethal influenza A/Hong Kong/8/68 virus (H3N1; passage 14). The animals were infected
with sufficient H3N1 virus to achieve a 30% mortality rate in the control animals. Animals were
observed for mortality and morbidity for 21 days following viral infection. Six mice from each
treatment group were sacrificed on days 3, 9, and 12 post-infection, and body, thymus, and wet
lung weights were recorded. Influenza viral titers were examined by sacrificing eight mice each
at 2 hours and at 1, 4, 6, 7, 8, 9, 10, and 11 days post infection.
Exposure to TCDD resulted in significantly (p < 0.05) increased mortality in the 10, 50,
and 100 ng/kg dose groups. No statistically significant difference in the percentage alive was
observed between these dose groups. TCDD doses of 1 and 5 ng/kg did not alter mortality in
influenza infected animals. A time-related increase in the wet weights of the lungs in infected
mice as a result of increased edema also was reflected in an increase in the lung weight-to-body
weight ratio. The study authors stated that this ratio was not altered as a result of TCDD
exposure. TCDD-only exposures at 1, 10, or 100 ng/kg did not affect thymus weight. Similarly,
animals infected with the influenza virus following TCDD exposure also showed no loss in
thymic weight. Enhanced mortality in TCDD-treated animals was not correlated with an
increase in influenza virus titers. Additionally, animals treated with 1, 10, 100, or 1,000 ng/kg
did not affect pulmonary viral titer assays on days 6, 7, and 8 post-infection. The authors also
concluded that TCDD did not alter Hong Kong virus replication or clearance.
Although these results support immunotoxic effects induced by TCDD, the findings were
not reproduced by Nohara et al. (2002) using the identical study design, and the translation of
these findings to humans is dubious. Thus, no LOAEL/NOAEL was established. A LOEL for
TCDD of 10 ng/kg for a single exposure is identified for significantly (p < 0.05) increased
mortality in mice infected 7 days later with the influenza virus. The NOEL for this study is
5 ng/kg.
2.4.2.3.2. Crofton et al., 2005.
Crofton et al. (2005) studied the impact of TCDD exposure in addition to the impact of
mixtures of thyroid disrupting chemicals and PCBs on serum total thyroxine (TT4)
concentration. Groups of female Long-Evans rats were dosed via oral gavage with 0, 0.1, 3, 10,
30, 100, 300, 1,000, 3,000, or 10,000 ng/kg-day TCDD (purity >99%) in corn oil (n = 14, 6, 12,
6, 6, 6, 6, 6, 6, and 4, respectively) for 4 consecutive days. On the day following the last dose,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-158 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
animals were sacrificed, trunk blood was collected, and serum obtained via centrifugation was
assayed for TT4 concentration using standard radioimmunoassay methods.
No visible signs of toxicity or changes in animal body weight as a result of TCDD
exposure were observed. Serum T4 levels showed a dose-dependent decrease, with the levels
dropping sharply beginning at 100 ng/kg-day dose. Percent serum T4 levels were 96.3, 98.6,
99.8, 93.3, 70.9, 62.5, 52.7, 54.7, and 49.1% in the 0.1, 3, 10, 30, 100, 300, 1,000, 3,000, and
10,000 ng /kg-day groups, respectively.
A LOAEL for TCDD of 100 ng/kg-day for 4 consecutive days of exposure is identified in
this study for a reduction in serum T4 levels (70.9% compared to 100% in controls). The
NOAEL for this study is 30 ng/kg-day.
2.4.2.3.3. Kitchin and Woods, 1979.
Female Sprague-Dawley rats (nine per control and four per treatment group) were
administered a single dose of 0, 0.6, 2, 4, 20, 60, 200, 600, 2,000, 5,000, or 20,000 ng/kg TCDD
(purity >99%) in corn oil. Animals were sacrificed 3 days after treatment and CYP level and
benzo(a)pyrene hydroxylase activity in the liver were measured. A significant (p < 0.05)
increase in cytochrome P450 levels occurred with doses of 600 ng/kg or greater and in
benzo(a)pyrene hydroxylase activity with doses of 2 ng/kg or greater. Cytochrome P450 was
significantly (p < 0.05) higher 1 month after a single exposure of 2,000 ng/kg (the only dose
measured), but not after 3 or 6 months. Aryl hydrocarbon hydralase (AHH; p < 0.05) and EROD
(p < 0.01) were both significantly increased through 3 months after treatment, and although
elevated at 6 months, the results were not significant.
CYP induction alone is not considered a significant toxicologically adverse effect given
that CYPs are induced as a means of hepatic processing of xenobiotic agents. Thus, no LOAEL
or NOAEL was established for this study because adverse endpoints (e.g., indicators of
hepatotoxicity) were not measured. The acute LOEL, however, is 2 ng/kg based on a significant
(p < 0.05) increase in benzo(a)pyrene hydroxylase activity (37% above control). The NOEL is
0.6 ng/kg.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-159 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.2.3.4. Li et al., 1997.
Female Sprague-Dawley rats (22 days old; 10 per treatment) were administered a single
oral dose of TCDD (>98% pure) in corn oil via gavage at doses of 3, 10, 30, 100, 300, 1,000,
3,000, 10,000, or 30,000 ng/kg. Vehicle controls received equivalent amounts of corn oil, while
naive controls were sham-treated only. In a preliminary time-course study, animals received a
single dose of 10,000 ng/kg and were sacrificed at 1, 2, 4, 8, 16, 24, 48, and 72 hours. The
time-course study showed two peaks in LH and FSH levels at 1 hour and 24 hours, with a
decrease to control values by 48 hours. Thus, in the dose-response study, animals were
sacrificed at 1 or 24 hours after treatment, blood was collected, and serum FSH and LH were
measured. The dose-response study demonstrated that the peak at 1 hour was related to the
vehicle as the peak also occurred in the vehicle controls, but did not occur in the naive controls.
At 24 hours, FSH was increased at 10 ng/kg and higher (>4-fold increase at 10 ng/kg). Doses of
10 to 1,000 ng/kg showed similar increases (not all reached statistical significance; p < 0.05). A
dose-dependent increase occurred for doses >3000 (p < 0.05) with a maximum increase of
20-fold over the vehicle control. At 24 hours, the LH response significantly (p < 0.05) increased
only for doses >300 ng/kg with a maximum increase of 15-fold above the vehicle control. The
study authors calculated an ED50 of 500 ng/kg for gonadotropin increase. The dose-dependent
release of LH was confirmed in in vitro studies, but did not occur with the same magnitude. The
increase did not occur in calcium-free medium and was unrelated to gonadotropin releasing
hormone.
Based on the increase in serum FSH, the LOAEL was 10 ng/kg and the NOAEL was
3 ng/kg.
2.4.2.3.5. Lucier et al., 1986.
Adult female Sprague-Dawley rats (six per treatment) were administered a single gavage
dose of TCDD (purity not specified) in either corn oil or contaminated soil at doses of 15, 40,
100, 200, 500, 1,000, 2,000, 5,000 (corn oil), or 5,500 (contaminated soil) ng/kg. Animals were
sacrificed 6 days later and livers were removed for analysis. No clinical signs of acute toxicity
or changes in body weight were observed at any dose. AHH increased in a dose-dependent
manner with significant (p < 0.05) increases observed at 15 ng/kg or greater in corn oil or
40 ng/kg or greater in contaminated soil. Cytochrome P450 was significantly (p < 0.05)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-160 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
increased with doses of 1,000 ng/kg or greater in corn oil or 500 ng/kg or greater in contaminated
soil. A dose-dependent increase was observed for UDP glucoronyltransferase (significance of
individual doses not reported), with the results twice as high with corn oil than with
contaminated soil. The authors state that the results indicate bioavailability from soils is 50%.
Because the association between AHH activity and TCDD-mediated hepatotoxicity is
unknown and no adverse endpoints were measured, a LOAEL or NOAEL was not determined
for this study. The acute LOEL for this study is 15 ng/kg, based on the significant (p < 0.05)
increase (80% above control) in AHH. No NOEL is established.
2.4.2.3.6. Nohara et ill., 2002.
Male and female B6C3F1 (C57BL/6 x C3H), BALB/c, C57BL/6N, and DBA2 mice
(10-40 per treatment group) were administered a single dose of 0, 5, 20, 100, or 500 ng/kg
TCDD in corn oil via gavage. Seven days following TCDD treatment, mice were infected with a
mouse-adapted strain of influenza (A/PR/34/8; H1N1) at a plaque forming unit dose designed to
target approximately 30% mortality in each strain. TCDD did not affect the body weight or
survival in any of the infected mouse strains at any dose.
Therefore, no LOAEL is established in this study. The NOAEL is 500 ng/kg.
2.4.2.3.7. Simanainett et al., 2003.
Simanainen et al. (2003) studied the short-term effects of TCDD exposure to determine
the efficacy and potency relationships among three differentially susceptible rat lines. The three
rat lines used were A, B, and C, which were selectively bred from TCDD-resistant Han/Wistar
and TCDD-sensitive Long-Evans rats. The study authors reported that Line A rats were most
resistant to TCDD acute lethality followed by Line B and C. Groups of five or six randomly
selected rats (sex not specified) were treated with a single oral dose of TCDD (purity >99%) in
corn oil by oral gavage. The dose of TCDD was reported to range between 30 ng/kg and
3,000 ng/kg for Line A, 30 ng/kg and 1,000 |ig/kg in Line B, and 30 ng/kg and 100 |ig/kg for
Line C. Control animals were similarly dosed with a corn oil vehicle. Rats were sacrificed on
day 8 post-exposure, and trunk blood was collected and serum separated. Liver and thymus were
removed and weighed, and liver samples were collected and preserved. Liver EROD activity,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-161 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
serum aspartate aminotransferase (ASAT) activity, free fatty acid (FFA) concentration, and total
bilirubin concentration were determined. Teeth were also examined.
Relative thymus weights were reduced 25% at 300 ng/kg relative to controls in Line B
rats. Liver enzyme (CYP1A1) induction, as measured by EROD activity, was evident at all
exposure levels; CYP induction is considered to be an adaptive effect and not adverse in itself.
No other endpoints were affected below 1 |ig/kg in any of the three rat lines.
A LOAEL for TCDD of 300 ng/kg is identified for decreased relative thymus weight in
Line B rats. A NOAEL of 100 ng/kg is identified for this study.
2.4.2.3.8. Simanainett et al., 2002.
To study the short-term effects of TCDD on hormone levels, adult female Long-Evans
(TCDD-sensitive) and Han/Wistar (TCDD-resistant) rats (// = 9-11/treatment) were administered
a single dose of TCDD (>99% pure) in corn oil via gavage at doses ranging from 30 ng/kg to
100 |ig/kg. Vehicle controls received an equivalent amount of corn oil. The study also
examined other polychlorinated dibenzo-p-dioxins outcomes. Rats were sacrificed on day 8
post-exposure, and trunk blood was collected and serum separated. Liver and thymus were
removed and weighed, and liver samples were collected and preserved. Liver EROD activity,
serum ASAT activity, FFA concentration, and total bilirubin concentration were determined.
Teeth were also examined.
Neither FFA or ASAT levels in Han/Wistar rats showed a dose-response relationship. In
Long-Evans rats, however, a significant (p < 0.05) dose-dependent increase in FFA occurred at
300 ng/kg TCDD. Serum ASAT sharply increased in Long-Evans rats between 3,000 and
10,000 ng/kg. Body weight change and relative thymus weights were significantly decreased
(p < 0.05) in Han/Wistar rats with doses >10,000 ng/kg and in Long-Evans rats with doses
>1,000 ng/kg. Liver EROD activity was significantly (p < 0.05) increased with all doses in both
strains. Serum T4 was significantly (p < 0.05) decreased in Long-Evans rats at concentrations
>300 ng/kg, but were not significantly affected in Han/Wistar rats. Serum bilirubin was
significantly (p < 0.05) increased with doses >10,000 ng/kg in Long-Evans rats and
>30,000 ng/kg in Hans/Wistar rats. Both strains of rat showed a dose-dependent increase in
mean severity of incisor tooth defects. The results indicate that TCDD was the most potent
congener tested in both rat strains.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-162 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
A LOAEL of 300 ng/kg for decreased T4 in the Long-Evans rat is identified for this
study. A NOAEL of 100 ng/kg is is established.
2.4.2.3.9. Smialowicz et al., 2004.
Smialowicz et al. (2004) examined the impact of TCDD exposure on immunosuppression
in mice. Groups of female (number not specified) C57BL/6N CYP1A2 (+/+) wild-type mice
were administered a single dose of 0, 30, 100, 300, 1,000, 3,000, or 10,000 ng/kg TCDD (purity
>99%) in corn oil via oral gavage. Control animals were similarly dosed with a corn oil vehicle.
To assess immune function, 7 days after TCDD administration, all mice were immunized with
sheep red blood cells (SRBCs) via injection into the lateral tail vein. Five days after
immunization, mice were sacrificed, blood was collected, and enzyme-linked immunosorbant
assays were performed. Additionally, spleen, thymus, and liver weights also were measured.
Body and spleen weights of the wild-type mice were unaffected by the TCDD exposure.
A decrease in thymus weights of the mice appeared to be dose related. Only mice treated with
10,000 ng/kg TCDD, however, showed a statistically significant (p < 0.05) decrease in thymus
weights compared to corresponding controls. Liver weights also showed a dose-related increase
with only animals treated with 3,000 and 10,000 ng/kg TCDD showing statistical significance
(p < 0.05) compared to the control group. The antibody response to SRBCs indicated a
dose-related suppression in the wild-type mice, with animals treated with 1,000, 3,000, and
10,000 ng/kg TCDD showing statistically significant (p < 0.05) suppression compared to the
controls.
A LOAEL for TCDD of 1,000 ng/kg is identified in female C57BL/6N CYP1A2 (+/+)
wild-type mice for significant (p < 0.05) suppression of SRBCs. The NOAEL for this study is
300 ng/kg.
2.4.2.3.10. Vanden Heuvel et al., 1994.
Vanden Heuvel et al. (1994) examined the dose-response relationship between TCDD
exposure and induction of hepatic mRNA. Groups of 10-week-old female Sprague-Dawley rats
were administered TCDD (purity -99%) in corn oil once at 0, 0.1, 0.05, 1, 10, 100, 1,000, or
10,000 ng/kg-BW. Four days after TCDD treatment, animals were sacrificed and livers were
excised and preserved. Total hepatic RNA was extracted using guanidine thiocyanate and DNA
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-163 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
was removed using standard phenol-chloroform-isoamyl alcohol partitioning procedures.
Quantitative competitive RNA-PCR method was used to analyze CYP1 Al,
UDP-glucuronosyltransferase I (UGT1), plasminogen activator inhibitor 2 (PAI2), P-actin, and
transforming growth factor a (TGFa). In addition to hepatic mRNA levels, microsomal protein
was assayed for EROD activity and livers were tested for TCDD concentration.
CYP1 Al mRNA induction levels in the TCDD-treated groups were low in the low-dose
region and sharply increased to plateaus at higher doses. The lowest dose that showed a
statistically significant (p < 0.05) difference compared to controls was the 1 ng/kg dose, which
showed a three-fold increase in CYP1 Al mRNA levels. In contrast, a 130-fold increase
occurred at 100 ng/kg and a 4,000- and 7,000-fold increase occurred at 1,000 and 10,000 ng/kg,
respectively. A slight increase in the CYP1 Al/p-actin levels was observed in the 0.1 ng/kg
group, but this increase was not significant. EROD activity exhibited a pattern similar to
CYP1A1 activity. EROD activity, however, was approximately 100-fold less sensitive
compared to mRNA levels in TCDD-treated groups. Statistical significance (p-value not
provided) in CYP1A1 level was observed at the 100 ng/kg dose compared to the 1 ng/kg dose.
The study authors reported that, despite this difference in CYP1 Al and EROD activity, the
correlation between CYP1 Al enzyme activity and mRNA levels was good. Dose-response
relationships for the induction of UGT1, PAI2, and TGFa mRNA differed from what had been
observed for CYP1 Al mRNA. UGT1 mRNA was induced, but at the much higher dose of
1,000 ng/kg. Additionally, the five-fold maximum induction of UGT1 mRNA was much less
than the 7,000-fold induction observed for CYP1 Al mRNA at the 10,000 ng /kg dose. The
authors state that this could be a result of the constitutive level of UGT1, which is much higher
than CYP1A1, which makes detecting induction of UGT1 in the low dose regions more difficult.
PAI2 and TGFa mRNA were not affected by TCDD in rat liver in the dose range tested. These
results indicate that dioxin-inducible genes have a quite dissimilar dose-response relationship.
Induction of CYP1A1 expression is not considered an adverse effect, as the role of
CYP1A1 in TCDD-mediated hepatotoxicity is unsettled. Therefore, in the absence of other
indicators of hepatotoxicity, a NOAEL/LOAEL cannot be determined for this study. A LOEL
for TCDD of 1 ng/kg for a single exposure was identified for statistically significant (p < 0.05)
increase in CYP1 Al mRNA levels. The NOEL for this study is 0.1 ng/kg.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-164 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.2.4. Subchronic Studies
2.4.2.4.1. Chu et al., 2001.
Adult female Sprague-Dawley rats (five per treatment group) were administered TCDD
(purity >99%) in corn oil by gavage at doses of 0, 2.5, 25, 250, or 1,000 ng/kg-day for 28 days
(Chu et al., 2001). The 1,000 ng/kg-day dose of TCDD caused a significant (p < 0.05) decrease
in body weight gain (36% lower than the control), increase in relative liver weight (40% greater
than the control), and decrease in relative thymus weight (50% lower than the control). There
was a significant (p < 0.05) increase in EROD activity, methoxy resoufin-O-deethylase (MROD)
activity, and UDP-glucuronosyl transferase (UDPGT) activity in the liver of female rats
receiving 250 or 1,000 ng/kg-day TCDD. In addition, significant (p < 0.05) increases in serum
cholesterol were observed in the 250 and 1,000 ng/kg-day dose groups, and liver ascorbic acid
(AA) also was significantly increased in the 1,000 ng/kg-day dose group. There was ~1.5-fold
increase in liver glutathione-»Y-transferase (GST), which was not statistically significant. Other
significant (p < 0.05) findings for the 1,000 ng/kg-day group included a decrease in liver
vitamin A (51% lower than the control), an increase in kidney vitamin A (15.5-fold increase
above the control), an increase in liver benzyloxy resoufin-O-deethylase (BROD, 30-fold
increase above control), a decrease in liver pentoxyresoufin-O-deethylase (PROD, 37% lower
than the control), increase in serum albumin (18% above the control), and a decrease in mean
corpuscular hemoglobin (MCH, 7% below the control) and mean corpuscular volume (MCV, 7%
below the control).
Based on the numerous significant (p < 0.05) liver-related biochemical changes and
significant (p < 0.05) increased relative liver weight, as well as significantly decreased body
weight and relative thymus weight, the LOAEL for 28 days of exposure in this study is
1,000 ng/kg-day and the NOAEL is 250 ng/kg-day.
2.4.2.4.2. Chu et al., 2007.
Chu et al. (2007) examined the potential impact of TCDD on various organs and the
toxicological impacts as a result of interactions between TCDD and PCBs in rats. Groups of
female Sprague-Dawley rats (n = 5 per treatment group) were treated daily for 28 days via
gavage with 0, 2.5, 25, 250, or 1,000 ng /kg-day TCDD (purity not specified) dissolved in corn
oil. Body weights were determined three times per week, and clinical observations were made
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-165 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
daily. At study termination, all animals were sacrificed and blood was analyzed for various
biochemical and hematological parameters. Liver, spleen, heart, thymus, brain, and kidneys
were removed and weighed. A small portion of the liver was homogenized and assayed for
BROD; EROD; MROD; and PROD. UDPGT, GST, and ascorbic acid levels also were
measured. Vitamin A levels in the liver, kidney, and lungs were analyzed as free retinol
(vitamin A), and histopathological analysis was conducted on various tissues.
Growth rate and thymic weights in rats treated with 1,000 ng/kg-day TCDD were
significantly (p < 0.05) inhibited compared to the control group. Enzyme analysis indicated that
measured levels of TCDD in the liver correlated with hepatic microsomal enzyme activity. The
authors reported that liver microsomal EROD and MROD activities were significantly (p < 0.05
for EROD activity, significance level for MROD not reported) increased in the 250 and
1,000 ng/kg-day TCDD dose groups compared to the control group. UDPGT levels were
significantly (significance level not reported) increased in the 250 and 1,000 ng/kg-day TCDD
dose groups compared to the controls. Serum albumin levels were significantly (p < 0.05)
increased in the 1,000 ng/kg-day TCDD dose group compared to the control group. Serum
cholesterol levels were significantly (level not reported) increased compared to the control group
at 250 ng/kg-day TCDD dose, while liver ascorbic acid concentrations were significantly (level
not reported) increased in the 1,000 ng/kg-day dose group. Hematological analysis indicated that
hemoglobin, packed cell volume, MCH, MCV, and platelet values were decreased in the
1,000 ng/kg-day TCDD dose group. Significant (p < 0.05) differences were observed only in
MCH and MCV levels compared to the control. Vitamin A levels in the liver and kidney were
significantly (p < 0.05) lower in the 1,000 ng/kg-day TCDD group compared to the control
group. Histopathological evaluation of various tissues indicated that liver, thyroid, and thymus
were the target organs. No TCDD-related affects were found in other tissues. A dose-dependent
alteration in the thymus consisted of reduced thymic cortex and increased medullar volume with
more animals exhibiting these changes at the 250 and 1,000 ng/kg-day dose level compared to
the control group. Alterations in thyroid included reduced follicles, reduced colloid density, and
increased epithelial height. A dose-dependent change in the thyroid was observed, with the
highest impact evident in reduced follicles and reduced colloid density beginning at a dose of
25 ng/kg-day TCDD. Changes in liver were characterized by accentuated hepatic zones,
anisokaryosis of hepatocytes, increased cytoplasmic density, and vacuolation. These changes
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-166 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
were also dose dependent, with more animals exhibiting these histopathological changes with
increasing TCDD dose. Based on these results, the study authors concluded that exposure to
TCDD resulted in a wide range of adverse effects with the thyroid proving to be most sensitive.
A LOAEL for TCDD of 25 ng/kg for a 28-day exposure is identified for alterations in
thyroid, thymus, and liver histopathology. The NOAEL for this study is 2.5 ng/kg-day.
2.4.2.4.3. DeCaprio et al., 1986.
Hartley guinea pigs (10 per sex per dose) were administered TCDD (purity not specified)
in the diet for 90 days at concentrations of 0, 2, 10, 76, or 430 ppt (equivalent to 0, 0.12, 0.61,
4.9, and 26 ng/kg-day in males and 0, 0.12, 0.68, 4.86, and 31 ng/kg-day in females calculated by
the study authors using food consumption and body weights). Other animals were administered
the high-dose diet (i.e., 430 ppt) for 11, 21, or 35 days and then administered the control diet
(i.e., no exposure) for the remainder of the 90 days for recovery analysis. Four high-dose males
died and two were sacrificed moribund by day 45; the remaining four animals were sacrificed on
day 46 for necropsy. Four high-dose females also died and two were sacrificed moribund by day
55 with the remaining females sacrificed on day 60 for necropsy. Animals in the 76- and
430-ppt groups had significantly (p < 0.05) reduced body weights. Organ weights were not
obtained in the 430-ppt group due to the early sacrifice, but in the 76-ppt group a significant
decrease in relative thymus weight (p < 0.05) was observed, and relative liver (p < 0.01) and
brain (p < 0.05) weights in males increased. Although a similar trend occurred in the females,
the results were not statistically significant. Males administered 76 ppt in the diet also had a
53% increase in triglycerides (p < 0.05). The same increase was observed in females, but was
not statistically significant. In the recovery groups, mortality during the recovery period after 11
or 21 days of treatment was 10% and after 35 days of treatment was 70%. Animals lost weight
during the treatment period. Although the body weight increased during the recovery period, the
body weight remained low compared to the control for the study duration.
The LOAEL from this study is 4.9 ng/kg-day for 90 days of exposure, based on
decreased body weight (12-15 %;p< 0 .05) and changes in organ weights (10-30%), significant
only in the males). The NOAEL is 0.61 ng/kg-day.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-167 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
2.4.2.4.4. Devito et at, 1994.
Female B6C3F1 mice (5 per treatment) were administered 0, 1.5, 4.5, 15, 45, or
150 ng/kg TCDD (98% pure) in corn oil via gavage, 5 days a week for 13 weeks. This dose is
equivalent to 0, 1.07, 3.21, 10.7, 32.1, 107 ng/kg-day (adjusted for continuous exposure,
administered dose multiplied by 5 and divided by 7). Body weight was recorded weekly and
animals were sacrificed 3 days after the last treatment. Examinations were performed on the
lung, skin, uterus, and liver. No differences were observed in the liver or uterus weights or in the
estrogen receptor levels in these two tissues. A dose-dependent increase in EROD activity (an
indicator of CYP1A1 [CYP] induction) in the lung, skin, and liver was observed, with significant
(p < 0.05) increases even at the lowest dose. The TCDD doses used did not achieve maximal
EROD induction. A significant (p < 0.05) increase in liver acetanilide-4-hydroxylase (ACOH;
an indicator of CYP1A2 induction) also was observed with all doses. A maximum induction of
ACOH occurred with doses of 3.21 ng/kg-day and greater. A dose-dependent increase in
specific phosphotyrosyl protein (pp) levels also was observed. Levels of pp34 and pp38 were
significantly (p < 0.05) increased even at the lowest dose, while pp32 reached statistical
significance (p < 0.05) with doses of 4.5 ng/kg-day and above.
The role of CYPs and phosphorylated pp32, pp34, and pp38 in TCDD-mediated toxicity
is unknown, and changes in the activity or function of these proteins are not considered adverse
Therefore, no LOAEL or NOAEL is established. The 13-week LOEL is 1.07 ng/kg-day, based
on a significant (p < 0.05) increase in EROD, ACOH, pp34, and pp38 levels (all increased by at
least 2-fold). No NOEL is established for this study.
2.4.2.4.5. Fattore et at, 2000.
Fattore et al. (2000) examined TCDD-induced reduction of hepatic vitamin A levels in a
subchronic rat bioassay on Sprague-Dawley rats. Four experiments were conducted;
Experiments 1, 2, and 3 were conducted in both male and female rats, while Experiment 4 was
conducted only in female rats. The dosing regimens for each experiment were as follows
Experiment 1: Groups of six Iva:SIV 50 rats (male and female) were maintained on a diet
consisting of 0, 200, 2,000, or 20,000 ng TCDD/kg diet and 3-[j,g vitamin A/kg diet for
13 weeks. Assuming food consumption of 10% of body weight per day, the average daily
doses are 0, 20, 200, and 2,000 ng/kg-day TCDD.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-168 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Experiment 2\ Groups of six male and female rats were treated with 0 or 200 ng
TCDD/kg-day and 3 jag vitamin A/kg diet for 13 weeks.
Experiment 3. Groups of six male and female rats were fed 0, 200, or 1,000 ng
TCDD/kg-day and 3 jag vitamin A/kg diet for 13 weeks.
Experiment 4. Groups of female rats (number not specified; IVA;SIV 50 Sprague-Dawley
strain) were treated with TCDD for 26 and 39 weeks in addition to a 13-week dietary
treatment with 0 or 100 ng TCDD/kg-day and 3 jag vitamin A/kg diet for 13 weeks.
For a 13-week exposure duration employed in all four experiments, male and female rats
were treated at 0, 20, 100 (females only), 200, 1,000, or 2,000 ng/kg-day. In all four
experiments, liver from control and treated animals was analyzed at termination for free
retinol content to determine hepatic vitamin A levels.
Results'.
Experiment 1: Liver and body weights in both treated males and females were significantly
affected at all but the lowest dose tested (20 ng/kg-day). Liver injury was severe, particularly
in female rats treated with 2,000 ng TCDD/kg-day. Dietary intake of vitamin A in male rats
was comparable to intake in controls, except in the 2,000 ng/kg-day group, which showed a
reduction of 16% in the dietary intake of vitamin A compared to controls. There was no
effect of TCDD on vitamin A intake in female rats. Hepatic vitamin A levels showed a
dose-dependent reduction with levels dropping sharply in the 200 and 2,000 ng/kg-day dose
groups, particularly in treated females. The reduction was significant at 200 ng/kg-day
(p < 0.05) and 2,000 ng/kg-day (p < 0.01) in males, and at 200 ng/kg-day (p < 0.5) and
2,000 ng/kg-day (p < 0.001) in females. The reductions ranged from 68-99% in males and
72-99% in females when compared to corresponding controls.
Experiment 2\ Changes in liver and body weights were not reported. Hepatic vitamin A
level in males and females were reduced by 70% and 99%, respectively, compared to
controls, in rats receiving 20 ng/kg-day (significance level in females: p < 0.01).
Experiment 3. Similar to the results of Experiments 1 and 2, a dose-related trend of
significantly (p < 0.001) reduced hepatic vitamin A level was observed in both males and
females, with males exhibiting a particularly sharp drop at the 1,000 ng/kg-day dose
compared to controls.
Experiment 4. Females treated with 100 ng/kg-day showed significant reductions in hepatic
vitamin A levels (p < 0.05-0.001) at all three treatment durations (13, 26, and 39 weeks).
A LOAEL for TCDD of 20 ng/kg-day for a 13-week subchronic exposure was identified
in this study for decreased hepatic vitamin A levels (27 and 24 % lower than the corresponding
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-169 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
control in female and male rats, respectively). This LOAEL is determined using data from
Experiment 1. A NOAEL was not identified in this study.
2.4.2.4.6. Fox et al., 1993.
Sprague-Dawley rats (6 per sex per dose) were gavaged with TCDD (purity not
specified) in corn oil using a dose-loading regime to achieve and maintain steady-state levels of
0.03, 30, or 150 ng/g in the liver. The regime consisted of an initial loading dose of 5, 2,500, or
12,000 ng/kg followed every 4 days with a maintenance dose of 0.9, 600, or 3,500 ng/kg.
Averaging the doses over the 14 days provides average daily doses of 0.55, 307, and
1,607 ng/kg-day (e.g., 5 ng/kg-day on day 1 and 0.9 ng/kg-day on days 5, 9, and 13 is 5 + 0.9
+ 0.9 + 0.9/14 = 0.55 ng/kg-day). Body weight, liver weight, and liver gene expression were
measured at 7 and 14 days. A significant (p < 0.05) decrease in body weight occurred in
high-dose males (at 14 weeks only) and females (at 7 and 14 days). A significant (p < 0.05)
increase in absolute and relative liver weights was observed in mid- and high-dose males and
females at both 7 and 14 days. Although the liver of treated animals indicated moderate
vacuolization and swelling, there was no indication of necrosis. An increase in gene expression
(clone 1, CYP1A1, CYP1A2, and albumin) was observed in the mid- and high-dose groups. A
significant (p < 0.05) decrease in labeling index (indication of cell proliferation) occurred in both
females (all doses) and males (high-dose only) during week 1, but not during week 2.
The 14-day LOAEL is 307 ng/kg-day for significant (p < 0.05) increases in absolute and
relative liver weights (25—34%). The NOAEL is 0.55 ng/kg-day.
2.4.2.4.7. Hassoun et al., 1998.
Female B6C3F1 mice (number not specified) received TCDD (>98% pure) in corn oil
5 days per week for 13 weeks via gavage at doses of 0, 0.45, 1.5, 15, or 150 ng/kg (equivalent to
0, 0.321, 1.07, 10.7, and 107 ng/kg-day adjusted for continuous exposure; administered dose
multiplied by 5 and divided by 7). Three days after the final dose, animals were sacrificed and
brains were removed for oxidative stress testing. Biomarkers for oxidative stress included
production of superoxide anion, lipid peroxidation, and DNA single-strand breaks. A significant
(p < 0.05) increase was observed in superoxide anion production, lipid peroxidation as measured
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-170 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
by thiobarbituric acid-reactive substances (TBARS), and DNA single-strand breaks with all
doses tested.
No other indicators of brain pathology were assessed, and it is unfeasible to link the
markers of oxidative stress to a TCDD-induced toxicological outcome in the brain. Thus, no
LOAEL/NOAEL was established. The subchronic (13-week) LOEL is 0.32 ng/kg-day, based on
significant (p < 0.05) increases in superoxide anion production (80% above control); lipid
peroxide production (25% above the control); and DNA single-strand breaks (2-fold over the
control). No NOEL is established.
2.4.2.4.8. Hassoun et al., 2000.
Hassoun et al. (2000) examined the effect of subchronic TCDD exposure on oxidative
stress in hepatic and brain tissues. Groups of 8-week-old female Harlan Sprague-Dawley rats
(6 rats/group) were administered TCDD (98% purity, dissolved in 1% acetone in corn oil) via
gavage at 0, 3, 10, 22, 46, or 100 ng/kg-day, 5 days/week for 13 weeks (0, 2.14, 7.14, 15.7, 32.9,
or 71.4 ng/kg-day adjusted for continuous exposure; administered doses were multiplied by 5
and divided by 7 days/week). Animals were sacrificed at the end of the study period, and brain
and liver tissues were collected and used to determine the production of reactive oxygen species,
lipid peroxidation, and DNA single-strand breaks (SSBs).
A dose-dependent effect was observed in both the liver and brain tissue as a result of
TCDD treatment. Based on the maximal induction of superoxide anion by various doses, more
production of superoxide anion was observed in the liver tissue when compared to the brain
tissue with an observed increase of 3.1- and 2.2-fold respectively, when compared to the control
group. A similar dose-dependent effect was observed in the induction of lipid peroxidation in
TCDD-treated animals with an approximately 1.8-fold increase in lipid peroxidation in both
tissues relative to the corresponding controls. A dose-dependent relationship was also observed
for DNA SSBs in both the hepatic and brain tissues at all TCDD-treated doses compared to
controls. Increases were statistically significant (p < 0.05) beginning at the lowest administered
dose.
Similar to the statement above, because no adverse endpoints were measured, no
LOAEL/NOAEL was established. However, a LOEL for TCDD of 2.14 ng/kg-day for a
13-week exposure duration was identified in this study for significant increases (p < 0.05) in
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-171 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
superoxide anion, lipid peroxidation, and DNA SSBs in the liver and brain tissues. A NOEL
cannot be determined for this study.
2.4.2.4.9. Hassoun et al, 2003.
Hassoun et al. (2003) examined the role of antioxidant enzymes in TCDD-induced
oxidative stress in various regions of the rat brain after subchronic exposure. Groups of
8-week-old female Harlan Sprague-Dawley rats (12 rats/group) were administered TCDD (98%
purity, dissolved in 1% acetone in corn oil) via gavage at 0, 10, 22, or 46 ng/kg-day (0, 7.14,
15.7, or 32.9 ng/kg-day adjusted for continuous exposure; administered doses were multiplied by
5 and divided by 7) daily for 13 weeks. Animals were sacrificed at the end of the study period
and the brain was immediately removed and dissected to the following regions: cerebral cortex
(Cc), hippocampus (H), cerebellum (C), and brain stem including midbrain, pons, and medulla.
Four pooled samples from each region per dose (i.e., 3 animals/pooled sample) were used in the
study. Dissected regions were subsequently assayed for lipid peroxidation (thiobarbituric acid
reactive substances, or TBARS), superoxide dismutase, catalase, and glutathione peroxidase.
Because the cytochrome c reduction method was used to determine superoxide anion (SA)
production in brain tissues, superoxide dismutase (SOD) was added to some of the brain tissue
samples that had the highest SA production (tissue homogenates from Cc and H from rats treated
with 46 ng/kg-day TCDD).
A dose-dependent increase in the production of SA was observed in the Cc and H, but
significant changes in SA production were not observed in either the C or the mid-brain, pons, or
medulla brain stem cells. Similar to SA production, there was a dose-dependent increase in the
production of TBARS in the Cc and H regions of the brain, but no significant changes were
observed in either the C or the B sections of the brain. The study authors also measured the
activities of various enzymes as a result of TCDD treatment and reported a dose-dependent
increase in SOD activity in the C and B sections, while there was dose-dependent suppression in
SOD activity in Cc and H. In contrast, catalase activity was significantly (p < 0.05) increased in
H and Cc at the 10 ng/kg-day TCDD dose level compared to controls and the mid- and high-dose
animals. Catalase activity also was increased in a dose-dependent manner in the C section, but
no significant changes in the activity of this enzyme were observed in the B section at any of the
three TCDD tested doses. The effects of subchronic exposure to different doses of TCDD on
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-172 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
glutathione stimulating hormone peroxidase (GSH-Px) showed a different response compared to
other enzymes. There was a dose-dependent increase in the activity of this enzyme in the C and
B regions of the brain, while a significant increase in the activity of GSH-Px occurred in Cc and
H only at the 10 ng/kg-day TCDD dose. In addition, the activity of this enzyme was suppressed
in a dose-dependent manner in the Cc and H at 22 and 46 ng/kg-day TCDD doses. Based on
these results, the study authors concluded that induction of oxidative stress by TCDD in the rat
brain occurs mainly in the Cc and H regions.
Similar to the statement above, because no adverse endpoints were measured, no
LOAEL/NOAEL was established. However, a LOEL for TCDD of 7.14 ng/kg-day for a
13-week exposure duration was identified for this study for increases in superoxide anion and
lipid peroxidation production, as well as increased activity in SOD, catalase, and GSH-Px.
2.4.2.4.10. Kociba et al., 1976.
Adult Sprague-Dawley rats (12 per sex per treatment group) were administered TCDD
(purity not reported) in corn oil via gavage 5 days per week at doses of 0, 1, 10, 100, or
1,000 ng/kg-day (equivalent to 0, 0.71, 7.14, 71.4, or 714 ng/kg-day averaged over 7 days; 5/7 of
dose). Five animals per group were sacrificed at the end of treatment, and the remaining animals
were observed over 13 weeks post treatment (only initial results for the post-treatment period
were provided in the report). Body weights and food consumption were measured semi-weekly.
Hematology and clinical chemistry were measured after 36-37 or 85-86 days of treatment and
59-60 days after termination of treatment. Forty-eight hour urine samples were collected from
select rats from 85-89 days of treatment and 52-56 days after cessation of treatment. Gross and
histopathological exams were conducted on the tissues.
Four high-dose females died during treatment. Two high-dose females and two
high-dose males died during the post-treatment period. Animals treated with 714 ng/kg-day
were less active during the treatment period, which became less evident during the
post-treatment period. Yellow discoloration of the external pinnae also was noted in this group,
both during treatment and during the post-treatment period. A significant (p < 0.05) reduction in
body weight and food consumption was observed in the 71.4 and 714 ng/kg-day groups. The
following significant (p < 0.05) hematology changes were observed in the high-dose
(714 ng/kg-day) males at all measured time points: decreased packed cell volume, decreased red
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-173 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
blood cells, decreased hemoglobin, increased reticulocytes, and decreased thrombocytes.
Significant (p < 0.05) changes also occurred in the high-dose females, but the only consistent
observation was a decrease in thrombocytes and increased leukocytes. Significant changes in
clinical chemistry (p < 0.05) and urinalysis (p < 0.05) were more consistent between the sexes in
the high-dose group and included increases in total and direct serum bilirubin; increase in serum
alkaline phosphatase; decreased urinary creatinine; and increased urinary coproporphyrin,
uroporphyrin, and delta-amino-levulinic. The following significant (p < 0.05) changes were
observed in the 71.4 ng/kg-day group: decreased packed cell volume (4-9%) in males; decreased
red blood cells (2-10%) in males; decreased hemoglobin (2—13%) in males; increased urinary
coproporyphin (2.2-fold increase during treatment) in females; increased urinary
delta-amino-levulinic (47% increase during treatment) in females; increased total and direct
serum bilirubin (48-61%) in females; and increased serum alkaline phosphatase (2-fold) in
females. The following significant (p < 0.05) changes in relative organ weights were observed
increased brain weight in 714 ng/kg-day males and females; increased liver weight in males
(71.4 and 714 ng/kg-day) and females (7.14, 71.4, and 714 ng/kg-day); increased spleen weight
in 714-ng/kg-day males and females; decreased thymus weight in 71.4 and 714 ng/kg males and
females; and increased testes weight in 714 ng/kg-day males. Microscopic changes were
observed in the thymus, and in other lymphoid tissues, and in the liver in rats treated with
71.4 ng/kg-day or greater.
The subchronic (13-week) LOAEL is 71.4 ng/kg-day, based on the numerous changes
noted in body weight, hematology, clinical chemistry, urinalysis, and histopathology. The
NOAEL is 7.14 ng/kg-day.
2.4.2.4.11. Matty and Chipman, 2002.
Female F344 rats (3 per treatment group) were administered TCDD at concentrations of
0, 2.5, 25, or 250 ng/kg in corn oil via gavage for either 3 consecutive days or 2 days per week
for 28 days (Mally and Chipman, 2002). The average daily doses for the 28-day study when
adjusted for 7 days a week were 0, 0.71, 7.1, and 71 ng/kg-day (i.e., 2/7 of administered dose).
No clinical signs of toxicity were observed. Histological examination of the liver revealed no
abnormalities. All doses of TCDD reduced the number of connexin (Cx) 32 plaques and Cx32
plaque area in the liver, which was considered the target tissue. The reductions were not
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-174 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
statistically significant after the 3-day treatment, but were significant after the 28-day treatment
(p < 0.05). TCDD also caused a reduction in the Cx32 plaque number and area in the thyroid
after 28 days, but the results were not statistically significant. Although the reduction in Cx32
plaque number and plaque area in the liver and thyroid occurred at all dose levels, there was no
relation to dose. TCDD did not induce hepatocyte proliferation.
In the absence of additional indicators of hepatotoxicity, changes in Cx32 plaques are not
clearly linked to TCDD-mediated hepatotoxicity, nor are they considered an adverse effect.
Additionally, no toxicologically-relevant endpoints were examined. Therefore, a NOAEL or
LOAEL cannot be determined. A 28-day LOEL at the lowest dose of 0.71 ng/kg-day for
significantly (p < 0.05) decreased Cx32 plaque area is evident (approximately 70% of the
controls).
2.4.2.4.12. Slezak et at, 2000.
Slezak et al. (2000) studied the impact of subchronic TCDD exposure on oxidative stress
in various organs of B6C3F1 female mice. Groups of 8- to 10-week-old female B6C3F1 mice
(number not specified) were administered TCDD (purity >98%, dissolved in corn oil) via gavage
at 0, 0.15, 0.45, 1.5, 15, or 150 ng/kg-day (0, 0.11, 0.32, 1.07, 10.7, or 107.14 ng/kg-day adjusted
for continuous exposure) 5 days per week for 13 weeks. Three days after the last treatment, the
animals were sacrificed and organs were removed for the measurement of oxidative stress
indicators including SA, lipid peroxidation (TBARS), and GSH-Px. Tissue TCDD
concentrations also were measured.
The study authors reported that TCDD dose range resulted in overlapping tissue
concentrations for liver, lung, kidney and spleen. Liver had the highest TCDD concentration,
with each tissue demonstrating a dose-dependent increase in TCDD concentration. Compared to
controls, SA production was significantly (p < 0.05) lower at the 0.15 ng/kg-day TCDD dose,
while it was significantly (p < 0.05) higher at 15 and 150 ng/kg-day. A dose-dependent increase
in hepatic TBARS production was observed, although the rate of production was significant
(p < 0.05) only at the highest TCDD administered dose (150 ng/kg-day) compared to controls.
AA also followed the same pattern observed for SA and TBARS with AA production
significantly (p < 0.05) increased at the 15 and 150 ng/kg-day TCDD doses. Contrary to the SA,
TBARS, and AA responses, GSH levels were decreased at 0.15 ng/kg-day, were increased at
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-175 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
0.45 and 150 ng/kg-day, and did not change at 1.5 or 15 ng/kg-day when compared to the control
group. Unlike the liver, there was no significant increase in SA production in the lung at any of
the TCDD tested doses; a dose dependent reduction, however, was observed at 0.45, 15, and
150 ng/kg-day compared to controls. GSH and AA production was decreased at 0.15 ng/kg-day,
while AA production was significantly (p < 0.05) increased at 15 and 150 ng/kg-day. Kidney
SA production showed a statistically significant (p < 0.05) increase only at the 15 and
150 ng/kg-day doses. GSH, like the liver and the lung, exhibited a decrease in production
following treatment at 0.15 ng/kg-day with this trend continuing at 0.45 and 1.5 ng/kg-day. AA
levels were significantly (p < 0.05) lower at all subchronic doses, except at 1.5 ng/kg-day dose.
SA levels in the spleen differed little from the control group at any of the TCDD doses. Total
GSH was higher only at the 150 ng/kg-day dose level, while the AA levels were significantly
(p < 0.05) decreased at 0.15, 1.5, and 150 ng/kg-day.
Similar to the statements regarding the Hassoun et al. studies above, because no adverse
endpoints were measured, no LOAEL/NOAEL was established. Therefore, a NOAEL or
LOAEL cannot be determined. However, a NOEL and LOEL of 1.07 and 10.7 ng/kg-day,
respectively, are identified in this study for increases in superoxide anion in the liver.
2.4.2.4.13. Smialowicz et al., 2008.
Female B6C3F1 mice (8-15 per treatment group) were administered TCDD (purity
>98%) in corn oil by gavage at doses of 0, 1.5, 15, 150, or 450 ng/kg-day, 5 days a week for
13 weeks (1.07, 10.7, 107, or 321 ng/kg-day, adjusted for continuous exposure; i.e., 5/7 of the
dose) (Smialowicz et al., 2008). Mice were immunized 3 days after the final TCDD exposure
with an intravenous injection of an optimal concentration of 4 x 107 SRBCs and sacrificed 4 days
later. No TCDD-related effects on body weight were observed. There was a dose-related
decrease in relative spleen weight (9-19% lower than control values) with statistically significant
(p < 0.05) decreases at all but the lowest dose. Additionally, there was a statistically significant
(p < 0.05) increase in relative liver weight (5—21%) in all treatment groups compared to controls.
Statistically significant dose-dependent decreases were observed in the antibody response to
SRBCs (24-89%) lower than control values), as measured by both the number of plaque forming
cells per 106 cells and plaque forming cells per spleen.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-176 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
The 13-week LOAEL for this study is 1.07 ng/kg-day based on a significant (p < 0.05)
increase in relative liver weight (10%) and a significant (p < 0.05) decrease in antibody response
to SRBCs (24%). A NOAEL cannot be determined for this study.
2.4.2.4.14. Van Birgelen et al 1995a, b.
Van Birgelen et al. (1995) studied the impact of TCDD exposure on various biochemical
endpoints in rats. Groups of 7-week-old female Sprague-Dawley rats (n = 8 per treatment group)
were treated with 0, 200, 400, 700, 5,000, or 20,000 ng/kg TCDD (purity >99%) in diet for
13 weeks. Daily TCDD intake based on food consumption, diet level, and mean weight was
estimated to be 0, 14, 26, 47, 320, or 1,024 ng/kg-day. Blood samples were collected from
treated animals and assayed for retinol (vitamin A), triiodothyronine, and total (TT4) and free
(FT4) thyroxine. At study termination, the animals were sacrificed and the liver, thymus, spleen,
and kidneys were removed and weighed. Parts of the liver were homogenized and assayed to
determine EROD; CYP1A1; CYP1A2; and UDPGT activity. Liver samples also were analyzed
for retinol content.
TCDD-treated animals showed a dose-related decrease in food consumption. Animals
treated with 1,024 ng/kg-day TCDD consumed 32% less food compared to controls. Similarly, a
dose-related decrease in body weight gain was observed in all animals treated with TCDD.
Animals treated with >47 ng/kg-day of TCDD showed a statistically significant (p < 0.05)
decrease in body weight gain. Relative liver weights were significantly (p < 0.05) increased in
the 320 and 1,024 ng/kg-day TCDD dose groups compared to the controls. Absolute and relative
thymus weights were significantly (p < 0.05) decreased at all TCDD dose groups compared to
the control group. Relative kidney and spleen weights were significantly (p < 0.05) higher in
animals dosed with >47 ng/kg-day of TCDD compared to the control group, with the greatest
increase occurring in animals treated with 1,024 ng/kg-day TCDD (121 and 173%) higher than
controls for kidney and spleen, respectively). Cytochrome P450 enzymes, including EROD,
CYP1A2, CYP1A1, and UDPGT, exhibited statistically significant (p < 0.05) increases in
activity at all TCDD dose groups compared to the control group. TT4 and FT4 thyroid hormone
concentrations were statistically significantly (p < 0.05) decreased only at TCDD doses
>47 ng/kg-day. A dose-dependent increase was observed in the plasma retinol concentrations
with significant (p < 0.05) increases occurring at >47 ng/kg-day TCDD after a 13-week
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-177 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
exposure. A dose-dependent reduction in liver retinoid levels also was observed after 13 weeks
of TCDD exposure with the levels dropping significantly (p < 0.05) at all TCDD-treated doses
compared to the control group.
A LOAEL for TCDD of 14 ng/kg for a 13-week exposure is identified for significantly
(p < 0.05) decreased absolute and relative thymus weights and significantly (p < 0.05) decreased
liver retinoid levels. A NOAEL cannot be determined for this study.
2.4.2.4.15. Vos et al., 1973.
Vos et al. (1973) conducted a study to examine the immune response in laboratory
animals treated with TCDD. In one experiment, 10 female Hartley strain guinea pigs were orally
treated with 8 weekly doses of 0, 8, 40, 200, and 1,000 ng/kg TCDD in corn oil (purity of TCDD
not specified) (0, 1.14, 5.71, 28.6, and 143 ng/kg-day adjusted for continuous exposure;
administered dose divided by 7). At study termination, the animals were sacrificed, and heart
blood was used to determine total leukocyte and differential leukocyte counts. In another
experiment, the effect of TCDD on humoral immunity was determined by injecting 0.1 mL of
tetanus toxoid into the right hind-foot pad on day 28 (1 left foot tetanus toxoid, aluminum
phosphate-adsorbed) and again on day 42 (1 left foot tetanus toxoid, unadsorbed). Blood was
collected (n = 10) on days 35 and 49, and the serum tetanus-antitoxin concentrations were
determined using a modified single radial immunodiffusion technique.
All guinea pigs receiving 1,000 ng/kg-day TCDD either died or were killed when
moribund between 24 and 32 days. These animals showed severe weight loss, lymphopenia, and
depletion of the lymphoid organs, especially the thymus. Microscopic observations revealed
severe atrophy of the thymic cortex with substantial destruction of lymphocytes, with the nuclear
debris being engulfed by macrophages. Large cystic Hassall bodies, filled with
polymorphonuclear leukocytes were observed in the medulla. All animals treated with 0, 8, 40,
or 200 ng/kg-day TCDD survived until study termination. Body weight gain was significantly
(p < 0.01) lower in the 200 ng/kg-day group. Absolute thymus weight was significantly reduced
in the 40 and 200 ng/kg-day treatment groups (p< 0.01 andp < 0.05, respectively). In contrast,
relative thymus weight was significantly (p < 0.01) reduced only in the 200 ng/kg-day dose
group. The absolute weight of the superficial cervical lymph nodes was significantly (p < 0.05)
decreased in the 200 ng/kg-day group, while the relative adrenal weight was significantly
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-178 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
(p < 0.05) increased in the 200 ng/kg-day dose group. Total leukocyte count was significantly
(p < 0.05) decreased in the 40 ng/kg-day dose group and total lymphocyte count was
significantly decreased at 8, 40, and 200 ng/kg-day (p < 0.01 ,p < 0.05, andp < 0.05,
respectively). A significant (/;-values not provided) monotonic dose-response relationship was
determined for body weight (decrease), relative thymus weight (decrease), relative adrenal
weight (increase), and total leukocyte and lymphocyte count (decrease). Microscopic
examination of the lymphoid organs and adrenals showed no effects, while slight cortical atrophy
of the thymus was observed at the 200 ng/kg-day dose.
Animals receiving the tetanus toxoid injection showed a small but significant increase in
serum tetanus antitoxin concentrations at the 8 and 40 ng/kg-day dose (p < 0.05 and p < 0.01,
respectively). Measurement at days 49 and 56 indicated that serum antitoxin levels had
decreased sharply and the significant (p < 0.05 on day 49 andp < 0.01 on day 56) effect was
seen only at the 200 ng/kg-day dose level.
A LOAEL for TCDD of 5.71 ng/kg-day for an 8-week exposure is identified in this study
for significantly (p < 0.01) reduced absolute thymus weight, significantly (p < 0.05) reduced
leukocyte and lymphocyte count, and significantly (p < 0.01) increased serum tetanus antitoxin
concentration. The NOAEL for this study is 1.14 ng/kg-day.
2.4.2.4.16. White et al., 1986.
White et al. (1986) studied the impact of TCDD exposure on serum complement levels.
Groups of female (C57BL/6 x C3H)F1(B6C3F1) mice were treated for 14 consecutive days with
TCDD in corn oil (purity of TCDD not specified) at doses of 0, 10, 50, 100, 500, 1,000 or
2,000 ng/kg-day via gastric intubation (n = 6-8). At study termination, blood was collected from
anesthetized animals and assayed for serum complement activity and complement component C3
levels.
Serum complement activity between the 10 and 100 ng/kg-day doses was between 69 and
59% compared to the vehicle control group, with all treatment groups being significantly
(p < 0.05) low compared to the vehicle control. In contrast, C3 levels were comparable to the
vehicle control with levels ranging between 98 and 94% of the control group. The higher doses
of 500, 1,000, and 2,000 ng/kg-day, however, produced a marked decrease of the component
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-179 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
hemolytic activity (45, 35, and 19% of the vehicle control) and of C3 levels (91, 81, and 74 % of
the vehicle control, respectively; significance level at/? < 0.05).
A LOAEL for TCDD of 10 ng/kg-day for a 14-day exposure is identified in this study for
significantly (p < 0.05) lower serum complement activity. A NOAEL cannot be determined for
this study.
2.4.2.5. Chronic Studies (Noncancer Endpoints)
2.4.2.5.1. Cantoni et al., 1981.
CD-COBS rats (4 per treatment) were orally administered TCDD (purity not specified)
dissolved in acetone:corn oil (1:6) at doses of 0 (vehicle alone), 10, 100, or 1,000 ng/kg per week
(equivalent to 1.43, 14.3, and 143 ng/kg-day adjusted for continuous exposure, administered
dose by dividing the dose by 7) for 45 weeks. Urine was collected several times during
treatment and tested for porphyrin excretion. Twenty-four hours after the final dose, animals
were sacrificed and their livers, spleens, and kidneys were removed for analysis of total
porphyrins. All treatment groups had a significant (p < 0.05) increase in coproporphyrin
excretion beginning at 6, 3, or 2 months, respectively. Uroporphyrin excretion was significantly
(p < 0.05) increased in the 14.3 ng/kg-day group at 10 months and in the 143 ng/kg-day group
beginning at 6 months. The high-dose group also had a significant (p < 0.05) increase in
excretion of heptacarboxylic methyl ester beginning at 6 months. The high-dose group had a
marked porphyric state beginning at 8 months as indicated by a 70-fold increase above controls
in total urinary porphyrin excretion. This group also had a significant (p < 0.05) increase in total
porphyrins in the liver, kidneys, and spleen.
The 45-week LOAEL for this study is 1.43 ng/kg-day, based on a 2- to 3-fold increase in
urinary coproporphyrin excretion. No NOAEL was established for this study.
2.4.2.5.2. Croutch et al., 2005.
Croutch et al. (2005) examined the impact of TCDD exposure on body weight via
insulin-like growth factor (IGF) signaling. Female Sprague-Dawley rats were randomly assigned
in groups of five to initial loading doses of TCDD (purity >98.5%, dissolved in corn oil) at 0,
12.5, 50, 200, 800, or 3,200 ng/kg-day, followed by treatment with maintenance doses equivalent
to 10%) of the initial loading dose every third day to maintain a pharmacokinetic steady state
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-180 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
throughout the entire study (equivalent to: 14-day average = 0, 1.25, 5, 20, 80, or 320 ng/kg-day;
28-day average = 0, 0.85, 3.4, 13.6, 54.3, or 217 ng/kg-day; 63-day average = 0, 0.60, 2.4, 9.5,
38, or 152 ng/kg-day; and 128-day average dose = 0, 0.51, 2.0, 8.1, 32.5, or 130 ng/kg-day).
Following 2, 4, 8, 16, 32, 64, or 128 days of initial dosing, the animals were sacrificed, livers
were removed and weighed, and trunk blood was collected to analyze glucose content. Rat liver
phosphoenolpyruvate carboxykinase (PEPCK) mRNA and protein levels also were analyzed, and
PEPCK activity was measured.
Body weights of TCDD-treated animals decreased after the second week of the
3,200 ng/kg-day TCDD loading dose, with significant differences beginning at week 9. There
was also a statistically significant (p < 0.05) difference in body weights at weeks 10, 11, 13, 18,
and 19 at the highest loading dose (3,200 ng/kg-day). PEPCK activity in the liver was also
decreased in a dose-dependent manner following TCDD administration at approximately
16 days. PEPCK inhibition was statistically significant (p < 0.05) on day 4 in rats treated with
either 800 or 3,200 ng/kg-day TCDD when compared to animals treated with a loading dose of
200 ng/kg-day. A similar statistically significant change was observed in animals treated with
3,200 ng/kg-day on day 16 when compared to the 200 ng/kg-day treatment group. In contrast,
differences in PEPCK activity at other doses or time points were not statistically significant. In
TCDD-treated animals, there was also a dose-dependent decrease in PEPCK mRNA expression
along with a decrease in PEPCK protein levels in the liver. In addition to body weight and
PEPCK activity changes, animals treated with 3,200 ng/kg-day TCDD showed a sharp decline in
circulating IGF-I levels on day 8 compared to the control group (corn oil) and TCDD-treated
animals at lower doses. In the highest dose animals, IGF-I levels continued to decline to 42% of
the control group by day 16 of the study. The IGF-I levels at the highest dose plateaued at an
average decrease of 66% through day 128 when compared to controls. Beginning at day 8, the
decrease in IGF-I was statistically significant at every time point through day 128 compared to
the control group, as well as groups treated with either 12.5 or 50 ng/kg-day TCDD. Similar
statistically significant decreases also were observed for the 800 ng/kg-day TCDD-treated groups
with an initial decrease of 37% on day 16 followed by a further decline to approximately 45%
thereafter compared to controls and the 12.5, 50, and 200 ng/kg-day dose groups. In contrast to
these results, circulating levels of insulin and glucose were unaffected by TCDD treatment, while
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-181 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
the active or phosphorylated form of AMPK-a protein increased with dose as a result of TCDD
treatment.
A LOAEL for TCDD of 217 ng/kg-day for a 28-day exposure duration (because this
represented the most sensitive time for elicitation of effects) was identified in this study for
decreased body weight, significant (p < 0.05) inhibition of PEPCK activity, and reduced IGF-I
levels (42% lower than the control group). A NOAEL of 54.3 ng/kg-day was identified in this
study.
2.4.2.5.3. Hassoun et al., 2002.
Hassoun et al. (2002) examined the potential of TCDD and other dioxin-like chemicals to
induce oxidative stress in a chronic rat bioassay. Groups of six Harlan Sprague-Dawley female
rats were treated with 0, 3, 10, 22, 46, or 100 ng/kg-day TCDD (98% purity), 5 days a week via
gavage for 30 weeks. The administered doses adjusted for continuous exposure were 0, 2.14,
7.14, 15.7, 32.9, and 71.4 ng/kg-day, respectively (administered doses were multiplied by 5 and
divided by 7). At study termination, hepatic and brain tissues from all treated rats were divided
into two portions and examined for the production of reactive oxygen species and SSBs in DNA.
When compared to controls, there was a dose-dependent increase in the production of
superoxide anion in TCDD-treated animals ranging from 21-998%) and 66-257%) in hepatic and
brain tissues, respectively. Hepatic tissues had statistically significant (p < 0.05) increases in
superoxide anion production at doses >7.14 ng/kg-day, while the brain tissue had a statistically
significant (p < 0.05) increase over controls at all doses. Similarly, increases in lipid
peroxidation were observed in hepatic and brain tissues with a 481%> increase (p < 0.05) at
71.4 ng/kg-day in the hepatic tissue when compared to controls. The increase in lipid oxidation
in brain tissue ranged from 33-188%) (p < 0.05) in the 2.14-71.4 ng/kg-day dose groups. DNA
SSBs were also observed in both hepatic and brain tissue in all treated groups. When compared
to the control group, there was a dose-dependent statistically significant (p < 0.05) increase in
DNA SSBs ranging from 58-322%) and 29-137%) in hepatic and brain tissues, respectively.
Nonmonotonic dose-response relationships were observed for superoxide production and lipid
peroxidation in liver tissues, with greater-than-linear increases in effect between the two highest
dose levels.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-182 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
As stated above, because no adverse endpoints were measured, no LOAEL/NOAEL was
established. However, a LOEL for TCDD of 2.14 ng/kg-day for a 30-week exposure duration is
identified in this study for significant (p < 0.05) increases in superoxide anion, lipid peroxidation
production, and DNA SSBs in the liver and brain tissues. A NOEL cannot be determined for this
study.
2.4.2.5.4. Kociba et al., 1978.
Sprague-Dawley rats (50 per sex per treatment group) were administered TCDD (purity
>99%) in the diet at doses of 0, 1, 10, or 100 ng/kg-day for 2 years. Body weights and food
consumption were routinely measured. Hematology, clinical chemistry, and urinalysis were
measured after 3, 12, or 23 months of treatment. Animals were routinely palpitated for tumors.
Gross and histopathological exams were conducted on the tissues of dead or dying animals or at
terminal sacrifice. Specific organs also were weighed.
The high-dose females had a statistically significant (p < 0.05) increase in mortality
compared to the controls during the second half of the study. Mortality changes in males were
variable and of questionable toxicological significance. A significant (p < 0.05) reduction in
body weight occurred in the 100 ng/kg-day males and females beginning at 6 months. Mid-dose
females also had reduced body weight, but to a lesser degree during the same time frame. There
were no consistent changes in food consumption. The following significant (p < 0.05)
hematology changes were observed in the high-dose animals: decreased packed cell volume in
males after 3 months and in females after 1 year, decreased red blood cells in females after
1 year and in males at terminal sacrifice, decreased hemoglobin in males after 3 months and in
females after 1 year, and decreased total white blood cell count in females after 1 year. Changes
in clinical chemistry (p < 0.05) occurred only in high-dose females and consisted of an increase
in serum alkaline phosphatase and gamma glutamyl transferase. Significant changes in
urinalysis occurred only in females and included increased urinary coproporphyrin in the mid-
and high-dose groups, increased urinary uroporphyrin in the mid- and high-dose groups, and
increased urinary delta-amino-levulinic acid in the high-dose group. Significant (p < 0.05)
changes in relative organ weights were observed, including increased liver weight in mid- and
high-dose females and decreased thymus weight in high-dose females. Mid- and high-dose rats
showed hepatocellular degeneration and inflammatory and necrotic changes in the liver. Thymic
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-183 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
and splenic atrophy were noted in high-dose females. An increase in non-neoplastic lung lesions
was noted in mid-dose females and high-dose males and females. High-dose females had an
increase in uterine changes. High-dose males had a significant (p < 0.05) increase in the
incidence of stratified squamous cell carcinomas of the tongue. High-dose males and females
had a significant (p < 0.05) increase in the incidence of squamous cell carcinomas of the hard
palate/turbinates.
The chronic (2-year) LOAEL is 10 ng/kg-day, based on the numerous significant
(p < 0.05) changes noted in coproporphyrin excretion (67% increase above control) and an
increase in liver and lung lesions in female rats. The NOAEL is 1 ng/kg-day.
2.4.2.5.5. Maronpot et al., 1993.
An initiation-promotion study was performed in female Sprague-Dawley rats (8-10 rats
per group). Rats were initiated with saline or diethylnitrosamine (DEN), followed 2 weeks later
by promotion with biweekly administration of TCDD (purity not specified) in corn oil via
gavage for 30 weeks. The doses were stated to be equivalent to 3.5, 10.7, 35.7, or
125 ng/kg-day. Rats were sacrificed 7 days after the final treatment. A significant (p < 0.05)
decrease in body weight occurred in the 125 ng/kg-day group. A significant (p < 0.05) increase
in relative liver weight occurred in the 35.7 and 125 ng/kg-day groups. There was a significant
(p < 0.05) increase in the labeling index in the 125 ng/kg-day group, but only with DEN
initiation. In the TCDD-alone group, a 2-fold increase in labeling index occurred in the
125 ng/kg-day group that did not reach statistical significance. A significant (p < 0.05) trend for
increased alkaline phosphatase levels was observed in TCDD-treated animals, but despite a 50%
increase in the highest dose group the increase was not statistically significant. Total cholesterol
and triglycerides were significantly (p < 0.05) higher in the 125 ng/kg-day TCDD-alone group.
A significant (p < 0.05) increase in 5'-nucleotidase occurred in the 35.7 and 125 ng/kg-day
TCDD-alone groups. A dose-dependent increase in the incidence and severity of liver toxicity as
measured by microscopic lesions was observed.
The 30-week LOAEL is 35.7 ng/kg-day, based on a significant (p < 0.05) increase in
relative liver weight (12%, accompanied by increases in incidence and severity of liver lesions).
The 30-week NOAEL is 10.7 ng/kg-day.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-184 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.2.5.6. National Toxicology Program, 1982.
National Toxicology Program (NTP, 1982) conducted a carcinogenic bioassay of TCDD
on rats and mice. Fifty male and female Osborne-Mendel rats and male and female B6C3F1
mice were treated twice per week with TCDD (purity not specified) in corn oil via oral gavage at
doses of 0, 5, 25, or 250 ng/kg for rats and male mice (1.4, 7.1, 71 ng/kg-day adjusted for
continuous exposure; administered doses multiplied by 2 and divided by 7) and 0, 20, 100, or
1,000 ng/kg for female mice (5.7, 28.6, or 286 ng/kg-day adjusted for continuous dosing;
administered doses multiplied by 2 and divided by 7) for 104 weeks. Seventy-five rats and mice
of each sex served as vehicle controls. One untreated control group of 25 rats and mice of each
sex was present in the TCDD treatment room and one untreated control group consisting of
25 rats and mice of each sex were present in the vehicle-control room. Animals surviving until
study termination were sacrificed at 105 or 108 weeks. A complete histopathological evaluation
was conducted on all animals.
Survival rates were not affected by TCDD exposure in rats or mice of either sex. Male
rats exhibited a dose-related depression in mean body weight after week 55, while the females
exhibited a dose-related body-weight depression after 45 weeks of TCDD exposure. However,
the magnitude of the body weight response is not indicated. Mean body weights in male and
female mice were comparable to the vehicle control group throughout the bioassay. Noncancer
histopathologic findings included increased incidences of liver lesions (termed toxic hepatitis)
from TCDD exposure, and were detected in the high-dose rats and high-dose mice of each sex.
A LOAEL for TCDD of 1.4 ng/kg-day for a 104-week exposure duration is identified for
increased incidences of liver lesions in mice of both sexes. A NOAEL cannot be determined for
this study.
2.4.2.5.7. National Toxicology Program, 2006.
Female Sprague-Dawley rats (81 control; 82 treatment group) were administered TCDD
(purity >98%) in corn oikacetone (99:1) via gavage at doses of 0, 3, 10, 22, 46, or
100 ng/kg-day, 5 days per week for 105 weeks (0, 2.14, 7.14, 15.7, 32.9, or 71.4 ng/kg-day,
adjusted for continuous exposure) (NTP, 2006). In addition to this primary group, a stop group
of 50 animals was administered 100 ng/kg-day TCDD in corn oikacetone (99:1) via gavage for
30 weeks and then just the vehicle for the remainder of the study. Up to 10 rats per dose group
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-185 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
from the primary study were sacrificed and evaluated at 14, 31, or 53 (n = 8) weeks for
biologically noteworthy changes in the incidences of neoplasms or non-neoplastic lesions in the
liver, lung, oral mucosa, uterus, pancreas, thymus, adrenal cortex, heart, clitoral gland, ovary,
kidney, forestomach, bone marrow, mesentery gland, and pituitary gland. All animals also
received a complete necropsy and microscopic examination, and the following organs were
weighed: the left kidney, liver, lung, left ovary, spleen, thymus (14 weeks only), and thyroid
gland. At study termination, the number of surviving animals had declined to 25 in the control
group and to 21, 23, 19, 22, and 21 in five treatment groups, respectively, due to accidental
deaths, moribund animals, or death due to natural causes.
Survival rate was not affected by TCDD treatment. Mean body weights in the high dose
primary study group and the 100 ng/kg stop group were less than the vehicle control group after
week 13 of the study. The mean body weights of animals in the 46 ng/kg-day group were less
than in the vehicle control at study termination (2 years), whereas animals in the 22 ng/kg-day
had lower mean body weights compared to controls during the last 10 weeks of study. In
addition to body weight changes, liver weights were also impacted as a result of TCDD
exposure. Absolute and relative liver weights were significantly (eitherp < 0.01 orp< 0.05)
higher in all dose groups compared to controls at the 14- and 31-week evaluation period, whereas
the relative liver weights were significantly (eitherp < 0.01 or p < 0.05) higher only at
>10 ng/kg-day at 53 weeks.
No clinical findings associated with TCDD treatment were observed. TCDD caused
changes in thyroid hormone levels at 14, 31, and 53 weeks. The following changes were
statistically significant (p < 0.05) compared to the vehicle control: decrease in TT4 at doses
>22 ng/kg-day at 14 and 31 weeks and at doses >46 ng/kg-day at 53 weeks; decrease in FT4 at
doses >22 ng/kg-day at 14 and 31 weeks; increase in total T3 at doses >46 ng/kg-day at 14 and
31 weeks and at doses >10 ng/kg-day at 53 weeks; and increase in TSH at doses >46 ng/kg-day
at 14 weeks. There was a statistically-significant (p < 0.05) increase in hepatocyte proliferation
at 14 weeks (22 ng/kg-day group only); 31 weeks (all doses); and 53 weeks (>46 ng/kg-day).
There were statistically significant (p < 0.01) dose-dependent increases in liver (includes EROD
[CYPlAl-associated] activity; 7-pentoxyresorufin-O-deethylase [PROD; CYP2B-associated]
activity; and acetanilide-4-hydroxylase [CYPlA2-associated] activity) and lung (EROD)
cytochrome P450 enzyme activities in all treatment groups at all three evaluation periods
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-186 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
compared to the vehicle control group. The largest effect was an 82-fold induction of hepatic
EROD activity in the 46 ng/kg-day group at 31 weeks.
TCDD was detected at the greatest concentration in the liver, followed by fat tissue, with
tissue concentration increasing in both of these tissues in a dose-dependent manner. TCDD
tissue levels generally remained constant after the first measurement at week 14. Pathological
examination at week 14 revealed increased incidences of hepatocellular hypertrophy in animals
administered >10 ng/kg-day TCDD. Examinations at weeks 31 and 53 indicated that incidence
and or severity of hepatocellular hypertrophy was increased at all treatment doses although
incidences were statistically significant (p < 0.05) only at >10 ng/kg-day doses. The incidence of
non-neoplastic hepatic lesions (including inflammation, necrosis, multiple eosinophilic focus,
diffuse fatty change, pigmentation, toxic hepatopathy) in the liver increased at doses
>22 ng/kg-day beginning at 14 weeks. Severity of the lesions increased at 14 weeks at doses
>46 ng/kg-day and were also observed at lower dose levels during later evaluation periods (31
and 53 weeks). By terminal sacrifice, numerous non-neoplastic changes were noted in TCDD
treated rats, even at the lowest dose tested.
Noncancer cardiovascular and pulmonary effects were evident after 2 years of TCDD
exposure. Significantly increased incidences of minimal to mild cardiomyopathy were seen in
male and female rats at >10 ng/kg-day. In the lung, there was a significant (p < 0.01)
dose-dependent increase, when compared to the vehicle control, in the incidence of bronchiolar
metaplasia of the alveolar epithelium at all dose groups in the primary study.
A LOAEL for TCDD of 2.14 ng/kg-day adjusted dose for a 105-week exposure duration
is identified in this study for significantly (eitherp < 0.01 orp< 0.05) increased absolute and
relative liver weights, increased incidence of hepatocellular hypertrophy, and increased incidence
of alveolar to bronchiolar epithelial metaplasia. A NOAEL cannot be determined for this study.
2.4.2.5.8. Rier et al., 2001a, b.
Female rhesus monkeys (8 per treatment group) were administered 0, 5, or 25 ppt TCDD
(purity not specified) in the diet for 4 years. Previously, Bowman et al. (1989b) determined that
these dietary concentrations were equivalent to 0, 0.15, and 0.67 ng/kg-day, respectively.
Thirteen years after termination of TCDD treatment, serum concentrations of TCDD and
dioxin-like polyhalogenated aromatic hydrocarbons (PHAH) were measured in six control
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-187 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
monkeys, six monkeys treated with 0.15 ng/kg-day, and three monkeys treated with
0.67 ng/kg-day (Rier et al., 2001a). Even after 13 years without treatment, there was
significantly (p < 0.05) elevated serum levels of TCDD and other dioxin-like compounds in
treated monkeys. There was a significant increase in triglycerides and total lipids in the serum of
monkeys treated with either 0.15 or 0.67 ng/kg-day, but not in cholesterol or phospholipids. In
addition to these 15 animals, 8 other female monkeys (4 treated with 0.67 ng/kg-day TCDD that
died 7 to 11 years after treatment and 4 lead-treated animals with no history of PHAH exposure)
were evaluated for endometriosis. Elevated serum concentrations of TCDD were not correlated
with endometriosis. Increased serum levels of 3,3',4,4'-tetrachlorobiphenyl (TCB), however,
were associated with the presence and severity of endometriosis (p < 0.05). TCB was found in
none of the animals without endometriosis, including TCDD-treated animals, nor was it found in
control animals with endometriosis. Animals with elevated serum levels of TCB,
pentachlorobiphenyl, and total serum analyte TCDD equivalents (TEQ) had an increased
incidence of endometriosis, but severity was associated only with increased levels of TCB. EPA
did not develop a LOAEL for TCDD for this study, because of DLC contamination.
In a separate study that evaluated the same 15 monkeys 13 years after exposure, Rier et
al. (2001b) examined effects on systemic immunity. Peripheral blood mononuclear cells
(PBMC) obtained from untreated monkeys secreted no detectable levels of TNF-a in response to
T-cell mitogen exposure. There was, however, a significant (p < 0.05) dose-dependent increase
in TNF-a production in PBMC from the TCDD-treated monkeys. Although PBMC from treated
monkeys with endometriosis produced more TNF-a than cells from unexposed controls without
the disease (median 128 pg/mL compared to not detected; p < 0.01), PBMC from TCDD-treated
animals without endometriosis also produced more TNF-a than controls (median 425 pg/mL,
p < 0.067). TNF-a production from the animals without endometriosis, however, was much
more variable and was not statistically significant compared to controls. In addition, there was a
dose-related but statistically insignificant decrease in PBMC cytotoxicity against natural
killer-sensitive RAJI cells in TCDD-treated animals compared to the unexposed controls. The
results were again related to TCDD exposure and not the presence of endometriosis. TCDD
alone was not associated with changes in PBMC surface antigen expression, but increased serum
levels of TCDD. 1,2,3,6,7,8-Hexachlorodibenzofuran and 3,3',4,4',5-pentachlorobiphenyl were
correlated with increased numbers of CD3+/CD25- and CD3-/CD25+ leukocytes, as well as
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-188 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
increased secretion of TNF-a in response to T-cell mitogen exposure. Although TNF-a
production is considered to be a general indicator of inflammation, relative adversity of increased
TNF-a secreted by PBMCs in and of itself cannot be substantiated in the absence of concurrent
physiological measurements of an inflammatory response. Therefore, neither a LOAEL nor
NOAEL can be determined for this study.
2.4.2.5.9. Sewall et al., 1993.
Sewall et al. (1993) examined the impact of TCDD exposure on the hepatic epidermal
growth factor receptor (EGFR) as a critical effect in hepatocarcinogenicity. In two separate
experiments, groups of 6- to 8-week-old female Sprague-Dawley rats were randomly assigned to
the following groups: control group, receiving saline and corn oil; a promoted group that
received four different doses of TCDD along with saline; a DEN-only initiated control group;
and a DEN and TCDD initiated and promoted group that received four different doses of TCDD.
DEN was administered via intraperitoneal injection at a dose of 175 mg/kg [saline (S) vehicle] as
the initiating agent to animals that were 70 days old. The control animals received saline only.
In the first experiment, each treatment group (S/TCDD and DEN/TCDD) that included
sham-operated or ovariectomized and intact animals were treated with TCDD (purity >98%) at
125 ng/kg-day. In the second dose-response experiment, DEN-initiated and saline control
treatment groups (intact animals, 84 days old) were administered TCDD (purity >98%) in corn
oil via oral gavage once every 2 weeks for 30 weeks at doses equivalent to 0, 3.5, 10.7, 35.7, or
125 ng/kg-day (n = 9). A week after the last treatment, all animals were sacrificed and livers
were harvested and fixed for immunohistochemistry. Sections of the fixed liver were tested for
EGFR binding, EGFR autophosphorylation, immunolocalization of EGFR, and hepatic cell
proliferation.
In the first experiment, intact animals treated with 125 ng/kg-day TCDD exhibited a 65%
reduction in EGFR binding capacity. In contrast, the EGFR equilibrium maximum binding
capacity (Bmax) of the ovariectomized rats was not statistically different from the ovariectomized
control rats, and no changes in the Kd were detected in any treatment group. In the
dose-response experiment with intact animals, a significant (p < 0.05) TCDD dose-dependent
decrease in the Bmax of EGFR was shown. A two-factor, five-level ANOVA indicated that the
effect of TCDD exposure on EGFR Bmax was significant (p = 0.0001), whereas, the effect of
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-189 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
DEN treatment on EGFR Bmax was not significant. Comparative analysis using Fisher's
protected least significant difference indicated that the lowest TCDD dose resulting in a
statistically significant (p < 0.05) decrease in the EGFR Bmax was 10.7 ng/kg-day S/TCDD
group. At the highest TCDD dose of 125 ng/kg-day, the EGFR Bmax was reduced by 38%
compared to controls in both the DEN initiated and noninitiated groups. A two-factor, five-level
ANOVA showed no significant effect on EGFR Kd in either the DEN- or the TCDD-treated
groups. The EGFR autophosphorylation assay indicated that, with increasing TCDD dose, the
amount of EGFR autophosphorylation in DEN/TCDD-treated animals decreased. The study
authors state that this decrease is similar to the dose-response alterations observed for the EGFR
Bmax. Additionally, EGFR autophosphorylation in control and 125 ng/kg-day noninitiated
animals was similar to the corresponding dose levels for the DEN-treated animals, suggesting
that DEN treatment did not affect the EGFR or the EGFR response to TCDD under the
experimental conditions. The immunolocalization assay indicated that staining was more
apparent in the centrilobular and midzonal regions of the liver in the DEN initiated control
animals, whereas, the amount of hepatocyte plasma membrane staining in DEN/TCDD treated
animals substantially decreased. The cell proliferation assay showed a decrease in the cell
labeling index in the 3.5 ng/kg-day DEN/TCDD dose group that was statistically less (p < 0.05)
than the labeling index for the control group. In contrast, the labeling index for the
125 ng/kg-day DEN/TCDD treatment group was significantly (p < 0.05) higher compared to
controls. Except for the low-dose (3.5 ng/kg-day) group, a clear dose-response trend (two
mid-level doses were not statistically significant) was observed in the other three TCDD treated
groups.
The role of EGFR in TCDD-mediated hepatotoxicity is unknown, and as such, this
endpoint cannot be unequivocally linked to TCDD-induced hepatotoxicity nor labeled as
adverse. Thus, no LOAEL/NOAEL was established. A LOEL for TCDD of 3.5 ng/kg-day for a
30-week exposure duration was identified in this study for a significant (p = 0.0001 using
ANOVA) decrease in EGFR Bmax levels. A NOEL cannot be determined for this study.
2.4.2.5.10. Sewall et al., 1995.
Sewall et al. (1995) studied the dose-response relationship for thyroid function alterations
in female rats as a result of TCDD exposure. Groups of female Sprague-Dawley rats were
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-190 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
initiated with DEN at 70 days of age at a dose of 175 mg/kg in a saline vehicle via an i.p.
injection. DEN was administered as a liver-initiating agent for a concurrent study to determine
TCDD promotion of hepatic preneoplastic foci. Saline-treated animals served as controls. At
84 days of age, both the DEN-initiated and the saline-noninitiated groups of animals were
administered TCDD (purity >98%) or corn oil vehicle via oral gavage once every 2 weeks for
30 weeks at dose levels equivalent to 0, 0.1, 0.35, 1.0, 3.5, 10.7, 35.7, or 125 ng/kg-day (n = 9
per group). One week after the last TCDD treatment, the animals were sacrificed and the thyroid
was removed and fixed for further analysis. Blood was drawn from the abdominal aortic vein,
and the serum was isolated and preserved for hormone analysis. Liver was also removed and
prepped for further analysis. Thyroid hormone analysis was performed to determine serum TSH,
T3, and T4 levels using radioimmunoassay kits. Histological examination was conducted on
eosin-stained sections of the thyroid tissue. RNA level in the hepatic tissue was determined
using a reverse transcription polymerase chain reaction (RT-PCR) technique.
TCDD treatment did not affect thyroid weight. A dose-dependent decrease in serum T4
levels was observed in both noninitiated and DEN-initiated animals with T4 levels dropping
significantly (p < 0.05) at the 35 and 125 ng/kg-day TCDD doses in the noninitiated group.
Compared to the noninitiated control group, DEN alone did not significantly affect T4 levels.
Serum T3 level in the 125 ng/kg-day treatment group was slightly elevated but was not
significantly different from levels in the control group. TSH levels in DEN initiated rats were
increased at a dose of 3.5 ng/kg-day. In the noninitiated group, TSH level in the 125 ng
TCDD/kg-day group was 3.27 ± 0.34 ng/mL (n = 9) compared to 1.3 ± 0.18 ng/mL in the corn
oil control group (n = 7). This result, in conjunction with the T4 data, demonstrates that TCDD
had a similar effect on thyroid hormone levels in both the noninitiated and DEN initiated groups.
Histological sections examined for nodular lesions or neoplasms exhibited thyroid follicular
adenoma in one DEN/corn oil control animal. The DEN/TCDD-treated animals exhibited
diffuse follicular hyperplasia, with the size of colloidal follicles decreasing with TCDD
treatment. Other qualitative DEN/TCDD-related changes included increased frequency of
abnormally shaped follicles. The study authors reported that image analysis demonstrated a
significant (p = 0.013) TCDD dose-related decrease in mean follicle size along with a significant
(p = 0.001) TCDD dose-related increase in parenchymal area. Additionally, like T4 and TSH
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-191 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
levels, DEN treatment alone or in combination with TCDD did not influence thyroid follicular or
C-cell morphology.
RT-PCR results for UGT1 and CYP1A1 mRNA levels indicated that the amount of
UGT1 mRNA at the 125 ng/kg-day dose was approximately 2.5-fold higher compared to the
concurrent controls. The study authors also stated that the maximal response for the UGT1
mRNA levels was reached at a dose between 1.0 and 3.5 ng TCDD/kg-day. In contrast, the
maximum induction of CYP1A1 mRNA was 260-fold higher at the 125 ng/kg-day compared to
the concurrent controls.
A LOAEL for TCDD of 35 ng/kg-day for a 30-week exposure duration was identified in
this study for a significant (p < 0.05) decrease in T4 levels. The NOAEL for this study is
10.7 ng/kg-day.
2.4.2.5.11. Toth etal., 1979.
Toth et al. (1979) examined the impact of TCDD exposure on the formation of liver
tumors in male mice. Ten-week-old, outbred Swiss/H/Riop male mice were administered
sunflower oil or TCDD (purity not specified; in sunflower oil) at 0, 7, 700 or 7,000 ng/kg (0, 1,
100, or 1,000 ng/kg-day adjusted for continuous dosing; administered dose divided by 7; n = 38,
44, 44, and 43, respectively) once per week via gastric tube for 1 year. Once exposure had
ceased, animals were followed for the rest of their lives. After spontaneous death or when mice
were moribund, autopsies were performed and all organs were examined histologically.
Average life span in the 1,000 ng/kg-day dose group decreased considerably (72%) when
compared to the control group. TCDD also caused dose-dependent, severe chronic and ulcerous
skin lesions (12, 30, and 58% in the 1, 100, and 1,000 ng/kg-day dose groups, respectively) that
was followed by generalized lethal amyloidosis (12, 23, and 40% in the 1, 100, and
1,000 ng/kg-day dose groups, respectively).
A LOAEL for TCDD of 1 ng/kg-day for 1-year exposure duration was identified in this
study for severe chronic and ulcerous skin lesions (12% higher than controls), and generalized
lethal amyloidosis (12% higher than controls). A NOAEL cannot be determined for this study.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-192 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.4.2.6. Chronic Studies (Cancer Endpoints)
2.4.2.6.1. Kociba et al, 1978.
As discussed above, Kociba et al. (1978) conducted a lifetime (2-year) feeding study of
male and female Sprague-Dawley rats using doses of 0, 1, 10, and 100 ng/kg-day. There were
50 males and 50 females in each group.
With respect to the cancer endpoints examined, the most significant finding was an
increase in hepatocellular hyperplastic nodules and hepatocellular carcinomas in female rats.
The incidence of hepatocellular carcinomas was significantly elevated above the control
incidence at the 100 ng/kg-day dose, whereas increased incidence of hyperplastic nodules was
evident in the 10 ng/kg-day dose group.
There have been two reevaluations of slides of liver sections from the Kociba et al. study
(Squire, 1980; Sauer, 1990; Goodman and Sauer, 1992). The Squire Review was requested by
EPA as an independent review of the slides. The Sauer Review was carried out using refined
criteria for the diagnosis of proliferative hepatocellular lesions (Maronpot et al., 1986, 1989).
Liver tumor incidences for the three evaluations are compared in Appendix F. Although there
are some quantitative differences between the evaluations, the lowest detectable effect for liver
tumor incidence is consistently observed at 10 ng/kg-day.
In the 10 ng/kg-day dose group, significant increases in the incidence of hyperplastic
nodules of the liver were observed in female rats (18/50 in the Kociba evaluation, 27/50 in the
Squire evaluation). Two females (2/50) had hepatocellular carcinomas. In the 1990 reevaluation
(Sauer, 1990; Goodman and Sauer, 1992), nine females (9/50) were identified with
hepatocellular adenomas and none with carcinomas; thus only one-third of the previously
observed "tumors" were identified when using the refined diagnostic criteria. As discussed
below, the tumor reclassification of Goodman and Sauer (1992) was used in the dose-response
modeling for the Kociba et al. (1978) data set.
In addition to nodules in the liver, increased incidence of stratified squamous cell
carcinoma of the tongue and nasal turbinates/hard palate, and keratinizing squamous cell
carcinoma of the lung were also observed in female rats in the 100 ng/kg-day dose group. One
possible cause for the induction of lung tumors in the Kociba feeding study may have been the
aspiration of dosed feed into the lungs. However the promotion of lung tumors has been
observed in mice treated systemically by intraperitoneal (i.p.) injections of TCDD (Beebe et al.,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-193 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
1995). In addition the induction of hyperplastic and metaplastic lesions in rats has been observed
following chronic oral gavage treatment with TCDD (Tritscher et al., 2000). More recently,
chronic oral exposure to HCDD resulted in the induction of lung tumors in treated female rats
(Rozman, 2000). These data indicate that the induction of lung tumors in the Kociba was most
likely primarily the result of systemic chronic dietary exposure to TCDD rather than due to a
localized exposure to aspired dosed feed.
There was no detectable increase in liver tumor incidences in male rats in any of the dose
groups. The mechanism responsible for dioxin-mediated sex specificity for
hepatocarcinogenesis in rats is not clear, but may involve ovarian hormones (Lucier et al., 1991).
Although there was no increase in liver tumors in male rats in this study, in the
100 ng/kg-day group, there was an increased incidence of stratified squamous cell carcinoma of
the hard palate/nasal turbinate, stratified squamous cell carcinoma of the tongue, and adenoma of
the adrenal cortex.
Kociba et al. (1978) had reported that chemically related increases in preneoplastic or
neoplastic lesions were not found in the 1 ng/kg-day dose group. However, Squire identified two
male rats in the 1 ng/kg-day dose group with squamous cell carcinoma of the nasal
turbinates/hard palate, and one of these male rats had a squamous cell carcinoma of the tongue.
These are both rare tumors in Sprague-Dawley rats, and these sites are targets for TCDD,
implying that 1 ng/kg-day may not represent a NOEL. However, no dose-response relationships
were evident for tumors at these sites (Huff et al., 1991)
There is considerable controversy concerning the possibility that TCDD-induced liver
tumors are a consequence of cytotoxicity. Goodman and Sauer (1992) have extended the
reevaluation of the Kociba slides to include liver toxicity data and have reported a correlation
between the presence of overt hepatotoxicity and the development of hepatocellular neoplasms in
female rats. With the exception of two tumors in controls and one each in the low- and mid-dose
groups, all liver tumors occurred in livers showing clear signs of toxicity. However, male rat
livers exhibit cytotoxicity in response to high TCDD doses, yet they do not develop liver tumors.
Moreover, both intact and ovariectomized female rats exhibit liver toxicity in response to TCDD,
yet TCDD is a more potent promoter in intact but not ovariectomized rats (Lucier et al., 1991).
Therefore, if cytotoxicity is playing a role in liver tumorigenesis, other factors must also be
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-194 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
involved. Also, there is little information on the role of cytotoxicity in TCDD-mediated cancer
at other sites such as the lung and thyroid.
2.4.2.6.2. Toth etal., 1979.
In a study of 10-week-old outbred male Swiss/H/Riop mice, Toth et al. (1979)
administered oral gavage TCDD doses of 0, 7, 700, and 7,000 ng/kg-day in sunflower oil weekly
for 1 year (0, 1, 100, or 1,000 ng/kg-day adjusted for continuous dosing; see details above). All
mice (100/group) were followed for their entire lives. The study authors identified the effective
number of mice in each group to be the number of surviving animals when the first
tumor-bearing animal was identified. The average lifespan of the control, low, mid and high
dose groups was 588, 649, 633, and 424 days, respectively.
In the 100 ng/kg-day dose group, liver tumor incidence was twice that of the control
group and was statistically significant (p < 0.01%). A dose-related increase in liver tumor
incidence was observed (18, 29, 48, and 30% in the control and three TCDD-treated groups,
respectively) in all treated mice. Increases were not statistically significant, however, at 1 and
1,000 ng/kg-day. The study authors also stated that spontaneous and induced liver tumors were
not histologically different. Additionally, the ratio of benign hepatomas to hepatocellular
carcinomas in the control group was not affected by treatment and an increase was observed only
in the absolute number of liver tumors. Cirrhosis was not observed with the tumors.
2.4.2.6.3. NTP, 1982.
As discussed above, the NTP (1982) study was conducted using Osborne-Mendel rats
and B6C3F1 mice (NTP, 1982). Groups of 50 male rats, 50 female rats, and 50 male mice
received TCDD as a suspension in corn oikacteone (9:1) by gavage twice each week at doses of
0, 5, 25, or 250 ng/kg-day (daily averaged doses of 0, 1.4, 7.1, or 71 ng/kg-day for rats and male
mice and doses of 0, 5.7, 28.6, or 286 ng/kg-day for female mice.
There were no statistically significant dose-related decreases in survival in any
sex-species group. TCDD-induced malignant liver tumors occurred in the high-dose female rats
and in male and female mice. These can be considered to result from TCDD exposure because
they are relatively uncommon lesions in control Osborne-Mendel rats (male, 1/208; female,
3/208), are seen in female rats and mice of both sexes, and their increasing incidence with
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-195 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
increasing dose is statistically significant (Cochran-Armitage trend test, p = 0.004). Because
liver tumors were increased in both sexes of mice, this effect is not female-specific as was
observed in rats. Interestingly, liver tumor incidences were decreased in female rats in both the
NTP and Kociba low doses (not statistically significant compared with controls). For example,
the combined control incidence data were 11/161 (7%) compared with 4/99 (4%) in the low-dose
group.
The incidences of thyroid gland (follicular cell) tumors were increased in all three dose
groups in male rats. Because the responses in the two highest dose groups are highly significant,
the statistically significant elevation of incidence in the lowest dose group (Fisher exact
p-walue = 0.042) is considered to be caused by exposure to TCDD, suggesting that thyroid tumor
incidence may be the most sensitive site for TCDD-mediated carcinogenesis. Because
71 ng/kg-day is above the maximum tolerated dose (MTD) (Huff et al., 1991), thyroid tumors
occur at doses more than 50 times lower than the MTD.
TCDD-induced neoplasms of the adrenal gland were observed in the 7.1 ng/kg-day/dose
group in male rats and in high-dose female rats. Fibrosarcomas of the subcutaneous tissue were
significantly elevated in high-dose female mice and female rats. One additional tumor type,
lymphoma, was seen in high-dose female mice. Lung tumors were elevated in high-dose female
mice; the increase was not statistically significant when compared with concurrent controls, but
the increase was dose related (Cochran-Armitage trend test,/? = 0.004).
Huff (1992) concluded, based on the NTP bioassay results, that TCDD was a complete
carcinogen and induced neoplasms in rats and mice of both sexes. As was observed in the
Kociba study (Kociba et al., 1978), liver tumors were observed with greater frequency in treated
female rats, but in male rats the thyroid appears to be the most sensitive (increased tumor
incidence at doses as low as 1.4 ng/kg-day).
2.4.2.6.4. NTP, 2006.
As discussed above, female Sprague-Dawley rats (81 control; 82 treatment group) were
administered TCDD (purity >98%) in corn oikacetone (99:1) via gavage at doses of 0, 3, 10, 22,
46, or 100 ng/kg-day, 5 days per weekfor 105 weeks (0, 2.14, 7.14, 15.7, 32.9, or
71.4 ng/kg-day, adjusted for continuous exposure) (NTP, 2006). In addition to this primary
group, a stop-dose group of 50 animals was administered 100 ng/kg-day TCDD in corn
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-196 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
oil:acetone (99:1) via gavage for 30 weeks and then just the vehicle for the remainder of the
study. At study termination, the number of surviving animals had declined to 25 in the control
group and to 21, 23, 19, 22, and 21 in five treatment groups, respectively, due to accidental
deaths, moribund animals, or death due to natural causes.
Incidence of hepatocellular adenomas was significantly (p < 0.001) increased in the
100 ng/kg-day dose group in the primary study and exceeded incidences seen in historical
vehicle control range at study termination. A dose-related increase in the incidence of
cholangiosarcoma was seen in the primary study group in animals receiving 22 ng/kg-day or
higher doses of TCDD. The high dose group of 100 ng/kg-day had the highest incidence of
cholangiosarcoma with a significant (p < 0.001) number of animals exhibiting multiple
cholangiosarcomas. Such an incidence was not seen in historical vehicle controls. In contrast,
only two cholangiosarcomas and hepatocellular adenomas were seen in the 100 ng/kg-day group
in the stop-exposure study.
In the lung, at 2 years, there was a significantly (p = 0.002) increased incidence of cystic
keratinizing epithelioma in the 100 ng/kg-day dose group of the primary study, while there were
no epitheliomas in the 100 ng/kg-day group of the stop-exposure study. There was also a
significant (p < 0.01) dose-dependent increase, when compared to the vehicle control, in the
incidence of bronchiolar metaplasia of the alveolar epithelium at all dose groups in the primary
study. Squamous metaplasia was also present in the 46 and 100 ng/kg-day dose groups in the
primary study, and was also observed in the 100 ng/kg-day dose group in the stop-exposure
study.
A positive trend in the incidence of gingival squamous cell carcinoma of the oral cavity
was seen at all doses (except 22 ng/kg-day), with the incidence significantly (p = 0.007) high in
the 100 ng/kg-day dose group. In addition, the occurrence of this lesion in the 46 and
100 ng/kg-day group of the primary study and 100 ng/kg-day group of the stop-exposure study
exceeded the historical control range. The incidence of gingival squamous hyperplasia was
significantly (either p < 0.01 or p< 0.05) increased in all dose groups of the primary study as
well as the 100 ng/kg-day group of the stop-exposure study.
In the uterus, at 2 years, there was a significantly (p = 0.032) higher rate of squamous cell
carcinoma in the 46 ng/kg-day group compared to vehicle controls. In addition there were two
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-197 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
squamous cell carcinomas in the 100 ng/kg-day group of the stop-exposure study. No squamous
cell carcinomas have been reported in historical vehicle controls.
These results indicate that TCDD is carcinogenic to female Sprague-Dawley rats and
causes tumors at multiple sites.
2.4.3. Summary of Key Data Set Selection for TCDD Dose-Response Modeling
To meet the NAS concerns regarding transparency and clarity in the identification of
TCDD studies for dose-response assessment, EPA has, in this section, developed and applied
two sets of criteria, for animal bioassays and epidemiologic studies. EPA has collected and
evaluated these studies, including studies from the 2003 Reassessment and newer studies found
via literature searches and through public submissions. Tables 2-4 and 2-5 contain the final lists
of key cancer and noncancer studies, respectively, that have met EPA's inclusion criteria for
epidemiologic data. Tables 2-6 and 2-7 provide the final lists of key studies that have met EPA's
inclusion criteria for animal bioassay data for cancer and noncancer studies, respectively.
Collectively, these four tables contain the final set of key studies that EPA has used to develop
noncancer and cancer dose-response assessments for TCDD in Sections 4 and 5 of this
document. In Sections 4 and 5, additional evaluations are made to determine which
study/endpoint data sets are the most appropriate for development of the RfD and OSF for
TCDD, using statistical criteria, dose-response modeling results and decisions regarding
toxicological relevance of the endpoints. The approaches taken to select the final candidate
study/endpoint data sets are discussed in Sections 4 and 5 and are illustrated in Figures 4-1 and
5-1 of those sections.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-198 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 2-1. Summary of epidemiologic cancer studies (key characteristics)
2
Publication
Length of
follow-up
Latency period
Half-life for TCDD
Fraction of TEQs
accounted for by TCDD
NIOSH cohort studies
Fingerhut et al., 1991
1942-1987
0, 20 years
N/A
N/A
Steenland et al., 1999
1942-1993
0, 15 years
N/A
N/A
Steenland et al., 2001
1942-1993
0, 15 years
8.7 years (Michalek et
al., 1996)
TCDD accounted for all
occupational TEQ; 10%
of background
Cheng et al., 2006
1942-1993
0, 10, 15 years
8.7 years (Michalek et
al., 1996), and CADM
(Aylward et al., 2005a)
N/A
Collins et al., 2009
1942-2003
None
7.2 years (Flesch-Janys
et al., 1996)
N/A
BASF cohort studies
Thiess et al., 1982
1953-1980
None
N/A
N/A
Zoberetal., 1990
1953-1987
Years since first
exposure: 0-9,
10-19, and 20+
N/A
N/A
Ott and Zober, 1996
1953-1991
None
5.8 years
N/A
Hamburg cohort studies
Manz et al., 1991
1952-1989
None, used
duration of
employment
(<20, >20 years)
N/A
N/A
Flesch-Janys et al.,
1995
1952-1992
None
7.2 years
(Flesch-Janys et al.,
1994)
Mean TEQ without
TCDD was 155 ng/kg;
mean TEQ with TCDD
was 296.5 ng/kg
Flesch-Janys et al.,
1998
1952-1992
None
7.2 years (Flesch-Janys
et al., 1996), also used
decay rates that were
function of age and fat
composition
Mean concentration of
TCDD was 101.3 ng/kg;
for TEQ (without TCDD)
mean exposure was 89.3
ng/kg
Becheretal., 1998
1952-1992
0, 5, 10, 15 and
20 years
7.2 years (Flesch-Janys
et al., 1996) took into
account age and fat
composition
Not described
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-199 DRAFT—DO NOT CITE OR QUOTE
-------
1
Table 2-1. Summary of epidemiologic cancer studies (key characteristics)
(continued)
Publication
Length of
follow-up
Latency period
Half-life for TCDD
Fraction of TEQs
accounted for byTCDD
Seveso cohort studies
Bertazzi et al., 2001
1976-1996
Periods
postexposure: 0,
0-4, 5-9,
10-14, 15-19
N/A
N/A
Consonni et al., 2008
1976-2001
Periods
postexposure: 0,
0-4, 5-9,
10-14, 15-19,
20-24
N/A
N/A
Warner et al., 2002
1976-1998
None
8 years (Pirkle et al.,
1989)
N/A
Chapaevsk cohort studies
Revich et al., 2001
Cross-
sectional
study
(1995-1998)
N/A
N/A
N/A
Ranch Hand cohort studies
Akhtar et al., 2004
1962-1999
None
N/A
N/A
Michalek and Pavuk,
2008
1962-2004
None, but
stratified by
period of service
7.6 years
N/A
New Zealand cohort studies
McBride et al., 2009a
1969-2004
None
N/A
N/A
McBride et al., 2009b
1969-2004
None
7 years
N/A
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-200 DRAFT—DO NOT CITE OR QUOTE
-------
Table 2-2. Epidemiology cancer study selection considerations and criteria
Methods
use to
ascertain
health
outcomes
were
unbiased,
highly
sensitive
and
specific.
Risk
estimates are
not
susceptible
to biases
from
confounding
exposures or
from study
design or
statistical
analysis.
Association
between
TCDD and
adverse
health effect,
with
exposure-
response
relationship.
Exposure
assessment
methodology
clear and
adequately
characterizes
individual-
level
exposures.
Limitations
and
uncertainties
in exposure
assessment
considered.
Study size
and follow-
up large
enough to
yield precise
estimates of
risk and
ensure
adequate
statistical
power.
Published
in peer-
reviewed
literature
with
appropriate
discussion
of
strengths,
limitations.
Exposure
primarily
TCDD and
quantified so
that dose-
response
relationship
can be
assessed.
Effective dose
& oral
exposure
estimable &
consistent w/
current
biological
understanding.
Latency and
appropriate
window(s) of
exposure
examined.
Pass for
dose-
response
analyses?
Cancer
Considerations
Criteria
Y/N
NIOSH Cohort Studies
Fingerhut et al., 1991
all cancer sites, site-specific analyses
a/
X
X
X
a/
a/
X
a/
N
Steenland et al.. 1999
all cancer sites combined, site-specific analyses
a/
a/
a/
a/
a/
a/
a/
a/
a
Steenland et al., 2001
all cancer sites combined
V
a/
a/
a/
V
V
a/
V
Y
Cheng et al., 2006
all cancer sites combined
V
V
V
V
V
V
V
N
a/
Y
Collins et al., 2009
all cancer sites combined, site-specific analyses
V
V
X
V
V
V
V
X
N
BASF Cohort Studies
Zoberetal., 1990
all cancer sites combined, site-specific analyses
V
V
X
X
X
V
X
X
N
Ott and Zober, 1996
all cancer sites combined
V
V
a/
a/
a/
V
a/
a/
Y
-------
o
>3*
§•
s;
3
s
>;*
5r
^s
^s
¦
*
^S
s;
K> 8
to g
s§
I
s^
3 §-
£> >;
a
H o
.. ^k
o
o
s
2! §-
O ^
H <£
hh Oq
H <3
W|
o ^
o
H
ffl
Table 2-2. Epidemiology cancer study selection considerations and criteria (continued)
Exposure
assessment
Risk
methodology
clear and
Effective dose
& oral
Methods
use to
estimates are
not
adequately
characterizes
Study size
and follow-
Published
exposure
estimable &
ascertain
health
susceptible
to biases
Association
between
individual-
level
up large
enough to
in peer-
reviewed
Exposure
primarily
consistent w /
current
outcomes
were
unbiased.
from
confounding
exposures or
TCDD and
adverse
health effect.
exposures.
Limitations
and
yield precise
estimates of
risk and
literature
with
appropriate
TCDD and
quantified so
that dose-
biological
understanding.
Latency and
highly
from study
with
uncertainties
ensure
discussion
response
appropriate
Pass for
sensitive
and
specific.
design or
statistical
analysis.
exposure-
response
relationship.
in exposure
assessment
considered.
adequate
statistical
power.
of
strengths,
limitations.
relationship
can be
assessed.
window(s) of
exposure
examined.
dose-
response
analyses?
Cancer
Considerations
Criteria
Y/N
Hamburg Cohort
Manz et al., 1991
all cancer sites combines, site-specific analyses
a/
a/
a/
a/
a/
a/
X
a/
N
Flesh-Janys et al.. 1995
all cancer sites combined
a/
a/
a/
a/
a/
a/
a/
X
N
Flesh-Janys et al., 1998
all cancer sites combined, site-specific analyses
V
V
V
V
V
V
a/
a/
b
Becher et al., 1998
all cancer sites combined
V
V
V
V
V
V
V
V
Y
Seveso Cohort
N
Bertazzi et al., 2001
all cancer sites combined, site-specific analyses
V
V
V
X
V
V
X
X
N
Pesatori et al., 2003
all cancer sites combined, site-specific analyses
V
V
X
X
V
V
X
X
N
Consonni et al., 2008
all cancer sites combined, site-specific analyses
V
V
a/
X
V
V
X
X
N
O
O
-------
Table 2-2. Epidemiology cancer study selection considerations and criteria (continued)
Exposure
assessment
Risk
methodology
clear and
Effective dose
& oral
Methods
use to
estimates are
not
adequately
characterizes
Study size
and follow-
Published
exposure
estimable &
ascertain
health
susceptible
to biases
Association
between
individual-
level
up large
enough to
in peer-
reviewed
Exposure
primarily
consistent w/
current
outcomes
were
unbiased,
highly
sensitive
and
specific.
from
confounding
exposures or
from study
design or
statistical
analysis.
TCDD and
adverse
health effect,
with
exposure-
response
relationship.
exposures.
Limitations
and
uncertainties
in exposure
assessment
considered.
yield precise
estimates of
risk and
ensure
adequate
statistical
power.
literature
with
appropriate
discussion
of
strengths,
limitations.
TCDD and
quantified so
that dose-
response
relationship
can be
assessed.
biological
understanding.
Latency and
appropriate
window(s) of
exposure
examined.
Pass for
dose-
response
analyses?
Cancer
Considerations
Criteria
Y/N
Baccarelli et al., 2006
site specific analysis
a/
a/
X
a/
a/
a/
a/
a/
C
Warner et al.. 2002
breast cancer incidence
a/
a/
a/
a/
a/
a/
a/
a/
Y
Chapaevsk Study
N
Revich et al., 2001
all cancer sites combined, site-specific analyses
X
X
X
X
V
X
X
X
N
Ranch Hands Cohort
Akhtar et al., 2004
all cancer sites combined, site-specific analyses
a/
X
a/
a/
V
a/
X
a/
N
Michalek and Pavuk, 2008
all cancer sites combined
V
X
V
V
V
V
X
a/
N
Others
tMannetje et al., 2005
all cancer sites combined, site-specific analyses
V
X
V
V
V
X
X
X
N
McBride et al., 2009b
all cancer sites combined, site-specific analyses
V
X
X
V
X
a/
X
X
N
-------
Table 2-2. Epidemiology cancer study selection considerations and criteria (continued)
> |5*
&
to
s
S
to
s
>;*
a
a,
Sf
TO
TO'
*
^5
Exposure
assessment
methodology
Effective dose
Risk
clear and
& oral
Methods
estimates are
adequately
Study size
exposure
use to
not
characterizes
and follow-
Published
estimable &
ascertain
susceptible
Association
individual-
up large
in peer-
Exposure
consistent w /
health
to biases
between
level
enough to
reviewed
primarily
current
outcomes
from
TCDD and
exposures.
yield precise
literature
TCDD and
biological
were
confounding
adverse
Limitations
estimates of
with
quantified so
understanding.
unbiased,
exposures or
health effect,
and
risk and
appropriate
that dose-
Latency and
highly
from study
with
uncertainties
ensure
discussion
response
appropriate
Pass for
sensitive
design or
exposure-
in exposure
adequate
of
relationship
window(s) of
dose-
and
statistical
response
assessment
statistical
strengths,
can be
exposure
response
specific.
analysis.
relationship.
considered.
power.
limitations.
assessed.
examined.
analyses?
Cancer
Considerations
Criteria
Y/N
McBride et al., 2009a
all cancer sites combined, site-specific analyses
a/
a/
X
a/
X
a/
a/
a/
d
Hooiveld et al., 1998
all cancer sites combined, site-specific analyses
a/
a/
a/
a/
X
a/
a/
X
N
K> 8
to t>
o
0
s
1
a,
TO
>1
s
o
^•k
TO
O
s
2! »
O ^
H r?
hh Oq
H to
W|
o
o
H
ffl
o
o
N
"This study has been superseded and updated by Steenland et al. (2001).
bBecher et al. (1998) assessed this same cohort taking cancer latency into account, thereby superseding this study.
°It is unknown whether the frequency of t(14;18)translocations in lymphocytes relates specifically to an increased risk of non-Hodgkin's lymphoma. Given this
lack of obvious adverse effect, dose-response analyses for this outcome were not conducted.
dNo dose-response associations were noted.
a/ = Consideration/criteria satisfied; X= Consideration/criteria not satisfied.
-------
Table 2-3. Epidemiology noncancer study selection considerations and criteria
Exposure
assessment
Effective dose
Risk
methodology
clear and
Study size
and
& oral exposure
estimable &
Methods
estimates
adequately
follow-up
consistent w/
use to
are not
characterizes
large
current
ascertain
health
outcomes
were
unbiased.
susceptible
to biases
from
confounding
exposures
Association
between
TCDD and
adverse
health effect.
individual-
level
exposures.
Limitations
and
enough to
yield
precise
estimates
of risk and
Published in
peer-
reviewed
literature
with
Exposure
primarily
TCDD and
quantified so
that dose-
biological
understanding.
Latency and
appropriate
window(s) of
highly
sensitive
and
specific.
or from
study design
or statistical
analysis.
with
exposure-
response
relationship.
uncertainties in
exposure
assessment
considered.
ensure
adequate
statistical
power.
appropriate
discussion ol
strengths,
limitations.
response
relationships
can be
assessed.
exposure
examined for a
Non-Fatal
endpoint.
Pass for
dose-
response
analyses?
Noncancer
Considerations
Criteria
Y/N
NIOSH Cohort
Steenland et al., 1999
mortality (noncancer) -ischemic heart disease
a/
X
a/
a/
a/
a/
X
X
N
Collins et al.. 2009
mortality (noncancer)
a/
a/
X
a/
a/
a/
a/
X
N
BASF Cohort
Ott and Zober, 1996
mortality (noncancer)
V
a/
X
V
V
V
a/
X
N
Hamburg Cohort
Flesch-Janys et al.. 1995
mortality (noncancer)
V
V
a/
V
V
V
V
X
N
Seveso Cohort
Eskenazi et al., 2002a
menstrual cycle characteristics
V
V
a/
V
V
V
V
a/
Y
Eskenazi et al., 2002b
endometriosis
X
X
X
V
X
V
V
X
N
-------
o
>3*
§•
s;
3
s
>;*
5r
^s
^s
¦
*
^S
s;
K> 8
to g
-§
I
3 §-
^
Tj S
H ©
3 §
2 §
o ^
H <£
°r^
hh Oq
H <3
MJ
o ^
o
H
W
Table 2-3. Epidemiology noncancer study selection considerations and criteria (continued)
Exposure
assessment
Effective dose
Risk
methodology
clear and
Study size
and
& oral exposure
estimable &
Methods
estimates
adequately
follow-up
consistent w /
use to
are not
characterizes
large
current
ascertain
health
outcomes
susceptible
to biases
from
Association
between
TCDD and
individual-
level
exposures.
enough to
yield
precise
Published in
peer-
reviewed
Exposure
primarily
TCDD and
biological
understanding.
Latency and
were
confounding
adverse
Limitations
estimates
literature
quantified so
appropriate
unbiased.
exposures
health effect.
and
of risk and
with
that dose-
window(s) of
highly
sensitive
and
specific.
or from
study design
or statistical
analysis.
with
exposure-
response
relationship.
uncertainties in
exposure
assessment
considered.
ensure
adequate
statistical
power.
appropriate
discussion ol
strengths,
limitations.
response
relationships
can be
assessed.
exposure
examined for a
Non-Fatal
endpoint.
Pass for
dose-
response
analyses?
Noncancer
Considerations
Criteria
Y/N
Eskenazi et al., 2003
birth outcomes
X
X
X
a/
a/
a/
a/
X
N
Warner et al.. 2004
age at menarch
a/
a/
X
a/
a/
a/
a/
X
N
Eskenazi et al., 2005
age at menopause
a/
a/
X
V
V
V
V
X
N
Warner et al.. 2007
ovarian function
V
V
X
V
V
V
V
X
N
Eskenazi et al., 2007
uterine leiomyoma
V
V
a/
V
V
V
V
X
Na
Mocarelli et al., 2008
semen quality
V
V
a/
V
V
V
V
a/
Y
Mocarelli et al., 2000
sex ratio
V
V
V
V
V
X
V
X
Nb
Baccarelli et al., 2008
neonatal thyroid function
V
V
V
X
V
a/
V
a/
Y
-------
Table 2-3. Epidemiology noncancer study selection
considerations and criteria (continued)
Exposure
assessment
Effective dose
Risk
methodology
clear and
Study size
and
& oral exposure
estimable &
Methods
estimates
adequately
follow-up
consistent w/
use to
are not
characterizes
large
current
ascertain
health
outcomes
were
unbiased.
susceptible
to biases
from
confounding
exposures
Association
between
TCDD and
adverse
health effect.
individual-
level
exposures.
Limitations
and
enough to
yield
precise
estimates
of risk and
Published in
peer-
reviewed
literature
with
Exposure
primarily
TCDD and
quantified so
that dose-
biological
understanding.
Latency and
appropriate
window(s) of
highly
sensitive
and
specific.
or from
study design
or statistical
analysis.
with
exposure-
response
relationship.
uncertainties in
exposure
assessment
considered.
ensure
adequate
statistical
power.
appropriate
discussion ol
strengths,
limitations.
response
relationships
can be
assessed.
exposure
examined for a
Non-Fatal
endpoint.
Pass for
dose-
response
analyses?
Alaluusua et al., 2004
oral hygiene
a/
a/
a/
a/
a/
a/
a/
a/
Y
Bertazzi et al.. 2001
mortality (noncancer)
a/
a/
X
X
a/
a/
X
X
N
Consonni et al., 2008
mortality (noncancer)
V
V
X
X
V
V
X
X
N
Baccarelli et al.. 2005
chloracne
V
V
a/
a/
V
V
a/
a/
C
Baccarelli et al., 2002, 2004
immunological effects
V
V
a/
a/
V
V
a/
X
N
Chapaevsk Study
N
Revich et al., 2001
mortality (noncancer) and reproductive health
V
X
X
X
V
V
X
X
N
Ranch Hands Cohort
Michalek and Pavuk, 2008
diabetes
V
X
a/
a/
V
V
X
a/
N
Other
McBride et al., 2009a
-------
Table 2-3. Epidemiology noncancer study selection considerations and criteria (continued)
> |5*
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
TO
TO'
*
^5
Exposure
assessment
Effective dose
methodology
Study size
& oral exposure
Risk
clear and
and
estimable &
Methods
estimates
adequately
follow-up
consistent w /
use to
are not
characterizes
large
current
ascertain
susceptible
Association
individual-
enough to
Published in
Exposure
biological
health
to biases
between
level
yield
peer-
primarily
understanding.
outcomes
from
TCDD and
exposures.
precise
reviewed
TCDD and
Latency and
were
confounding
adverse
Limitations
estimates
literature
quantified so
appropriate
unbiased,
exposures
health effect,
and
of risk and
with
that dose-
window(s) of
highly
or from
with
uncertainties in
ensure
appropriate
response
exposure
Pass for
sensitive
study design
exposure-
exposure
adequate
discussion of
relationships
examined for a
dose-
and
or statistical
response
assessment
statistical
strengths,
can be
Non-Fatal
response
specific.
analysis.
relationship.
considered.
power.
limitations.
assessed.
endpoint.
analyses?
mortality (noncancer)
X
X
X
a/
X
a/
a/
X
N
McBride et al., 2009b
mortality (noncancer)
X
a/
X
a/
X
a/
X
X
N
Ryan et al., 2002
sex ratio
X
X
X
X
a/
V
X
X
N
K> 8
K 1 ^
N) >>
O
00
>3
0
1
a,
TO
>1
s
o
^k
TO
O
s
2! »
O ^
H r?
hh Oq
H to
W|
o
o
H
ffl
o
o
aCategorical measures of TCDD suggest an inverse association between TCDD exposure and uterine fibroids. The observed direction of the reported
associations precluded quantitative dose-response modeling.
bThe somewhat arbitrary cut off age of 19 for statistically significant exposure associations results in a highly uncertain critical exposure window. It is difficult
to determine whether effects are a consequence of the initial high exposure during childhood or a function of the cumulative exposure for this entire exposure
window. The differences between these two dose estimates are quite large.
°Chloracne is recognized to occur following high TCDD exposure levels. This study provides limited relevance to TCDD RfD development, as exposure levels
observed in the general population are much lower.
a/ = Consideration/criteria satisfied. X= Consideration/criteria not satisfied.
-------
Table 2-4. Epidemiology studies selected for TCDD cancer dose-response modeling
Location,
No. of
Health
time
Cohort
Exposure
Exposure
cases/
Effect Measure/
outcome
period
description
assessment
measures
deaths
RR (95% CI)
Risk factors
Comments
Reference
Mortality
USA,
NIOSH cohort
Cumulative
No exposure
256
The slope (P) was
Available: age,
Confounding by
Cheng et
from all
1942-1993
including 3,538
serum lipid
categories
cancer
3.3 x 10"6 for lag
year of birth, and
smoking was
al., 2006
cancers
occupationally
TCDD
provided
deaths
of 15 years
race
considered indirectly
exposed male
concentrations
excluding upper
by analysis of
workers at 8
(CSLC) based
5% of TCDD
Risks adjusted for:
smoking-related and
plants in the
on work
exposures.
year of birth, age,
smoking-unrelated
United States;
histories, job-
The slopes ranged
and race
cancers.
256 cancer
exposure
two orders of
Other occupational
deaths
matrix, and
magnitude
exposures were
concentration
depending on
considered indirectly
and age-
modeling
by repeated analyses
dependent two-
assumption
removing one plant
compartment
at a time.
model of
Based on indirect
elimination
evaluation, there
kinetics
was no clear
evidence of
confounding.
Mortality
USA,
NIOSH cohort
CSLC based on
CSLC
Available: date of
Included in
Steenland et
from all
1942-1993
including 3,538
work histories,
(ppt-years)
birth and age
U.S. EPA (2003)
al., 2001
cancers
male workers,
job-exposure
<335
64
1.00
256 cancer
matrix, and a
335-520
29
1.26 (0.79-2.00)
Adjusted for: date
deaths
simple one-
520-1,212
22
1.02 (0.62-1.65)
of birth, and age
compartment
1,212-2,896
30
1.43 (0.91-2.25)
was used as time
first-order
2,896-7,568
31
1.46 (0.93-2.30)
scale in Cox
pharmacokineti
7,568-20,455
32
1.82(1.18-2,82)
model
c elimination
>20,455
48
1.62 (1.03-2,56)
model with 8.7-
year half-life
-------
o ^
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
^s
TO
TO'
*
^5
«
*> s
TO
>;
0
s
1
a,
§•
>3
o
o
s
>3
Table 2-4. Epidemiology studies selected for TCDD cancer dose-response modeling (continued)
Location,
No. of
Health
time
Cohort
Exposure
Exposure
cases/
Effect Measure/
outcome
period
description
assessment
measures
deaths
RR (95% CI)
Risk factors
Comments
Reference
Mortality
Hamburg,
Boehringer
Cumulative
Categorical
124
Included in
Becher et
from all
Germany,
cohort including
TCDD serum
exposures
U.S. EPA (2003)
al., 1998
cancers
production
approximately
lipid
(Cox model)
combined
period was
1,189 workers
concentrations
0-
-------
Table 2-4. Epidemiology studies selected for TCDD cancer dose-response modeling (continued)
o <5;
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
TO
TO'
*
^5
s
TO
0
s
1
a,
>3
o
o
s
>3
to
o
o
O S"
H r?
<*^
l-H CfQ
H
W|
o
o
H
W
Location,
No. of
Health
time
Cohort
Exposure
Exposure
cases/
Effect Measure/
outcome
period
description
assessment
measures
deaths
RR (95% CI)
Risk factors
Comments
Reference
Mortality
Ludwig-
BASF cohort,
Cumulative
Internal
Internal
Available: age,
Included in
Ottand
and
shafen,
243 men
TCDD serum
comparisons
cohort
BMI, smoking
U.S. EPA (2003)
Zober, 1996
incidence
Germany,
exposed from
lipid
based on
analysis
status and history
for all
1954-1992
accidental
concentrations
continuous
Date of 1st TCDD
of occupational
cancers
release that
expressed in
measure of
31
exposure
exposure to
QfrtTTldtlP
Positive associations
combined,
occurred in
|ig/kg based on
TCDD.
cancer
1.22 (95% CI:
dlUllldllL'
noted for digestive
as well as
1953 during
TCDD half-life
deaths
1.00-1.50)
amines and
cancer, but not for
for specific
production of
of 5.1-8.9 years,
asbestos
respiratory cancer
cancer sites
trichlorophenol,
Cox regression
47
or who were
model
incident
1.11 (95% CI:
Associated between
involved in
cancers
0.91-1.35)
TCDD and
clean-up
increased SMRs
activities
External
found only among
External
cohort
current smokers
comparisons
analyses
exposure
Last published
categories:
Deaths
SMRs
account of this
<0.1,
8
0.8 (0.4-1.6)
cohort
0.1-0.99,
8
1.2 (0.5-2.3)
1.0-1.99
8
1.4 (0.6-2.7)
>2 f-ig/kg
7
2.0 (0.8-4.0)
-------
Table 2-4. Epidemiology studies selected for TCDD cancer dose-response modeling (continued)
Location,
No. of
Health
time
Cohort
Exposure
Exposure
cases/
Effect Measure/
outcome
period
description
assessment
measures
deaths
RR (95% CI)
Risk factors
Comments
Reference
Breast
Italy
981 women
TCDD serum
Cases
Available:
Included in
Warner et
cancer
1976-1998
from zones A
lipid
<20 ppt
1
1.0
gravidity, parity,
U.S. EPA (2003)
al., 2002
incidence
and B with
concentrations
20.1-44 ppt
2
1.0 (0.1-10.8)
age at first
available
(ppt) collected
44.1-100 ppt
7
4.5 (0.6-36.8)
pregnancy, age at
archive serum
between 1976
>100 ppt
5
3.3 (0.4-28.0)
last pregnancy,
samples, 15
and 1981. For
lactation, family
breast cancer
most samples
LogioTCDD
15
2.1 (1.0-4.6)
history of breast
cases
collected after
also modeled
cancer, age at
1977, serum
as continuous
menarche, current
TCDD levels
variable
body mass index,
were back-
oral contraceptive
extrapolated
use, menarcheal
using a first-
status at
order kinetic
explosion,
model with a 9-
menopause status
year half-life.
at diagnosis,
height, smoking,
alcohol
consumption.
Adjusted for age,
which was used as
time scale in Cox
model; other
covariates were
evaluated but
were not
identified as
confounders.
-------
Table 2-5. Epidemiology studies selected for TCDD noncancer dose-response modeling
No. of
Health
Location,
Cohort
Exposure
Exposure
cases/
Effect Measure/
outcome
time period
description
assessment
measures
deaths
RR (95% CI)
Risk factors
Comments
Reference
b-TSH
Italy, 1976;
Population-
Based on zone
Population-
Population-based
Available: gender,
An association with
Baccarelli
measured 72
children,
based study:
of residence,
based study:
study
birth weight, birth
serum TCDD levels
et al., 2008
hours after
1994-2005
1,041
estimated mean
Mean b-TSH
order, maternal age
of mothers was
birth from a
singletons
values from a
at delivery,
found with b-TSH
heel pick
(56 from
previous study.
Reference
533
Reference:
hospital, type of
among the 51 births
(routine
zone A, 425
Maternal
births
0.98 (95% CI:
delivery.
in the plasma dioxin
screening for
from zone B
plasma TCDD
0.90-1.08)
study.
all newborns in
and 533 from
levels estimated
Zone B
425
Zone B:
There was limited
the region
reference)
at the date of
delivery using a
first-order
pharmacokineti
c model and
births
1.66 (95% CI:
evidence of
born between
1.19-2.31)
confounding, so
Jan. 1, 1994-
Zone A
56
Zone A:
mean TSH results
June 30,
births
1.35 (95% CI:
presented here are
2005.
elimination rate
1.22-1.49)
unadjusted.
Plasma
estimated in
dioxin study:
Seveso women
Plasma
Association
51 children
born to 38
women of
fertile age
who were
(half-life =9.8
years).
dioxin
study:
Continuous
maternal
between neonatal
b-TSH with
plasma TCDD:
adjusted p = 0.75
part of the
plasma
(p< 0.001)
Seveso
TCDD
Chloracne
Study.
-------
o ^
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
^s
TO
TO'
*
^5
«
*> s
TO
>3
Table 2-5. Epidemiology studies selected for TCDD noncancer dose-response modeling (continued)
No. of
Health
Location,
Cohort
Exposure
Exposure
cases/
Effect Measure/
outcome
time period
description
assessment
measures
deaths
RR (95% CI)
Risk factors
Comments
Reference
Sperm conc.
Italy, 1976,
135 exposed
Serum TCDD
TCDD
Mean values
Available: age,
Results stratified by
Mocarelli
(million/ mL)
1998
(from zone
(in ppt) from
quartiles
were compared
abstinence time,
timing of exposure
et al., 2008
Progressive
A) and 184
1976-1977
between the
smoking status,
(1-9 yrs old vs.
motility (%)
non-exposed
samples (for
exposed and
education, alcohol
10-17 yrs old in
Serum E2
men aged
exposed men);
comparison
use, maternal
1976).
(pmol/L)
1-26 in 1976
background
groups for sperm
smoking during
were
values were
concentration,
pregnancy,
included.
assumed for
volume, motility
employment status,
These
unexposed men
and count, FSH,
BMI, chronic
subjects were
based on serum
E2, LH, and
exposure to
selected from
analysis of
InhibinB.
solvents and other
the cohort of
residents in
toxic substances.
257 exposed
uncontaminated
and 372
areas.
Adjusted for
unexposed
smoking status,
people.
organic solvents,
age at time of tests,
BMI, alcohol use,
education,
employment status
and abstinence
(days) for sperm
data.
Hormone data not
adjusted for
education level,
employment status,
and abstinence
time.
to
0
s
1
a,
>3
o
o
s
>3
o
o
O S"
H r?
<*^
l-H CfQ
H
W|
O
o
H
W
-------
Table 2-5. Epidemiology studies selected for TCDD noncancer dose-response modeling (continued)
o <5;
&
to
s
S
to
s
>;*
a
a,
Sf
TO
TO'
*
^5
K> 8
K 1 ^
NJ Co
£ §
I
a,
3 §-
TO
nr\ S
H ©
3 §
2 §
o ^
H r?
hh Oq
H to
MJ
o
o
H
W
No. of
Health
Location,
Cohort
Exposure
Exposure
cases/
Effect Measure/
outcome
time period
description
assessment
measures
deaths
RR (95% CI)
Risk factors
Comments
Reference
Dental defects
Seveso, Italy,
65 subjects
Serum TCDD
Non-ABR
Dental defect %
Available: medical
Dose-response
Alaluusua
Dental
<9.5 years
(ng/kg) from
Zone
10/39
26%
history, age, sex,
pattern observed
et al., 2004
exams
old at time of
1976 samples
31-226
1/10
education, smoking
with dental defects
administered
Seveso
for those who
ng/kg
10%
in the ABR zone;
in 2001
explosion
resided in Zone
238-592
5/11
however, the control
among those
and residing
ABR; no serum
ng/kg
45%
population had a
exposed to
in zones
levels for non-
700-26000
9/15
much higher
TCDD in
ABR; 130
ABR residents
ng/kg
60%
prevalence of dental
1976
subjects
(unexposed).
defects (26%) than
recruited
TCDD
<5 years of
25/75
those in the lowest
from the
exposure
age at time
exposure group
non-ABR
represent levels
of accident
(10%).
region
as of 1976
Odds Ratios
(unexposed)
(after accident)
Non-ABR
(among those <5
Also assessed
Zone or
years of age at
hypodontia and
31-226
time of accident)
other dental and oral
ng/kg serum
1.0
aberrations, but
TCDD
these were too rare
238-26,000
to allow modeling
ng/kg serum
2.4(1.3-4.5)
by ABR zone.
TCDD
-------
Table 2-5. Epidemiology studies selected for TCDD noncancer dose-response modeling (continued)
o <5;
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
TO
TO'
*
^5
s
TO
>3
to
ON §
I
Si,
2 §-
TO
nr\ S
H ©
3 §
2 §
o ^
H r?
HH Oq
H TO
W|
o
o
H
W
No. of
Health
Location,
Cohort
Exposure
Exposure
cases/
Effect Measure/
outcome
time period
description
assessment
measures
deaths
RR (95% CI)
Risk factors
Comments
Reference
Menstraal
Seveso, Italy,
Women who
Serum TCDD
Interquartile
Lengthening of
Interview data:
Eskenazi et
cycle
follow-up
were <40
(ng/kg) from
range was
the menstrual
medical history,
al., 2002a
characteristics:
interview
years from
1976 samples.
64-322 ppt
cycle by 0.93
personal habits,
menstrual cycle
conducted in
zones A or B
TCDD
days (95% CI: -
work history,
length.
1996-1997 of
in 1976,
exposure level
TCDD
0.01, 1.86)
reproductive
women
A positive
was back-
examined as
history, age,
exposed to
association
extrapolated to
continuous
smoking, body
TCDD in the
found among
1976 using the
measure
mass index, alcohol
1976
women who
Filser or the
(per 10-fold
and coffee
accident
were pre-
first-order
increase in
consumption,
menarcheal
kinetic models.
serum
exercise, illness,
at the time of
levels).
abdominal
accident
surgeries.
(n=134)
-------
Table 2-6. Animal bioassays selected for cancer dose-response modeling
Species/strain
Sex
exposure
route/duration
n
Average daily
dose levels
(ng/kg-day)
Cancer types
Statistical significant tumors
(pairwise with controls or trend tests)
Reference
Rat/Sprague-
Dawley
Male/female
Oral-lifetime
feeding; 2 years
50 each
(86 each in
vehicle control
group)
0, 1, 10, or 100
Females: liver, lung,
oral cavity
Males: adrenal, oral
cavity, tongue
Adrenal cortex: adenoma
Liver: hepatocellular adenoma(s) or
carcinoma(s); hyperplastic nodules
Lung: keratinizing squamous cell carcinoma
Oral cavity: stratified squamous cell
carcinoma of hard palate or nasal turbinates
Tongue: stratified squamous cell carcinoma
Kociba et al.,
1978; Female
liver tumors
analysis
updated in
Goodman and
Sauer, 1992
Mouse/
B6C3F1
Male/female
Oral-gavage twice
per week; 104
weeks
50 each
(75 each in
vehicle control
group)
0, 1.4, 7.1,or71
for males;
0, 5.7, 28.6, or
286 for females
Females:
hematopoietic system,
liver, subcutaneous
tissue, thyroid
Males: liver, lung
Hematopoietic system: lymphoma or leukemia
Liver: hepatocellular adenoma or carcinoma
Lung: alveolar/bronchiolar adenoma or
carcinoma
Subcutaneous tissue: fibrosarcoma
Thyroid: follicular-cell adenoma
NTP, 1982a
Rat/Osborne-
Mendel
Male/female
Oral-gavage twice
per week; 104
weeks
50 each
(75 each in
vehicle control
group)
0, 1.4, 7.1,or71
Females: adrenal, liver,
subcutaneous tissue,
thyroid
Males: adrenal, liver,
thyroid
Adrenal: cortical adenoma, or carcinoma or
adenoma, NOS
Liver: neoplastic nodule or hepatocellular
carcinoma
Subcutaneous tissue: fibrosarcoma
Liver: neoplastic nodule or hepatocellular
carcinoma
Thyroid: follicular-cell adenoma or carcinoma
NTP, 1982a
Rat/Harlan
Sprague-
Dawley
Female
Oral-gavage
5 days per week;
2 years
53 or 54
0,2.14,7.14,
15.7,32.9, or
71.4
Liver
Lung
Oral mucosa
Pancreas
Liver: hepatocellular adenoma
Liver: cholangiocarcinoma
Lung: cystic keratinizing epithelioma
Oral mucosa: squamous cell carcinoma
Pancreas: adenoma or carcinoma
NTP, 2006
Mouse/Outbred
Swiss/H/Riop
Male
Gastric intubation
once per week; 1
year
43 or 44
(vehicle control
group = 38)
0, 1, 100, or
1,000
Liver
Liver: tumors
Toth et al.,
1979
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Reproductive toxicity studies
Monkey/
Rhesus
Daily dietary
exposure in
female
monkeys
(3.5-4 years)
F (F0, Fl,
F2, F3)
3 to 7 (Fl)
0,0.15, or 0.67
0.15
0.67
Reproductive
and
developmental
effects
Neurobehavioral effects
(e.g., discrimination-
reversal learning
affected)
Bowman et
al., 1989a, b
(and related
Schantz and
Bowman,
1989; Schantz
et al., 1986)
Mink
Daily dietary
exposure
(132 days)
F
12
0.03 (control),
0.8, 2.65, 9, or
70
None
2.65
Reproductive
effects
Reduced kit survival
Hochstein et
al., 2001
Rat/Holtzman
Corn oil
gavage (initial
loading dose
followed by
weekly dose
during mating,
pregnancy, and
lactation-
about
10 weeks)
F(F0)
F andM
(Fl and
F2)
12 (F0)
Not specified
(Fl andF2)
0 or 16.5
None
16.5
(maternal
exposure)
Reproductive
and
developmental
effects
Decreased development
of the ventral prostrate
(Fl), decreased sex ratio
(percentage of males)
(F2)
Ikeda et al.,
2005a
Mouse/ICR
Sesame oil
gavage (initial
loading dose
followed by
weekly doses
for 5 weeks)
M(F0)
42 or 43
0, 0.095, or
950
0.1
100
Reproductive
effects
Decreased male/female
sex ratio (percentage of
males) (Fl)
Ishihara et al.,
2007
-------
o ^
&
to
s
S
to
s
>;*
a
a,
Sf
^s
TO
TO'
*
^5
«
K> 8
K 1 ^
NJ Co
S §
I
a,
2 §•
C^ TO
J> ^5
nr\ S
H ©
3 §
2 §
o ^
H r?
<*^
hh Oq
H to
W|
o £?
o
H
W
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Reproductive toxicity studies (continued)
Rat/Wistar
albino
Olive oil
gavage (daily
for 45 days)
M
6
0, 1, 10, or 100
None
1
Reproductive
effects
Reduced sperm
production, decreased
reproductive organ
weights
Latchoumy-
candane and
Mathur, 2002
(and related
Latchoumy-
candane et al.,
2002a, b,
2003)
Rat/Sprague-
Dawley
Daily dietary
exposure
(3 generations)
F andM,
(FO)
F andM,
(F1 and
F2)
10-32 (FO)
22 (Fl)
28 (F2)
0, 1, 10, or 100
1
10
Reproductive
and
developmental
effects
Decrease in fertility,
decrease in the number
of live pups, decrease in
gestational survival;
decrease in postnatal
survival, decreased
postnatal body weight in
one or more generations
Murray et al.,
1979
Monkey/
Rhesus
Daily dietary
exposure
(4 years)
F
8
0,0.15, or 0.67
None
0.15
Reproductive
effects
Increased incidence of
endometriosis (disease
ranged from moderate to
severe)
Rier et al.,
1993, 1995
Rat/Sprague-
Dawley
Maternal corn
oil gavage
(weekly on
GD 14 and 21;
PND 7 and 14)
Offspring corn
oil gavage
(weekly for
11 months)
F (FO)
F (Fl)
3 (FO)
10 (Fl)
0,0.14,0.71,
7.14, or 28.6
0.14
0.71
Reproductive
effects
Decrease serum
estradiol levels (Fl)
Shi et al.,
2007
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Reproductive toxicity studies (continued)
Rhesus
monkey/
Cynomolgus
Fed gelatin
capsules
(5 days/week
for 12 months)
F
6 (treatment)
5 (controls)
0,0.71, 3.57,
or 17.86
17.86
None
Endometriosis
effects
Increased endometrial
implant survival,
increased maximum and
minimum implant
diameters, growth
regulatory cytokine
dysregulation
Yang et al.,
2000
Developmental toxicity studies
Rat/Harlan
Sprague-
Dawley
Corn oil
gavage (GD
10-16
F (F0)
80-88 (Fl)
0, 25, or 100
None
25
Developmental
effects
Decreased preference in
the consumption of
0.25% saccharin
solution (Fl)
Amin et al.,
2000
Rat/CRL:WI
(Han)
Maternal daily
dietary
exposure for an
estimated
20 weeks
(12 weeks
prior to mating
through
parturition)
F (F0)
M (Fl)
65 (F0
treatments)
75 (F0
controls) at
study
initiation;
following
interim
sacrifice
~30 animals
were allowed
to litter; Fl
onPND 21
was ~7
0,2.4, 8, or 46
None
2.4
(maternal
exposure)
Reproductive
and
developmental
effects
Delayed BPS (Fl)
Bell et al.,
2007a
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
l-H CfQ
H
W|
o
o
H
W
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Developmental toxicity studies (continued)
Rat/Sprague-
Dawley
Maternal corn
oil gavage
(GD 14 and
21; PND 7 and
14)
Offspring corn
oil gavage
(weekly for
8 months)
F (FO and
Fl)
2 or 3 (FO)
7 (Fl)
0,7.14, or 28.6
None
7.14
Developmental
effects
Decreased serum
estradiol levels (Fl)
Franczak et
al., 2006
Rat/Sprague-
Dawley
Maternal
single corn oil
gavage (GD 8)
Offspring
exposed during
gestation and
lactation
(35 days)
F (FO)
F andM
(Fl)
12 (FO)
50 or 60 (Fl)
0, 20, 60, or
180
None
20
(maternal
exposure)
Developmental
effects
Abrogation of sexually
dimorphic neuro-
behavioral responses
(Fl)
Hojo et al.,
2002 (and
related Zareba
et al., 2002)
Rat/
Han/Wistar
and Long-
Evans
Maternal
single corn oil
gavage
(GD 15)
F (FO)
F andM
(Fl)
4 to 8 (FO)
3F/3M per
treatment
group (Fl)
0, 30, 100,
300, or 1,000
None
30
(maternal
exposure)
Developmental
effects
Reduced mesiodistal
length of the lower third
molar (Fl)
Kattainen et
al., 2001
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Developmental toxicity studies (continued)
Mouse/
C57BL/6J,
BALB/cByJ,
A/J, CBA/J,
C3H/HeJ, and
C57BL/10J
Maternal
single corn oil
gavage
(GD 13)
F(F0)
F andM
(Fla, b, c)
Dams not
specified
(FO);
23-36 (Fla);
4-5 (Fib);
107-110
(Flc)
0, 10, 100, or
1,000
None
10
(maternal
exposure)
Developmental
effects
Variation in Ml
morphology in
C57BL/10J males and
females (Fla);
decreased mandible
shape and size in
C3H/HeJ males (Fib);
variation in molar shape
in C3H/HeJ males (Flc)
Keller et al.,
2007, 2008a, b
Mouse/ddY
Maternal olive
oil gavage
(weekly for
8 weeks prior
to mating)
F(F0)
M (Fl)
7 (FO)
3 (Fl
immuno-
cytochemical
analysis)
6 (Fl cell
number
count)
0, 0.7, or 70
None
0.7
(LOEL)
(maternal
exposure)
Neurotoxicity
Decreased serotonin-
immunoreactive neurons
in raphe nuclei of male
offspring (Fl)
Kuchiiwa et
al., 2002
Mouse/NIH
(pregnant and
pseudo-
pregnant)
Maternal
sesame oil
gavage daily
for 8 days
(GD 1-8)
F
10
0, 2, 50, or 100
None
2
Developmental
effects
Decreased progesterone
and increased serum
estradiol levels
Li et al., 2006
Rat/Holtzman
Maternal
single olive oil
gavage
(GD 18)
F (FO and
Fl)
4-7 (FO and
Fl)
0, 20, 60, or
180
None
20
(maternal
exposure)
Behavioral
effects
Decreased training
responses (Fl)
Markowski et
al., 2001
Rat/Line C
Maternal
single corn oil
gavage
(GD 15)
F (FO)
F andM
(Fl)
24-32
(treatment)
12-48
(controls)
0, 30, 100,
300, or 1,000
None
30
(maternal
exposure)
Developmental
effects
Increase in dental caries
(Fl)
Miettinen et
al., 2006
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Developmental toxicity studies (continued)
Rat/Holtzman
Maternal
single corn oil
gavage
(GD 15)
F (F0)
M (Fl)
Not specified
(F0)
5 males and
3 females
(Fl)
0, 12.5, 50,
200, or 800
800
(maternal
exposure)
None
Immunotoxicity
Decreased spleen
cellularity (Fl)
Nohara et al.,
2000
Rat/Holtzman
Maternal
single corn oil
gavage
(GD 15)
F (F0)
M (Fl)
6 (F0)
5 males and
3 females
(Fl)
0, 12.5, 50,
200, or 800
12.5
(maternal
exposure)
50
(maternal
exposure)
Developmental
effects
Decreased anogenital
distance (Fl)
Ohsako et al.,
2001
Rat/Harlan
Sprague-
Dawley
Maternal corn
oil gavage
(GD 10-16
F(F0)
~4 (F0);
80-88 (Fl)
0, 25, or 100
Not
determined
Not
determined
Developmental
effects
Facilitatory effect on
radial arm maze learning
(Fl)
Schantz et al.,
1996
Rat/Sprague-
Dawley
Maternal corn
oil gavage
(GD 10-16)
F andM
(Fl)
-15 (F0);
5-9 (Fl)
0, 25, or 100
25
100
Developmental
effects
Decreased thymus
weight
Seo et al.,
1995
Rat/TCDD-
resistant
Han/Wistar
bred with
TCDD-
sensitive
Long-Evans
Maternal corn
oil gavage
(GD 15)
F (F0)
M (Fl)
5-8 (F0)
0, 30, 100,
300, or 1,000
100
300
Reproductive
effects
Reduction in daily
sperm production and
cauda epididymal sperm
reserves
Simanainen et
al., 2004
Mouse/C57/6
NCji
Maternal
drinking water
exposure
(daily for
17-day
lactational
period)
F (F0)
F andM
(Fl)
8 (F0)
Not specified
(Fl)
0, 1.14, or 11.3
1.14
(NOEL)
(maternal
exposure)
11.3
(LOEL)
(maternal
exposure)
Immunotoxicity
Increased susceptibility
to Listeria (Fl males
and females); increase in
thymic CD4+ cells
(Fl males); decreased
spleen weight
(Fl males)
Sugita-
Konishi et al.,
2003
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
O
H r?
<*^
l-H CfQ
H
W|
o
o
H
W
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Acute toxicity studies
Mouse/B6C3Fl
Corn oil
gavage (single
exposure)
F
20
0, 1, 5, 10, 50,
100, or 6,000
5
10
Immunotoxicity
Increased mortality from
influenza infection
7 days after a single
TCDD exposure
Burleson et
al., 1996
Rat/Long-
Evans
Corn oil
gavage
(4 consecutive
days)
F
14, 6, 12, 6,
6, 6, 6, 6, 6,
and 4,
respectively
in control
and treated
groups
0,0.1,3, 10,
30, 100, 300,
1,000, 3,000,
or 10,000
30
100
Thyroid effects
Reduction in serum T4
levels
Crofton et al.,
2005
Rat/Sprague-
Dawley
Corn oil
gavage (single
dose)
F
4 (treated);
9 (control)
0, 0.6, 2, 4, 20,
60, 200, 600,
2,000, 5,000,
or 20,000
0.6
(NOEL)
2
(LOEL)
Enzyme
induction
Increased
benzo(a)pyrene
hydroxylase (BPH)
Kitchin and
Woods, 1979
Rat/Sprague-
Dawley
Corn oil dose
via oral gastric
intubation
(single dose)
F
10
0, 3, 10, 30,
100, 300,
1,000, 3,000,
10,000, or
30,000
3
10
Hormonal
effects
Increased serum FSH
Li et al., 1997
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Acute toxicity studies (continued)
Rat/Sprague-
Dawley
Corn oil
gavage or
TCDD-
contaminated
soil (single
dose)
F
6
0, 15, 40, 100,
200, 500,
1,000, 2,000,
or 5,000 in
corn oil
0, 15, 44, 100,
220, 500,
1,100, 2,000,
or 5,500 in
contaminated
soil
None
15
(LOEL)
Enzyme
induction
Induction of
arylhyrocarbon
hydroxylase (at low
dose in both treatment
protocols)
Lucier et al.,
1986
Mouse/
B6C3F1
(BALB/c
(C57BL/6N
(and DBA2
Corn oil
gavage (single
dose)
M, F
10-40
0, 5, 20, 100,
or 500
500
None
Mortality and
body weight
changes
No increased mortality
of virus-infected mice or
treatment-related
changes in body weight
Nohara et al.,
2002
Rat/TCDD-
resistant
Han/Wistar
bred; TCDD-
sensitive
Long-Evans
Corn oil
gavage (single
dose)
M, F
9-11
30-100,000
100
300
General
toxicological
endpoints, organ
weights, dental
defects
Reduction in serum T4
levels
Simanainen et
al., 2002
Rat/TCDD-
resistant
Han/Wistar
bred with
TCDD-
sensitive
Long-Evans
Corn oil
gavage (single
dose)
M, F
5-6
Line A:
30-3,000,000
Line B:
30-1,000,000
Line C:
30-100,000
100
300
General
toxicological
endpoints, organ
weights, dental
defects
Decreased thymus
weight
Simanainen et
al., 2003
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Acute toxicity studies (continued)
Mouse/
C57BL/6N
CYP1A2 (+/+)
wild-type
Corn oil
gavage (single
dose)
F
Not specified
0, 30, 100,
300, 1000,
3000, or
10,000
300
1,000
Immunotoxicity
Decreased antibody
response to SRBCs
Smialowicz et
al., 2004
Rat/Sprague-
Dawley
Corn oil
gavage (single
dose)
F
5-15
0,0.05,0.1, 1,
10, 100, 1,000,
or 10,000
0.1
(NOEL)
1
(LOEL)
Liver effects
Increase in hepatic
EROD activity and
CYP1A1 mRNA levels
Vanden
Heuvel et al.,
1994
Subchronic toxicity studies
Rat/Sprague-
Dawley
Corn oil
gavage (daily
for 28 days)
F
5
0, 2.5, 25, 250,
or 1,000
250
1,000
Body and organ
weight changes
Decreased body weight,
increased relative liver
weight and related
biochemical changes,
decreased relative
thymus weight
Chu et al.,
2001
Rat/Sprague-
Dawley
Corn oil
gavage (daily
for 28 days)
F
5
0, 2.5, 25, 250,
or 1,000
2.5
25
Liver effects
Alterations in thyroid,
thymus, and liver
histopathology
Chu et al.,
2007
Guinea pig/
Hartley
Daily dietary
exposure
(90 days)
M, F
10/sex
0,0.12,0.61,
4.9, or 26
(males); 0,
0.12,0.68,
4.86, or 31
(females)
0.61
4.9
Body and organ
weight changes
Decreased body weight
(male and females);
increased relative liver
weights (males);
decreased relative
thymus weight (males)
DeCaprio et
al., 1986
Mice/B6C3F1
Corn oil
gavage
(5 days/week
for 13 weeks)
F
5
0, 1.07, 3.21,
10.7, 32.1, or
107
None
1.07
(LOEL)
Body and organ
weight changes;
enzyme
induction
Increased EROD,
ACOH and
phosphotyrosyl proteins
at all doses
Devito et al.,
1994
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Subchronic toxicity studies (continued)
Rat/Tva:SIV
50-Sprague-
Dawley
Daily dietary
exposure
(13 weeks)
M, F
6
0, 20, 200, or
2,000
None
20
Liver effects
Reduced hepatic
vitamin A levels
Fattore et al.,
2000
Daily dietary
exposure
(13 weeks)
M, F
6
0 or 200
Daily dietary
exposure
(13 weeks)
M, F
6
0, 200, or
1,000
Daily dietary
exposure
(13 weeks, 26,
and 39 weeks)
F
6
0 or 100
Rat/Sprague-
Dawley
Gavage
loading/
maintenance
doses (every
4 days for
14 days)
M, F
6
0, 0.55, 307, or
1,607
0.57
327
Body and liver
weight changes;
hepatic cell
proliferation
Increased absolute and
relative liver weight
Fox et al.,
1993
Mouse/
B6C3F1
Corn oil
gavage
(5 days/week
for 13 weeks)
F
Not
specified
0,0.32, 1.07,
10.7, or 107
None
0.32
(LOEL)
Brain effects
Induction of biomarkers
of oxidative stress at all
doses
Hassoun et al.,
1998
Rat/Harlan
Sprague-
Dawley
Corn oil
gavage
(5 days/week
for 13 weeks)
F
6
0,2.14,7.14,
15.7,32.9, or
71.4
None
2.14
(LOEL)
Liver and brain
effects
Induction of biomarkers
of oxidative stress at all
doses in liver and brain
Hassoun et al.,
2000
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Subchronic toxicity studies (continued)
Rat/Harlan
Sprague-
Dawley
Corn oil
gavage
(5 days/week
for 13 weeks)
F
12
0,7.14, 15.7,
or 32.9
None
7.14
(LOEL)
Brain effects
Induction of biomarkers
of oxidative stress at all
doses
Hassoun et al.,
2003
Rat/Sprague-
Dawley
Corn oil
gavage
(5 days/week
for 13 weeks)
M, F
12
0,0.71,7.14,
71.4, or 714
7.14
71.4
Liver effects,
body weight
changes, and
hematologic and
clinical effects
Reduced body weight
and food consumption,
slight liver degeneration,
lymphoid depletion,
increased urinary
porphyrins and delta
aminolevulinic acid,
increased serum alkaline
phosphatase and
bilirubin
Kociba et al.,
1976
Rat/F344
Corn oil
gavage
(2 days/week
for 28 days)
F
3
0,0.71,7.14,
or71.4
None
0.71
(LOEL)
Clinical signs
and
histopathology
Decreased Cx32 plaque
number and area in the
liver
Mally and
Chipman,
2002
Mouse/
B6C3F1
Corn oil
gavage
(5 days/week
for 13 weeks)
F
Not specified
0,0.11,0.32,
1.07, 10.7, or
107.14
1.07
(NOEL)
10.7
(LOEL)
Liver, lung,
kidney, and
spleen effects
Increased hepatic
superoxide anion
Slezak et al.,
2000
Mouse/
B6C3F1
Corn oil
gavage
(5 days/week
for 13 weeks)
F
8-15
0, 1.07, 10.7,
107, or 321
None
1.07
Immunotoxicity
and organ
weight
Reduced antibody
response to SRBC,
increased relative liver
weight
Smialowicz et
al., 2008
Rat/Sprague-
Dawley
TCDD in diet
(13 weeks)
F
8
0, 14, 26, 47,
320, or 1,024
None
14
Multiple end-
points
Decreased absolute and
relative thymus weights,
decreased liver retinoid
levels
Van Birgelen
1995a, b
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Subchronic toxicity studies (continued)
Guinea pig/
Hartley
Corn oil
gavage
(weekly for
8 weeks)
F
10
0, 1.14, 5.71,
28.6, or 143
1.14
5.71
Immunotoxicity
Decreased total
leukocytes and
lymphocyte count,
decreased absolute
thymus and weight,
increase in primary
serum tetanus antitoxin
Vos et al.,
1973
Mouse/
B6C3F1
Corn oil
gavage (daily
for 14 days)
F
6-8
0, 10, 50, 100,
500, 1,000, or
2,000
None
10
Immunotoxicity
Reduction of serum
complement activity
White et al.,
1986
Chronic toxicity studies
Rat/CD-
COBS
Corn oil
gavage
(weekly for
45 weeks)
F
4
0, 1.43, 14.3,
or 143
None
1.43
Hepatic
porphyria
Increased urinary
porphyrin excretion
Cantoni et al.,
1981
Rat/Sprague-
Dawley
Loading/
maintenance
dose (every
3 days for
different
durations up to
128 days)
F
5
0,0.85,3.4,
13.6, 54.3, or
217
(28-day
duration)
54.3
(28-day
duration)
217
(28-day
duration)
Body weight
changes and
changes in
PEPCK activity
and IGF-I levels
Decreased body weight,
decreased PEPCK
activity, and reduced
IGF-I levels
Croutch et al.,
2005
Rat/Sprague-
Dawley
Corn oil
gavage
(5 days/week
for 30 weeks)
F
6
0,2.14,7.14,
15.7,32.9, or
71.4
None
2.14
(LOEL)
Brain effects
Induction of biomarkers
of oxidative stress at all
doses
Hassoun et al.,
2002
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Chronic toxicity studies (continued)
Rat/Sprague-
Dawley
Daily dietary
exposure
(2 years)
M, F
50
0, 1, 10, or 100
1
10
Multiple
endpoints
measured
Increased urinary
porphyrins,
hepatocellular nodules,
and focal alveolar
hyperplasia
Kociba et al.,
1978
Rat/Sprague-
Dawley
Biweekly
gavage
(30 weeks)
F
9
0,3.5, 10.7,
35, or 125
10.7
35
Body and organ
weight changes,
clinical
chemistry,
hepatocellular
proliferation
Increased relative liver
weight
Maronpot et
al., 1993
Mouse/
B6C3F1;
Rat/Osborne
Mendel
Corn oil
gavage
(2 days/week
for 104 weeks)
M, F
50
0, 1.4, 7.1, or
71 for rats and
male mice; 0,
5.7, 28.6, or
286 for female
mice
None
1.4
Liver and body
weight changes
Increased incidences of
liver lesions in mice
(males and females)
NTP, 1982
Rat/Sprague-
Dawley
Corn oil
gavage
(5 days/week
for 105 weeks)
F
53
0,2.14,7.14,
15.7,32.9, or
71.4
None
2.14
Liver and lung
effects
Increased absolute and
relative liver weights,
increased incidence of
hepatocellular
hypertrophy, increased
incidence of alveolar to
bronchiolar epithelial
metaplasia
NTP, 2006
Monkey/
Rhesus
Daily dietary
exposure
(4 years)
F
8
0,0.15, or 0.67
None
0.15
General
toxicological
endpoints and
reproductive
effects
Elevated serum
triglycerides and total
lipids
Rier et al.,
2001a, b
-------
Table 2-7. Animal bioassay studies selected for noncancer dose-response modeling (continued)
> |5*
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
TO
TO'
*
^5
Species/
strain
Exposure
protocol
Sex
(exposure
group)
n
Average daily
dose levels
(ng/kg-day)
NOAEL
(ng/kg-day)
LOAEL
(ng/kg-day)
Endpoint(s)
examined
LOAEL/NOAEL
Endpoint(s)
Reference
Chronic toxicity studies (continued)
Rat/Sprague-
Dawley
Biweekly
gavage
(30 weeks)
F
9
0,3.5, 10.7,
35, or 125
None
3.5
(LOEL)
EGFR kinetics
and auto-
phosphorylation,
hepatocellular
proliferation
Decrease in EGFR
maximum binding
capacity
Sewall et al.,
1993
Rat/Sprague-
Dawley
Biweekly
gavage
(30 weeks)
F
9
0,0.1,0.35, 1,
3.5, 10.7, 35,
or 125
10.7
35
Thyroid
function
Decreased serum T4
levels
Sewall et al.,
1995
Mouse/Swiss/
H/Riop
Sunflower oil
gavage
(weekly for
1 year)
M
38-44
0, 1, 100, or
1,000
None
1
Skin effects
Dermal amyloidosis and
skin lesions
Toth et al.,
1979
K> 8
K 1 ^
N) >>
U>
>3
0
1
a,
TO
>;
s
o
^•k
TO
O
s
2! »
O ^
H r?
hh Oq
H to
W|
o
o
H
ffl
o
o
ND = not determined.
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
Criteria
met?
No
Yes
Studies excluded
from quantitative
dose-response
analysis of TCDD
Studies Included in final list of
key cancer and noncancer
studies for quantitative dose-
response analysis of TCDD
Final development of two sets of TCDD study inclusion criteria:
For in vivo mammalian bioassays
For epidemiologic studies
Initial TCDD-specific study inclusion criteria
development for in vivo mammalian bioassays
Final literature collection (October, 2009)
Federal Register Notice; Web publication of literature
search for public comment and submissions
Dioxin workshop (2009) and expert refinement of
TCDD study inclusion criteria for in vivo mammalian bioassays
Literature search for in vivo mammalian bioassays and
epidemiologic TCDD studies (2000-2008)
Studies screened using TCDD study inclusion criteria:
Studies cited in 2003 Reassessment
Studies identified via literature search results
Studies submitted by the public
Studies collected by EPA in 2009
Figure 2-1. EPA's process to select and identify in vivo mammalian and
epidemiologic studies for use in the dose-response analysis of TCDD. EPA
first conducted a literature search to identify studies published since the 2003
Reassessment. Results were published and additional study submissions were accepted
from the public. Next EPA developed TCDD-specific study inclusion criteria for in vivo
mammalian studies and held a Dioxin Workshop where these criteria were discussed and
refined. Third, EPA developed two final sets of study inclusion criteria, one for in vivo
mammalian studies and another for epidemiologic studies. Finally, EPA applied these
two sets of criteria to all studies from the literature search, public submissions, 2003
Reassessment, and additional studies identified by EPA after the Dioxin Workshop
through October 2009. The studies that met these criteria formed a list of key studies for
EPA's consideration in TCDD dose-response assessment.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-232 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Most
elements of the five
considerations
^^^satisfied?
No
Yes
No
Study
in peer-reviewed
literature? _
Yes
Exposure
primarily to TCDD
and quantified?
No
Yes
Long-term
exposures and
latency information
available for cancer
^Xassessment?/
Exposure
windows and
latency information
available for RfD
^xassessmenf-V^
No
No
Yes
Yes
List of available epidemiologic studies on TCDD and DLCs
Key study included
for TCDD cancer
dose-response
assessment
Key study included
for TCDD noncancer
dose-response
assessment
Study excluded
from TCDD
dose-response
assessment
Evaluate study using five considerations:
• Methods used to ascertain health outcomes are clear and unbiased.
• Confounding and other potential sources of bias are addressed.
• Exposures based on individual-level estimates and uncertainties described.
• Statistical precision, power and study follow-up are sufficient.
• Exposure methods are described including exposure duration and latency.
Figure 2-2. EPA's process to evaluate available epidemiologic studies using
study inclusion criteria for use in the dose-response analysis of TCDD. EPA
applied its TCDD-specific epidemiologic study inclusion criteria to all studies published
on TCDD and DLCs. The studies were initially evaluated using five considerations
regarded as providing the most relevant kind of information needed for quantitative
human health risk analyses. For each study that satisfied most of these considerations
and was published in the peer-reviewed literature, EPA then examined whether the
exposures were primarily to TCDD and if the TCDD exposures could be quantified so
that dose-response analyses could be conducted. Finally, EPA required that the effective
dose and oral exposure be estimable: (1) for cancer, information is required on long-term
exposures, (2) for noncancer, information is required regarding the appropriate time
window of exposure that is relevant for a specific, nonfatal health endpoint, and (3) for
all endpoints, the latency period between TCDD exposure and the onset of the effect is
needed. Only studies meeting these criteria were included in EPA's TCDD
dose-response analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-233 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
Study
in peer-reviewed
literature?
Yes
No
x Lowest
dose tested for
cancer endpoint <1
s. ijg/kg-day? ^
Lowest dose
tested for noncancer
endpoint <30
ng/kg-day?
No
No
Yes
Yes
No
No
Yes
Were
elements of the four
considerations
^^satisfied?
Oral
exposure to TCDD
only with purity
^specified?
Yes^r^^
Study excluded
from TCDD
dose-response
assessment
List of available in vivo mammalian bioassay studies on TCDD
Key study included
for TCDD cancer and/or noncancer
dose-response assessment
Evaluate study further using four considerations:
Strain, gender, and age of test species is identified.
Testing protocol, including duration and timing of dosing is clear.
Study design is consistent with standard toxicological practices.
Magnitude of animal responses is outside range of normal variability.
Figure 2-3. EPA's process to evaluate available animal bioassay studies using study
inclusion criteria for use in the dose-response analysis of TCDD. EPA evaluated all
available in vivo mammalian bioassay studies on TCDD. Studies had to be published in
the peer-reviewed literature. Next, to ensure working in the low-dose range for TCDD
dose-response analysis, EPA applied dose requirements to the lowest tested average daily
doses in each study, with specific requirements for cancer (<1 |ig/kg-day) and noncancer
(<30 ng/kg-day) studies. Third, EPA required that the animals were exposed via the oral
route to only TCDD and that the purity of the TCDD was specified. Finally, the studies
were evaluated using four considerations regarded as providing the most relevant kind of
information needed for quantitative human health risk analyses from animal bioassay
data. Only studies meeting all of these criteria and considerations were included in
EPA's TCDD dose-response analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 2-234 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
3. THE USE OF TOXICOKINETICS IN THE DOSE-RESPONSE MODELING FOR
CANCER AND NONCANCER ENDPOINTS
A key recommendation from the National Academy of Sciences (NAS) for improving the
2003 Reassessment was that U.S. Environmental Protection Agency (EPA) should justify its
approaches to dose-response modeling for cancer and noncancer endpoints. Further, the NAS
suggested that EPA incorporate the most up-to-date and relevant state of the science for
2,3,7,8-tetrachl orodibenzo-/>dioxin (TCDD) dose-response assessment.
While EPA believes that at the time of its release, the 2003 Reassessment offered a
substantial improvement over the general state-of-the-science regarding dose-response modeling,
EPA agrees that the justification of the approaches to dose-response modeling can be improved
and the methodologies updated to reflect the most current EPA practices and science. In
Section 3, EPA describes the use of toxicokinetic (TK) information in the dose-response
modeling assessment of TCDD. Section 3.1 summarizes the NAS comments regarding the use
of TK in the dose-response approaches for TCDD. Section 3.2 overviews EPA's responses to
the NAS comments. Section 3.3 discusses TCDD kinetics, including TK models developed to
simulate disposition of this compound in rodents and humans (see Section 3.3.4), alternative
measures of dose that could be used in a TCDD dose-response analysis and uncertainties in the
TCDD dose estimates (see Section 3.3.5). Sections 4 and 5 of this document incorporate the TK
information into noncancer and cancer dose response modeling, respectively.
3.1. SUMMARY OF NAS COMMENTS ON THE USE OF TOXICOKINETICS IN
DOSE-RESPONSE MODELING APPROACHES FOR TCDD
The NAS commented on the appropriate use of TK models in dose-response modeling
for TCDD. Specifically, the committee requested that EPA consider using such models to
provide refined estimates of dose, for example, as the underlying science and predictive
capabilities of these models improved.
[Discussing Kinetic models],. .the committee encourages further development and
use of these models as data become available to validate and further develop them
(NAS, 2006a, p. 59).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Although the NAS basically agreed with EPA's use of body burden as a dose metric in
the 2003 Reassessment (e.g., see NAS, 2006a, p. 7), the NAS was concerned about the
limitations of first order kinetic models, such as the one used in the 2003 Reassessment, to
estimate TCDD body burdens.
TCDD, other dioxins, and DLCs act as potent inducers of CYP, a property that
can affect both the hepatic sequestration of these compounds and their half-lives.
Hepatic sequestration of dioxin may influence the quantitative extrapolation of the
rodent liver tumor results because the body-burden distribution pattern in highly
dosed rats would differ from the corresponding distribution in humans subject to
background levels of exposure. EPA should consider the possible quantitative
influence of dose-dependent toxicokinetics on the interpretation of animal
toxicological data (NAS, 2006a, p. 129).
The NAS also asked EPA to evaluate the impact of kinetic uncertainty and variability on
dose-response assessment. The NAS committee asked EPA to use TK models to examine both
interspecies and human interindividual differences in the disposition of TCDD, which would
better justify EPA dose-response modeling choices.
The Reassessment does not adequately consider the use of a PBPK model to
define species differences in tissue distribution in relation to total body burden for
either cancer or noncancer end points (NAS, 2006a, p. 62).
EPA ... should consider physiologically based pharmacokinetic modeling as a
means to adjust for differences in body fat composition and for other differences
between rodents and humans (NAS, 2006a, p. 10).
The Reassessment does not provide details about the magnitudes of the various
uncertainties surrounding the decisions EPA makes in relation to dose metrics
(e.g., the impact of species differences in percentage of body fat on the steady-
state concentrations present in nonadipose tissues). The committee recommends
that EPA use simple PBPK models to define the magnitude of any differences
between humans and rodents in the relationship between total body burden at
steady-state concentrations (as calculated from the intake, half-life,
bioavailability) and tissue concentrations. The same model could be used to
explore human variability in kinetics in relation to elimination half-life. EPA
should modify the estimated human equivalent intakes when necessary (NAS,
2006a, p. 73).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Finally, the NAS asked EPA to use TK considerations to better justify its choice of dose
metric.
EPA makes a number of assumptions about the appropriate dose metric and
mathematical functions to use in the Reassessment's dose-response analysis ...
but does not adequately comment on the extent to which each of these
assumptions could affect the resulting risk estimates.. EPA did not quantitatively
describe how this particular selection affected its estimates of exposure and
therefore provided no overall quantitative perspective on the relative importance
of the selection (NAS, 2006a, p. 51).
3.2. OVERVIEW OF EPA'S RESPONSE TO THE NAS COMMENTS ON THE USE OF
TOXICOKENTICS IN DOSE-RESPONSE MODELING APPROACHES FOR
TCDD
In response to the NAS recommendations regarding TCDD kinetics and choice of dose
metrics, this document presents an in depth evaluation of TCDD TK models, exploring their
differences and commonalities and their possible application for the derivation of dose metrics
relevant to TCDD. Initially, EPA discusses the application of first order kinetics to estimate
body burden as a dose metric for TCDD. This first order kinetic model is used to predict TCDD
body burden for all of the studies identified as Key Studies (see Section 2.4); this model uses a
constant half-life to simulate the elimination of TCDD from the body. However, given the
observed data indicating early influence of cytochrome P450 1A2 (CYP1A2) induction and
binding to TCDD in the liver and later redistribution of TCDD to fat tissue, the use of a constant
half-life for TCDD clearance following long term or chronic TCDD exposure is not biologically
supported. Therefore, using half-life estimates based on observed terminal steady state levels of
TCDD will not account for the possibility of an accelerated dose-dependent clearance of the
chemical during early stages following elevated TCDD exposures. The biological processes
leading to dose-dependent TCDD excretion are better described using physiologically based
pharmacokinetic (PBPK) models than by simple first order kinetic models. Additionally, as part
of its preparation for developing this document, EPA evaluated recent TCDD kinetic studies as
NAS advocated. Although the NAS agreed with continued use of body burden metric as the
dose metric of choice, EPA believes that the state-of-the-practice has advanced sufficiently to
justify the consideration of alternative dose metrics (other than administered dose) based on an
application of a physiologically-based TK model.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
EPA identified a number of advances in the overall scientific understanding of TCDD
disposition; many of these are documented in a summary discussion introducing the section on
TCDD kinetics (see Section 3.3). The increased understanding warranted an evaluation of
current kinetic modeling of TCDD to determine if the use of such models would improve the
dose-response assessment for TCDD. Justification of the final PBPK model choice is detailed in
Section 3.3. Through the choice of a published PBPK model to estimate dose metrics for dioxin,
EPA has addressed several of the NAS concerns. The PBPK model can be applied to estimate
dose metrics other than body burden that may be more directly related to response, e.g., tissue
levels, serum levels, blood concentrations, or dose metrics related to TCDD-protein receptor
binding. The selected PBPK model included explicit description of physiological and
biochemical parameters, therefore, it can also provide an excellent tool for investigating
differences in species uptake and disposition of TCDD. One of the criteria used to select a
PBPK model for TCDD kinetics was the availability of both human and animal models so that
differences in species uptake and disposition of TCDD can be investigated. Additionally, the
PBPK model includes quantitative information that is suitable for addressing the impact of
physiological (e.g., body weight [BW] or fat tissue volume), or biochemical (e.g., induction of
CYP1A2) variability on overall risk of TCDD between species, in response to another area of
concern in the NAS report. The sensitivity analysis and uncertainty in dose metrics derived for
the risk assessment of TCDD are also presented in Section 3.3. Detailed discussion on the
uncertainty in choice of PBPK model-driven dose metrics is also provided in Section 3.3.
3.3. PHARMACOKINETICS (PK) AND PK MODELING
3.3.1. PK Data and Models in TCDD Dose-Response Modeling: Overview and Scope
In general, the use of measures of internal dose in dose-response modeling is considered
to be superior to that of administered dose (or uptake) because the former is more closely related
to the response. The evaluation of internal dose, or dose metric, in exposed humans and other
animals is facilitated by an understanding of pharmacokinetics (i.e., absorption, distribution,
metabolism, and excretion). When measurements of internal dose (e.g., blood concentration,
tissue concentration) are not available in animals and humans, pharmacokinetic models can be
used to estimate them. The available data on the pharmacokinetics of TCDD in animals and
humans have been reviewed (Van Birgelen and Van den Berg, 2000; U.S. EPA, 2003; NRC,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
2006). It is evident based on these reviews and other analyses that three distinctive features of
TCDD play important roles in determining its pharmacokinetic behavior, as discussed below:
TCDD is very highly lipophilic and thus is more soluble in fat or other relatively nonpolar
organic media than in water. The //-octanol/water partition coefficient is a commonly-
used measure of lipophilicity equal to the equilibrium ratio of a substance's concentration
in //-octanol (a surrogate for biotic lipid) to the substance's concentration in water (Leo
et al., 1971). For TCDD, this coefficient is on the order of 10,000,000 or more (ATSDR,
1998). It follows that the solubility of TCDD in the body's lipid fraction, i.e., the fatty
portions of various tissues, including adipose, organs, and blood, is extremely high.
TCDD is very slowly metabolized compared to many other organic compounds, with an
elimination half life in humans on the order of years following an initial period of
distribution in the body (Carrier et al., 1995a; Michalek et al., 2002). Most laboratory
animals used for toxicologic testing tend to eliminate TCDD much more quickly than
people, although even in animals TCDD is eliminated much more slowly than most other
chemicals.
¦ TCDD induces binding proteins in the liver that have the effect of sequestering some
of the TCDD. The ability of TCDD to alter gene expression and the demonstration that
the induction of CYP1A2 is responsible for hepatic TCDD sequestration suggest that
both pharmacokinetic and pharmacodynamic events must be incorporated for a
quantitative description of TCDD disposition (Santostefano et al., 1998). The induction
of these proteins implies that TCDD tends to be eliminated more rapidly in the early
years following short-term, high-level exposures than it is after those initial levels have
declined. Recent efforts of pharmacokinetic modeling have supported the concentration-
dependent elimination of TCDD in animals and humans (Aylward et al., 2005b; Emond
et al., 2006).
Sections 3.3.2 and 3.3.3 of this section present the salient features of TCDD
pharmacokinetics in animals and humans, with particular focus on mechanisms and data of
relevance to interspecies and intraspecies variability. Section 3.3.4 describes the various dose
metrics for the dose-response modeling of TCDD and the characteristics of pharmacokinetic
models potentially useful for estimating these metrics. Finally, Sections 3.3.5 and 3.3.6
summarize the results of application of pharmacokinetic models to derive dose metrics as well as
the uncertainty associated with the predictions of dose metrics used in dose-response modeling.
Dose metrics derived via PBPK modeling approaches are utilized in Sections 4 and 5 of this
document for noncancer and cancer TCDD dose-response modeling, respectively.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
3.3.2. PK of TCDD in Animals and Humans
3.3.2.1. Absorption and Bioavailability
When administered via the oral route in the dissolved form, TCDD appears to be well
absorbed. Animal studies indicate that oral exposure to TCDD in the diet or in an oil vehicle
results in the absorption of >50% of the administered dose (Nolan et al., 1979; Olson et al.,
1980). Human data from Poiger and Schlatter (1986) indicate that >87% of the oral dose (after
ingestion of 105 ng [3H]-2,3,7,8-TCDD [1.14 ng/kg BW] in 6 mL corn oil) was absorbed from
the gastrointestinal tract. Lakshmanan et al. (1986), investigating the oral absorption of TCDD,
suggested that it is absorbed primarily by the lymphatic route and transported predominantly by
chylomicrons.
Oral absorption is generally less efficient when TCDD is more tightly bound in soil
matrices. Based on experiments in miniature swine, Wittsiepe et al. (2007) reported an
approximately 70% reduction in bioavailability when TCDD was administered in the form of
contaminated soil, relative to TCDD after extraction from the same soil matrix with solvents.
Working with soil from the prominent contamination site at Times Beach, Missouri, Shu et al.
(1988) reported an oral bioavailability of approximately 43% based on experiments in rats.
Percent dose absorbed by the dermal route is reported to be less than the oral route, whereas
absorption of TCDD by the transpulmonary route appears to be efficient (see, for example,
Banks et al., 1990; Banks and Birnbaum, 1991; Nessel et al., 1992; Diliberto et al., 1996;
U.S. EPA, 2003; Roy et al., 2008).
3.3.2.2. Distribution
TCDD in systemic circulation equilibrates and moves into the tissues where it is then
accumulated, bound, or eliminated. Whereas the bulk of the body tissues are expected to
equilibrate in a matter of hours, the adipose tissue will approach equilibrium concentrations with
blood much more slowly. Consistent with these assertions, a number of experimental and
modeling studies in rats and humans have shown that TCDD has a large volume of distribution
(Vd), i.e., the apparent volume in which it is distributed. The Vd corresponds to the volume of
blood plus the product of internal tissue volumes and the corresponding tissue:blood partition
coefficients. This parameter is a key determinant of the elimination rate of TCDD in exposed
organisms. The tissue:blood partition coefficients of TCDD, in turn, are determined by the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
relative solubility of TCDD in tissue and blood components (including neutral lipids,
phospholipids, and water).
Column 1 in Table 3-1 presents the tissue:blood partition coefficients for TCDD (Wang
et al., 1997; Emond et al., 2005). Column 3 of this table lists the physical volume of each tissue,
scaled to a person weighing 60 kg. The last column shows the implications of the tissue volumes
and tissue:blood partition coefficients for the effective volumes of distribution for each tissue
and for the body as a whole. It can be seen that, purely on the basis of solubility space, the fat
should be expected to contain about 94% of the TCDD in the body, and that the body as a whole
behaves as if it is about 1,200 liters in terms of blood-equivalents (i.e., approximately 22-fold
larger than its physical volume).
Maruyama et al. (2002) have published another set of tissue/blood partition coefficients
for TCDD and other dioxin congeners based in part on observations of tissue concentrations
measured in autopsy specimens from eight Japanese people without known unusual exposures to
TCDD. Their estimates of TCDD partition coefficients seem to be rather large and variable,
with a fat:blood value of 247 ± 78 (standard deviation [SD]), a livenblood value of 9.8 ± 5.7 and
a muscle:blood value of 18 ± 10.6. Depending on time of autopsy, tissue samples may not be an
accurate source of information on observed, in vivo partition coefficients because weight loss is
likely to occur pre and post mortem. In particular, a decline in fat stores volume could lead to an
increased concentration of dioxin in fat in autopsy specimens relative to what would be observed
in vivo.
The calculations shown in Table 3-1 do not include the additional amount that will be
bound to induced proteins in the liver. That induction and binding will tend to increase the
contribution of the liver on the effective volume of distribution (Birnbaum, 1986).
It is also of interest to point out some basic implications of the data in Table 3-1 for the
expected rates of perfusion-mediated transfer of TCDD between blood and each of the
organ/tissues. The rate of loss from a tissue (occurring primarily via blood flow) and the
corresponding half-life can be calculated using the following equations:
Rate constant for loss (hour"') = Blood flow (hters / hour) (Eq 3.,,
Tissue volume (liters) x Tissue/Blood Partition Coefficent
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
t1/2 for tissue perfusion loss = —
Rate constant for loss
(Eq. 3-2)
_ ln(2) x Tissue volume (liters) x Tissue/Blood Partition Coefficent
Blood flow (liters/hour)
Because TCDD is highly lipophilic, its concentration in the aqueous portion of the blood
is very small, and TCDD tends to partition from blood components into cellular membranes and
tissues, probably in large part via diffusion. As a result, full equilibrium concentrations of
TCDD are not attained by the end of the transit time through organs from the arterial to venous
blood. For organs in which this occurs, diffusion coefficients or "permeability factors" have
been estimated to assess the fractional attainment of equilibrium concentration that occurs by the
time the blood leaving each organ reaches the venous circulation. Table 3-2 presents the
permeability factors and implications for perfusion half-lives for TCDD, per Emond et al. (2005,
2006).
Despite the high lipid bioconcentration potential of TCDD, it does not always occur at
the highest concentration in the adipose tissue (Poiger and Schlatter, 1986; Geyer et al., 1986;
Abraham et al., 1988). Further, the ratios of tissue:tissue concentrations of TCDD and related
compounds (e.g., the livenadipose ratio) may not remain constant during nonsteady-state
conditions. TCDD concentrations have been observed to decrease more rapidly in the liver than
in adipose tissue. For example, Abraham et al. (1988) found that the livenadipose tissue
concentration ratio in female Wistar rats exposed to a subcutaneous TCDD dose of 300 ng/kg
decreased from 10.3 at 1 day postexposure to 0.5 at 91 days postexposure. It should be noted
that even at a ratio of 0.5, the amount of TCDD in the liver is greater than that based on lipid
content of the tissue alone, consistent with the presence of hepatic TCDD binding proteins. The
liver/adipose tissue ratio also was dose-dependent, such that the liver TCDD burden increased
from ~11% of the administered dose at low doses (i.e., 1-10 ng/kg) to -37% of the dose at an
exposure level of 300 ng/kg. The increase in TCDD levels in liver, accompanied by a decrease
in concentration in the adipose tissue, is a particular behavior to be considered in high dose to
low dose extrapolations. This behavior is essentially a result of dose-dependent hepatic
processes, as described below.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
3.3.2.3. Metabolism and Protein Binding
The metabolism of TCDD is slow, particularly in humans, and it is thought to be
mediated by the CYP1A2 enzyme that is inducible by TCDD (Ramsey et al., 1982; Wendling et
al., 1990; Olson et al., 1994; Weber et al., 1997). The low rate of metabolism in combination
with sequestration appear to account for the retention of TCDD in liver, and these processes
collectively contribute to the long half-life for elimination of TCDD from the body.
Dynamic changes in TCDD binding in liver and partitioning to fat have been studied
extensively in rats and mice (Diliberto et al., 1995, 2001). Figure 3-1 shows observations by
Diliberto et al. (1995) of the ratio of liver concentrations to adipose tissue concentrations for
mice given doses spread over a 100-fold range and studied at four different times following
exposure. It can be seen that even for the lowest dose studied the livenfat concentration ratio is
higher than would be expected based on the lipid contents of the tissues (i.e., 0.06:1,
corresponding to the ratio of human liver:blood and fat:blood partition coefficients; see
Table 3-1). Moreover, the relative concentration in the liver consistently rises with dose, with
the steepest rise observed during the first two weeks after dosing. If the distribution of TCDD
were governed solely by passive partitioning into fat, there should be no such change in relative
concentrations with dose. However, data presented in Figure 3-1 illustrate that at longer time
points, the ratio of TCDD in the liver to TCDD in fat decreases, indicating that a redistribution of
the chemical occurs as time goes on for each applied dose. The redistribution of TCDD tissue
levels from liver to fat with increasing time suggests that binding of the chemical in the liver
(including via induction of CYP1A2) is an important kinetic consideration at early exposure
points with relatively high applied doses. At steady state levels (longer than 35 days, and low
applied doses), there seems to be a tendency for TCDD to redistribute to fat tissue.
Experiments with CYP1A2 "knock-out" mice (i.e., congenic strains differing in only a
single gene that is "knocked out" in one of the strains) indicate that the inducible binding of
TCDD is attributable to CYP1A2 (Diliberto et al., 1997, 1999). As noted previously, this
enzyme is believed to make an important contribution to metabolism of TCDD. Given the
critical role of CYP1A2 induction in the kinetics of TCDD, dose-and time-dependent induction
of this protein in rats has been examined and modeled (Wang et al., 1997; Santostefano et al.,
1998; Emond et al., 2004, 2006). Accordingly, the amount of CYP1A2 in the liver can be
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
computed as the time-integrated product of inducible production and a simple first-order loss
process (Wang et al., 1997):
dCYP-
dt
2A1 = S(t)K0 - K2CA2t (Eq. 3-3)
where CYP2ai is the concentration of the enzyme, K2 is the rate constant for the first order loss,
Ca2i is the concentration of CYP1A2 in the liver, K{) is the basal rate of production of CYP1A2 in
the liver, and S(t) is a multiplicative stimulation factor for CYP1A2 production in the form of a
Hill-type function:
^nA2 (^Ah-TCDD )
(IC A2f+(CAh_TCDD)h
S(t) = 1 + h rCDD „ (Eq. 3-4)
where ICa2 corresponds to the concentration of the aryl hydrocarbon (Ah)-TCDD complex at
which half of the maximum fold stimulation of CYP2A production is reached, and h, the Hill
exponent, determines the curvature of the stimulation in relation to concentration of the
Ah-TCDD complex at relatively low doses. A value of 0.6 as the Hill exponent has been used by
Wang et al. (1997, 2000) and Emond et al. (2004, 2005, 2006), indicative of a negative
cooperation, i.e., the curve is convex-upward (supralinear), depicting a faster increase in the low-
dose region compared to a straight line. Additional parameters in this expression include IriA2,
the maximum fold increase in the CYP1A2 synthesis rate over the basal rate that can occur at
high levels of TCDD, and (Cah-tcdd), the concentration of TCDD bound to the aryl hydrocarbon
receptor (AhR). This concentration in turn depends on the concentration of TCDD in the liver
(Cnf), the concentration of the AhR (Ah/ ,) in liver, and the dissociation constant for the
Ah-TCDD receptor complex, KDAh-
Ahr. x CTif
C,™ = „ * (Eq. 3-5)
^DAh + Lif
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
3.3.2.4. Elimination
Elimination half-lives (i.e., the time taken for the concentration to be reduced to one-half
of its initial level) of TCDD range from 11 days in the hamster to 2,120 days in humans
(U.S. EPA, 2003). Hepatic metabolism and binding processes, fecal excretion, and accumulation
in adipose tissue collectively determine the dose-dependent elimination half-lives in various
species. Aylward et al. (2005a) depicted the relationship between the elimination rate versus
initial level of lipid-corrected TCDD in serum for 36 people (see Figure 3-2). Even though this
analysis was done using the initial TCDD level, rather than the geometric mean or midpoint level
in the decline for each person, it indicated a concentration-dependency of the half-life and
elimination of TCDD in exposed individuals.
3.3.2.5. Interspecies Differences and Similarities
Among the pharmacokinetic determinants of TCDD, some are known to vary markedly
between species whereas others are not characterized sufficiently in this regard. Overall, the
qualitative determinants of the body burden and elimination half-lives appear to be similar across
species. Based on empirical observations for TCDD as well as with other PCDFs, Carrier et al.
(1995a,b) argued that in rats, monkeys, and humans, the dose-dependent changes in the fraction
contained in liver and adipose tissue follow a similar pattern across species. The authors
suggested that the half-saturation body burden is around 100 ng/kg and the plateau of liver dose
(as fraction of body burden) appears to occur around 1,000 ng/kg. Literature also indicates that
AhR is conserved phylogenetically (Nebert et al., 1991; Fujii-Kuriyama et al., 1995; Harper
et al., 2002) and is present in mammalian species, including experimental animals and humans
(Roberts et al., 1985, 1986; Manchester et al., 1987; Lorenzen and Okey, 1991; Okey et al.,
1994). These qualitative similarities in pharmacokinetic determinants and outcome support the
use of animal data to infer general patterns of the pharmacokinetic behavior of TCDD in humans.
However, quantitative differences in determinants, including physiological, physicochemical,
and biochemical, need to be taken into account. Even though species-specific physiological
parameters can be obtained from the literature, key data on species-specific biochemical
parameters (particularly binding constants, maximal capacity, induction rates, and other
parameters) are not available for humans at this time. However, these can be inferred by using a
pharmacokinetic model fit to in vivo data on the rate of TCDD elimination from specific
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_i i DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
compartments in humans (Carrier et al., 1995a, b; Emond et al., 2004, 2005, 2006; Aylward
et al., 2005b).
3.3.3. PK of TCDD in Humans: Interindividual Variability
The pharmacokinetics and tissue dose of TCDD in humans can exhibit variability within
a population as a function of the interindividual variability in the key determinants. If a chronic,
lifetime exposure is considered to be the relevant scenario for developing guidance for human
exposure standards, the key determinants of concern are clearance, binding, and temporal
changes in volume of distribution. When considering the interindividual variability in
pharmacokinetics and dose metrics of TCDD, it is important to recognize that higher lipid-
corrected serum concentrations in highly exposed persons are associated with greater elimination
rates, probably due to greater degrees of induction of CYP1A2 in the liver and possibly other
related metabolic enzymes (Grassman et al., 2000; Abraham et al., 2002; Aylward et al., 2005b;
Emond et al., 2006).
The interindividual variability in fat content is a critical parameter given the
pharmacokinetic characteristics of TCDD (see Section 3.3.2). Both metabolic elimination and
elimination via the GI tract depend on the fraction of TCDD in the body that is available outside
of adipose tissue. As body fat content rises, a smaller portion of the total body TCDD will be
contained in the relatively available fraction outside of fat. Because elimination of TCDD by
both metabolism and fecal excretion depends on the small proportion of TCDD that exists
outside of fat tissue, people with larger proportions of body fat—including many older people—
will tend to require longer times to reduce TCDD levels by a given proportion than leaner people
(Van der Molen et al., 1996, 1998; Rohde et al., 1999; Emond et al., 2006).
The sections that follow highlight key aspects of interindividual variability in TCDD
pharmacokinetics, with an emphasis on the available data related to elimination half-lives and
volume of distribution.
3.3.3.1. Life Stage and Gender
The influence of the variability of fat content in human population on the distribution and
clearance of TCDD has been evaluated by several investigators. Figure 3-3 shows the results of
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_i2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
a rare experiment in which TCDD elimination via feces was measured in six people in relation to
their body fat content (Rohde et al., 1999).
There are data which suggestive an inverse dependency of half life on percent body fat.
Observations of TCDD elimination rates in a small number of men and women in the Seveso
cohort provide a modest opportunity to compare TCDD elimination rates on this basis. Based on
the partition coefficients reported by Emond et al. (2006), the elimination rates for the men in the
sampled group are expected to be greater than the elimination rates in the women. Based on
calculations similar to those shown in Table 3-2, and fat proportions inferred from body mass
indices using the equations of Lean et al. (1996), the Seveso men studied are expected to have an
overall average of about 3.92% of their TCDD body burden outside of fat, whereas the women
are expected to have an average of only 2.36% outside of fat. On this basis, the TCDD
elimination rates in the men are expected to be 3.92/2.36 = 1.66 times faster than the elimination
rates in the women. By comparison, Michalek et al. (2002) reported observed elimination rates
in men and women that result in a slightly lower ratio:
men:0. Ill year1 ±0.010 (std.error) , ^
— = 1.56 (Eq. 3-6)
women :0.071 year ± 0.010 (std. error)
The central estimates for the elimination rates correspond to half lives of 6.5 and 9.6 years for
men and women, respectively.
A further point of comparison can be derived using the observed body mass index (BMI)8
and TCDD elimination rate of each of the male Ranch Hand military veterans, whose TCDD
elimination rates were observed between 9 and 33 years after their time in Vietnam. The average
BMI over that time was 29.44 (based on 287 measurements for the 97 veterans, tabulated in three
periods by Michalek et al., 2002), and their average age was about 44.5 for the measurements.
Based on these data, the corresponding average estimated percent body fat is 29.7% using the
Lean et al. (1996) formula for men. The observed average TCDD elimination rate constant for
these men for the period was 0.092 year-1 ± 0.004 (standard error), corresponding to a half life of
7.5 years. This half life is slightly longer than the central estimate of the half life of 6.2 years
8The body mass index, or BMI, is calculated as the body weight in kilograms divided by the square of the height in
meters.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_i3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
(i.e., ln(2)/0. Ill) for the smaller group of Seveso males with their slightly smaller estimated
percent body fat. Figure 3-4 shows a simple plot of these data and a fitted unweighted regression
line characterizing the relationship between estimated fat content and TCDD elimination rates.
Variation in metabolic enzyme activities and other routes of loss is also likely to be important,
but there is little human quantitative information available on these issues.
More recently, Kerger et al. (2006) estimated the slope of the relationship between half-
life and age to be 0.12 years (95% confidence interval, 0.10-0.14), which corresponds to the rate
of increase in TCDD half-life for each year of age. The authors speculated that although age
explained most of the variance in the individual half-life trends, it was also correlated with
TCDD concentration, BMI, and body fat mass. The regression model developed by these
authors discriminated between the high and low TCDD exposures or concentrations. Thus, after
accounting for the TCDD (concentration x age) term's effect on the slope of age, the final model
for TCDD concentration < 700 ppt was
where t\n is the half-life and Age is the age at time of subsequent sampling. Pharmacokinetic
information relevant to specific age groups is presented in the sections that follow.
3.3.3.1.1. Prenatal period.
Data to estimate TCDD elimination rates for fetuses are not available. Levels of TCDD
in fetal tissues for rats were experimentally estimated at different gestational periods and utilized
in a developmental model by Emond et al. (2004). There is information on body composition
that is relevant to prediction of TCDD dose to fetus. These data, summarized as part of the
radiation dosimetry model of the International Commission on Radiological Protection, are
consistent with the idea that early fetuses are nearly all water and less than 1% lipid, and lipid
levels rise toward parity with protein near the time of normal delivery.
t\/i = 0.35 + 0.12 x Age
For TCDD concentration >700 ppt, the final model was:
(Eq. 3-7)
t\ / 2 = 0.35 + 0.088 x Age
(Eq. 3-8)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_i4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Bell et al. (2007b) reported that the disposition of TCDD into the fetus shows dose
dependency, with a greater proportion of the dose reaching the fetus at lower doses of TCDD.
Further, both CYP1A1 and CYP1A2 are highly inducible (~103-fold) in fetal liver, whereas
CYP1A2 shows much lower induction (10-fold) in maternal liver. It has been speculated that
this is due to the lower basal levels of CYP1A2 in fetal liver, as compared to maternal liver
(Bell et al., 2007b).
3.3.3.1.2. Infancy and childhood.
Hattis et al. (2003) describe the general pattern of change of body fat content with age in
children. Central tendency values for percent body fat begin at about 12% at birth and rise
steeply to reach about 26% near the middle of the first year of life. Fat content then falls to reach
a minimum of approximately 15% at 5-8 years of age, followed by a sex-dependent "adiposity
rebound" that takes females to about 26% body fat while the males remain near 16—17% on
average by age 20. The interindividual variability distributions about these central values are
complex, as some children experience the "adiposity rebound" earlier than others, and this
creates patterns that are not simply interpretable as unimodal normal distributions. Hattis et al.
(2003) did find it possible to fit distributions of body fat content inferred from NHANES skin
fold measures to mixtures of two normal distributions for children between age 5 and 18.
At least two groups of authors have published PBPK modeling results indicating
generally more rapid clearance of TCDD in children than in adults, a trend that is consistent with
the generally lower fat content of children (Kreuzer et al., 1997; Van der Molen et al., 2000;
Leung et al., 2006). The rapid expansion of the adipose tissue compartment can contribute, in
part, to the reduced apparent half-life in children (Clewell et al., 2004). This reduction may also
be due to varying rates of metabolism and/or fecal lipid excretion (Abraham et al., 1996;
Kerger et al., 2007).
Furthermore, very young children have different modes and quantities of exposure
compared to adults. Lakind et al. (2000) characterize distributions of milk intake for nursing
infants to characterize distributions of TCDD exposure. This is also a corresponding route of
loss of TCDD stores for lactating women, as described in Section 3.3.3.2 below.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
3.3.3.1.3. Adulthood and old age.
The fraction of fat in relation to body weight in adulthood and old age can be computed
as a function of the BMI and age (e.g., Lean et al., 1996):
% Body Fat (males) = 1.33 x BMI + 0.236 x Age - 20.2 (Eq. 3-9)
% Body Fat (females) = 1.21 x BMI + 0.262 x Age - 6.7 (Eq. 3-10)
The above equations are the result of analysis of data based on underwater weighing of
63 men and 84 women (age range 16.8-65.4). The salient observation with respect to TCDD for
these data is that age and BMI-dependent variability in fat content have implications for the
variability in TCDD elimination rates and internal dose among adults.
3.3.3.2. Physiological States: Pregnancy and Lactation
Data on body fat content in pregnant women at various stages of gestation (Pipe et al.,
1979) have potential implications for TCDD elimination rates during pregnancy, even though the
relationship between these parameters has not been formally analyzed.
Lactation is viewed as an additional route of elimination for some chemicals such as
TCDD. According to a recent study, a breast-feeding woman expels through lactation an
estimated 8.76 kg fat per year [^/(kg/day), 0.8 kg milk/day with an average 3% lipid], and the
partition coefficient between blood lipid and milk fat (Kbm) for TCDD is 0.92 (Wittsiepe et al.,
2007; Milbrath et al., 2009). The estimated rate of elimination of TCDD due to breast-feeding
(kbfed) can then be computed as follows (Milbrath et al., 2009):
k -
bfed
qf x At,
bfed
K xPM.xBw
100
(Eq. 3-11)
where
A^bfed (unitless) = the fraction of the year during which the woman was actively breast-
feeding;
pbf = woman's percent body fat; and
BW = woman's body weight in kg.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Assuming no interaction between breast-feeding and other half-life determinants
Milbrath et al. (2009), the authors predicted a half-life of 4.3 years for TCDD in a 30-year-old,
nonsmoking woman with 30% body fat if she did not breast-feed that year, and a half-life of
1.8 years if she breast-fed for 6 months.
3.3.3.3. Lifestyle and Habits
One of the factors related to lifestyle and habits that could influence TCDD kinetics is
smoking. Smoking has been reported to enhance the elimination of dioxin and dioxin-like
compounds (Flesch-Janys et al., 1996; Ferriby et al., 2007). Milbrath et al. (2009) accounted for
interindividual variation in body composition as well as smoking habits in an empirical model.
The predicted half-life (years) for an individual / as a function of age, smoking status, and
percent body fat i was as follows
t\i2 (age, smoke, p¥X = W{0age) + P(age) x age, ]x SFt x ^ '— (Eq. 3-12)
PbJrtf(aget)
where
P(Stage)
P (age)
aget
Pbf
Pty"ref( aget)
SFt
= intercept constant derived from regressed data;
= slope constant derived from regressed data;
= specific age i (years);
= individual percent body fat;
= reference percent body fat; and
= the unitless, multiplicative smoking factor.
3.3.3.4. Genetic Traits and Polymorphism
One particular genetic locus that is potentially related to TCDD pharmacokinetics and
tissue dose is the gene for the AhR. Eight candidate AhR polymorphisms have been identified to
date (Harper et al., 2002; Connor and Aylward, 2006). Given the role of AhR in regulating the
induction of CYP1 isozymes (Baron et al., 1998; Toide et al., 2003; Connor and Aylward, 2006),
the polymorphism might lead to interindividual differences in metabolic clearance, the
significance of which would depend upon the dose, fat content, and exposure scenario. In this
regard, it should be noted that the inducibility of aromatic hydrocarbon hydroxylase in human
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
tissues has been reported to be highly variable, up to 100-fold (Wong et al., 1986; Smart and
Daly, 2000; Connor and Aylward, 2006).
Finally, the scientific literature contains values of K& (the dissociation constant of the
TCDD-AhR complex) ranging from about 1 to much higher values (corresponding to lower
binding affinity) (reviewed in Connor and Aylward, 2006). This provides suggestive evidence
for a heterogeneous human AhR, with functionally important polymorphisms (Roberts et al.,
1986; Micka et al., 1997), even though some of the range may be attributed to experimental
procedural differences and to other factors (Manchester et al., 1987; Lorenzen and Okey, 1991;
Harper et al., 2002; Connor and Aylmard, 2006).
The various pharmacokinetic processes and determinants (see Sections 3.3.2 and 3.3.3),
individually or together, might influence the dose metrics of relevance to the dose-response
modeling of TCDD.
3.3.4. Dose Metrics and Pharmacokinetic Models for TCDD
3.3.4.1. Alternative Dose Metrics for Dose-Response Modeling
The dose metric related to a toxicologic endpoint can range from the maximal
concentration, the area under a time-course curve (AUC), or the time-averaged concentration of
the toxic moiety in the body, blood, or target tissue, to an appropriate measure of the resulting
interactions in the target tissue (e.g., receptor occupancy or functional biomarkers related to
specific effects). A single dose metric, however, is unlikely to be sufficient for all endpoints and
exposure durations. Further, the ideal dose metric chosen on the basis of the mode of action
(MOA) may not be the dose metric for which model predictions can be obtained with a high
level of confidence. Consideration of these issues is critical to the selection of the dose metrics
of relevance to dose-response modeling of TCDD.
Figure 3-5 lists a range of alternative dose metrics for TCDD in terms of their relevance
based on considerations of pharmacokinetic mechanisms and MO A. The administered dose or
daily intake (ng/kg-day) is the least relevant dose metric for dose-response modeling of TCDD.
This dose adjusts only for body weight differences between species. When used with an
additional uncertainty factor for kinetics (or BW3 4) and for dynamics, it can also account for
allometrically-predicted pharmacokinetic (clearance) and pharmacodynamic differences between
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
species in deriving the human equivalent dose (HED). In effect, it facilitates the computation of
administered dose associated with the same steady-state blood concentration of parent chemical
in humans and rats by accounting for differences in metabolic clearance (assumed to be related
to body surface area, with no corresponding temporal changes in the volume of distribution; see,
for example, Krishnan and Andersen, 1991). Such calculations of HED for TCDD may not be
appropriate given that (1) steady-state was not attained in all critical toxicological studies chosen
for the assessment, (2) the clearance is mainly due to enzyme(s) and processes whose levels/rates
do not necessarily vary across species or life stages as a function of body surface differences, and
(3) there is a likelihood of change in volume of distribution over time. Furthermore, the use of
administered dose does not explicitly account for the dose-dependent elimination of TCDD from
tissues as demonstrated in multiple studies (reviewed in Sections 3.3.2 and 3.3.4). The use of
administered dose in TCDD dose-response modeling is unlikely to facilitate the characterization
of the true relationship between the response and the relevant measures of internal dose that are
influenced by dose-dependent elimination and binding processes. Additionally, the use of
administered dose to extrapolate across species or life stages would not effectively take into
account the differences in fat content or the demonstrated dose-dependent and species-dependent
differences in elimination half-life of TCDD.
The other alternative dose metrics for TCDD include absorbed dose, body burden, serum
or whole blood concentration, tissue concentration, and functional-related metrics of relevance to
the MOA (e.g., receptor occupancy, change in protein levels). These measures can be calculated
as a current (terminal), average (over a defined period), or integral quantity. The applicability of
the integral measures, such as the AUC (i.e., the area under the curve of a plot of blood or
plasma concentration vs. time), traditionally used for analyzing chronic toxicity data, is
questionable in the case of TCDD because of differences in lifespan and uncertainties regarding
the appropriateness of the duration to be specified for averaging the AUC in experimental
animals and humans for certain critical effects (NRC, 2006).
Among the alternative dose metrics, the absorbed dose accounts for differences in body
weight as well as species-specific differences in bioavailability. However, this dose measure,
while more closely related to the endpoint than administered dose, does not account for the
physiological and biochemical mechanisms responsible for interspecies differences in internal
dose.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_i9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Body burden, or more appropriately the body concentration, represents the amount of
TCDD per kg body weight. TCDD body burdens, like other dose measures, can be determined
as the peak, the average over the period of the bioassays, or the level at the end of the
experiments. Thus, the terminal or average body burdens can be obtained either using data or
pharmacokinetic models and used in dose-response modeling. The body burden is a measure of
TCDD dose that reflects the net impact of bioavailability, uptake, distribution, and elimination
processes in the organism. It is essentially a function of the volume of distribution and clearance
processes, and as such it does take into account the temporal changes in volume of distribution as
well as the concentration-dependent clearance. These are phenomena that are critical to the
understanding of TCDD dose to the target. However, the body burden may not accurately reflect
the tissue dose (NRC, 2006), and as such does not allow for analysis of species-specific
differences in target organ sensitivity to TCDD. In essence, the body burden represents only an
"overall average" of TCDD concentration in the body, without regard to the differential
partitioning and accumulation in specific tissues, including the target tissue(s).
Serum (or blood) concentration of TCDD is a dose metric that reflects both the body
burden and the dose to target tissues. Serum or blood concentration, at steady-state, would be
reflective of the impact of clearance processes, and expected to be directly proportional to the
tissue concentrations of TCDD (NRC, 2006). This dose metric for lipophilic chemicals such as
TCDD is often expressed as a lipid-normalized value, to adjust for varying serum lipid content
(e.g., DeKoning and Karamus, 2000; Niskar et al., 2009), particularly in human biomonitoring
studies, thus of relevance to dose-response modeling; however, the serum lipid-normalized
concentrations of TCDD are not routinely collected and reported in animal toxicologic studies.
The lipid-adjusted serum concentration, however, would be reflective of the lipid-adjusted
concentration of TCDD in other organs (reviewed in Aylward et al., 2008) depending upon the
extent of steady-state attained and the similarity of lipid composition across tissues in each
species. In essence, the serum lipid-normalized measure is representative of the amount of
TCDD per specified volume of total lipids, whereas the whole blood measure will be reflective
of the ensemble of free, lipid-bound and protein-bound TCDD in plasma and erythrocytes, which
may be species-specific. Even though these dose metrics are thought to be more closely and
directly related to the tissue concentrations associated with an effect, a less direct association
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_20 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
might occur at increasing doses when nonlinear processes dominate the kinetics and distribution
of TCDD into organs such as the liver.
Tissue concentration of TCDD, as free, bound, or total TCDD, is a more relevant
pharmacokinetic measure of dose, given that it provides a measure of exposure of the target cells
to the chemical. In this regard, the CYP1A2-bound fraction may be considered as a relevant
dose metric for certain toxic effects; however, the available data contain mixed results regarding
the mechanistic linkage of this dose metric to toxicity and carcinogenicity (reviewed in Budinsky
et al., 2006). In such cases, the use of alternative dose metrics (e.g., bound concentration as well
as the serum concentration) in dose-response modeling could be considered. Other function-
related biomarkers and dose metrics could facilitate the additional consideration of
pharmacodynamic aspects reflecting tissue- and species-specific sensitivity. These metrics
represent the most relevant measures of tissue exposure and sensitivity to TCDD.
Empirical time-course data on the alternative dose metrics of TCDD associated with
epidemiologic and experimental (animal) studies are not available, requiring the use of
pharmacokinetic models to obtain estimates of these dose metrics. These models may be simple,
based on first order kinetics (see Section 3.3.4.2), or more complex based on physiochemical,
biochemical, and physiological parameters for simulating uptake, distribution (including
sequestration to proteins), and clearance of TCDD (see Section 3.3.4.3).
3.3.4.2. First-Order Kinetic Modeling
Figure 3-6 illustrates the process of estimating a human-equivalent TCDD oral exposure
from an experimental animal-administered dose, based on the assumption that body burden is the
effective dose metric for TK equivalence across species. The primary assumption is that the
time-weighted average (TWA) TCDD body burden over some critical time period is the
proximate toxicokinetically-effective dose eliciting a toxicologic effect.9 The process consists of
estimating the effective average body burden in the experimental animal over some time tA
(generally the experimental duration) using a TK model, then "back-calculating" a daily human
exposure level that would result in that average body burden over some time fo- (the human
equivalent to tj\).
9The conversion depicted in Figure 3-6 does not account for toxicodynamic differences between species.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_2l DRAFT—DO NOT CITE OR QUOTE
-------
1 The following closed-form equation is the general formula used to calculate a TCDD
2 terminal body burden in an experimental animal or human at time (t).
3
4 BB(t) = BB(0) +d(l~e (Eq. 3-13)
k
5
6 where
7 BB(t) = the body burden at time t (ng/kg);
8 BB{0) = the initial body burden (ng/kg);
9 d = the daily dose (ng/kg-day);
10 k = the whole-body elimination rate (days1);
11 t = the time at which the body burden is determined (days); and
12 fa = the fraction of oral dose absorbed (unitless).
13
7 d a(\ - fa a
14 For the experimental animal, BB(t) is RR \ (/) = RR \ (0)e A ¦ ' H , and for
kA
15 humans, this parameter is BBH(t) = BBH(0)e~kHtH + ^11 ^— \fajj_
kH
16
17 Setting BBn(f) = BBa(1) obtains the following expression:
18
19 BB„(0)e~k>"» +dH(l-e-kH'H)f<>H = (0)e-^ +dA(l-e-k^A)faA ^ 3_[4)
kH kA
20
21 Rearranging yields the general solution for dH.
22
23 dH = dA kfI faA 0~e AA) + BBa(oy-kAtA _BB (o)e~kHtH (Eq 3.15)
kA faH (1 -e~kHtH )
24
25 Assuming that initial body burdens are very small compared to BB(t) and that the fraction of
26 TCDD absorbed is the same for humans and experimental animals, and using the relationship
i = M2)
27 tl'2 , where /u is the whole-body half-life, a simplified solution for dH is obtained.
28
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
t (\ — e k '!'' )
dH = dA —— (Eq. 3-16)
H V(i-^)
The term l-e~fais the daily fraction eliminated. Therefore, dH can be seen to be the
average daily administered dose to the experimental animal times the ratio of the animal :human
half-life times the ratio of the animal :human daily fraction eliminated over the respective times,
tA and in. For both species at (theoretical) steady state (I —~ co; daily fraction eliminated —>¦ 1),
the latter ratio approaches unity, reducing the animal :human conversion factor to the ratio of the
half-lives. The latter approach was used in the 2003 Reassessment for conversion of animal
cancer slope factors to the human equivalent, where only lifetime exposures are relevant.10
However, for less-than-lifetime exposures eliciting noncancer effects, specific values for
tA and tH must be considered. Furthermore, Eq. 3-16 computes dH on the basis of terminal body
burdens at times tA and in. The more representative metric for toxicokinetic equivalence based
on average body burden over the respective time periods is given in Eq. 3-17.
BB(t) = BB(0)- fe-kTdr + d^-~ f (1 - e)dz = 55(0)^—^ + d&- 1 - ^ g )
t i k t { kt k kt
(Eq. 3-17)
On the basis of average body burden as given in Eq. 3-17, is transformed again assuming
minimal initial body burden (BB(0) ~ 0), as follows:
djj — d
A
hi 2 A
1 (l - e~kAtA )
kAtA
h/lR
1-
%0
ie~kHtH0 _e~
kHtH )
kHtH
(Eq. 3-18)
where tHo is the initial human exposure time.
The value of is the duration of the experimental exposure period. For some gestational
exposures, if a critical exposure window is defined, tA will be the duration of the critical
10No conversions to human-equivalent exposures were attempted for other effects in the 2003 Reassessment.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
exposure window. The value of tH is the human-equivalent duration corresponding to /,,¦.
However, for less than lifetime exposure in humans, tn does not begin at 0, but must end at
25,550 days (70 years) to include the terminal (pseudo) steady-state level, at which the BBH(t):
dH ratio is highest. The average is determined from the terminal end of the human exposure
period because the daily exposure achieving the target blood concentration is smaller than for the
same exposure period beginning at birth and is health protective for effects occurring after
shorter-term exposure.11 Figure 3-7 depicts the relationship of daily dose to TWA body burden
graphically for several exposure duration scenarios. For shorter durations occurring later in life,
the average body burden over the exposure period does not differ substantially from the steady-
state value. Even for half-lifetime exposures, the deviation of the average from steady state is
minimal. Only for lifetime exposures does the difference become more marked, but only by
about 15%. Note that in the 2003 Reassessment, a constant value of 3,000 was used for BBH(t):
dH, based on the relationship of continuous exposure to theoretical steady-state body burden
(t = lifetime, t<, = 2,593 days); this approach, while conservative, does not account for exposure
scenarios of different durations and does not strictly reflect the average body burden dose metric.
The simulation in Figure 3-7 is based on a unit daily exposure to humans, such that the
target body burden represents BBH(tH)'dH as a general scalar for calculating dH from any given
dA. Table 3-3 shows the resulting TK conversion factors for the rodent species and strains
comprising the bulk of the experimental animals in TCDD studies. Monkey and mink values are
not shown in this table because, for the former, only chronic exposures were evaluated and, for
the latter, no TCDD half-life information is available. Monkey (Rhesus) half-life estimates
range from about 200-500 days. A representative value of 365 days is used for this TCDD
assessment. The dA to dH conversion factor for the chronic monkey exposures (3.5-4 years) in
TCDD studies is 9.2-9.7 (BBA\dA = 279-263).
Application of first order kinetics for the risk assessment of TCDD can only be used to
estimate total body burdens or back-calculate administered dose from experimental data. Body
burden calculations using first order kinetics is based on the assumption of a first order decrease
in the levels of administered dose as function of time. In that sense, any loss of TCDD from the
body is described by using a rate constant that is not specific to any biological process. This
11 See the following section (3.3.4.3) for a more detailed discussion of this concept.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
constant is usually estimated from estimates of half-life of TCDD. Assuming a constant half-life
value for the clearance for long-term or chronic TCDD exposure is not biologically supported
given the observed data indicating early influence of CYP1A2 induction and binding to TCDD
and later redistribution of TCDD to fat tissue. Abraham et al. (1988) found that the liver:adipose
tissue concentration ratio in female Wistar rats exposed to a subcutaneous TCDD dose of
300 ng/kg decreased from 10.3 at 1 day postexposure to 0.5 at 91 days postexposure.
Consequently, using half-life estimates based on observed steady-state levels of TCDD will not
account for the possibility of accelerated dose-dependent clearance of the chemical at the early
stages and thus would result in estimation of lower administered levels of the chemical. The
dynamic change in half-life due to dose-dependent elimination at the early stages of TCDD
exposure and its later redistribution to fat tissues for steady-state levels is better described using
biologically-based models, such as the PBPK models and concentration- and age-dependent
elimination (CADM) models (Carrier et al., 1995a, b; Emond et al., 2004, 2005, 2006; Aylward
et al., 2005b). Additionally, these models provide estimates for other dose metrics (e.g., serum
or tissue levels) that are more biologically relevant to response than administered dose or total
body burden (see Section 3.3.4.3).
3.3.4.3. Biologically-Based Kinetic Models
The development and evolution of biologically-based kinetic models for TCDD have
been reviewed by EPA (2003) and Reddy et al. (2005). The initial PBPK model of Leung et al.
(1988) was developed with the consideration of TCDD binding to CYP1A2 in the liver. The
next level of PBPK models by Andersen et al. (1993) and Wang et al. (1997) used diffusion-
limited uptake and described protein induction by interaction of DNA binding sites. The models
of Kohn et al. (1993) and Andersen et al. (1997) further incorporated extensive hepatic
biochemistry and described zonal induction of CYP by TCDD. TCDD PBPK models have
evolved to include detailed descriptions of gastrointestinal uptake, lipoprotein transport, and
mobilization of fat, as well as biochemical interactions of relevance to organ-level effects (Kohn
et al., 1996; Roth et al., 1994). Subsequently, developed PBPK models either used constant
hepatic clearance rate (Wang et al., 1997, 2000; Maruyama et al., 2002) or implemented varying
elimination rates as an empirical function of body composition or dose (Andersen et al., 1993,
1997; Kohn et al., 1996; Van der Molen et al., 1998, 2000). The more recent pharmacokinetic
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
models explicitly characterize the concentration-dependent elimination of TCDD (Carrier et al.,
1995a, b; Emond et al., 2004, 2005, 2006; Aylward et al., 2005b). The biologically-based
pharmacokinetic models describing the concentration-dependent elimination (i.e., the
pharmacokinetic models of Aylward et al. [2005b] and Emond et al. [2005, 2006]) are relevant
for application to simulate the TCDD dose metrics in humans and animals exposed via the oral
route. The rationale for considering the application of Aylward et al. (2005b) and Emond et al.
(2004, 2005, 2006) models for estimating dose metrics for possible application to TCDD risk
assessment is based on the following considerations.
• Both models represent research results from the more recent peer-reviewed publications.
• Both models are relatively simple and less parameterized than earlier kinetic models for
TCDD. The Aylward et al. (2005b) model is based on two-time scale TCDD kinetics
described by Carrier et al. (1995a), and the Emond et al. (2004, 2005, 2006) PBPK
models are reduced versions of earlier complex PBPK models. Although simple, both
the Aylward et al. (2005b) and Emond et al. (2004, 2005, 2006) models are still inclusive
of important kinetic determinants of TCDD disposition.
• Both models are uniquely formulated with dose-dependent hepatic elimination consistent
with the physiological interpretations commonly accepted by the scientific community.
• Both models and extrapolated human versions were tested against human data collected
in a variety of human exposure scenarios (Alyward et al., 2005b; Emond et al., 2005).
• Both models are capable of deriving one or more of the candidate dose-metrics that are of
interest to EPA's dose-response assessment of TCDD.
3.3.4.3.1. CADM model.
3.3.4.3.1.1. Model structure.
The pharmacokinetic model of Aylward et al. (2005b), referred to as the CADM model in
this report, is based on an earlier model developed by Carrier et al. (1995a,b) that describes the
dose-dependent elimination and half-lives of polychlorinated dibenzo-p-dioxins and furans. This
model describes the TCDD levels in blood (body), liver, and adipose tissue. Blood itself is not
characterized physically as a separate compartment within the model, and the distribution of
TCDD to tissues other than adipose tissue and liver (usually less than 4%) is not accounted for
by the model. The original structure of the Carrier et al. (1995a, b) model was modified by
Aylward et al. (2005b) to include TCDD elimination through partitioning from circulating lipids
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_26 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
across the lumen of the large intestine into the fecal content (see Figure 3-8). The most recent
version of the Carrier model (Aylward et al., 2005b, 2008) includes fecal excretion of TCDD
from two routes: (1) elimination from circulating blood lipid through partitioning into the
intestinal lumen; and (2) elimination of unabsorbed TCDD from dietary intake.
A basic assumption of this model is that metabolic elimination of TCDD is a function of
its current concentration in the liver. The current concentration of TCDD in the liver increases
with increasing body burden in a nonlinear fashion as a result of the induction of (and binding of
TCDD to) specific proteins (i.e., CYP1A2). Consequently, the fraction of TCDD body burden
contained in the liver increases nonlinearly (with a corresponding decrease in the fraction
contained in adipose tissues) with increasing body burden of TCDD (Carrier et al., 2005a).
Of particular note is that the adipose tissue compartment of the model is considered to
represent the lipid contained throughout the body. It then assumes that the concentrations of
TCDD in lipids of plasma and various organs is essentially equivalent to that of adipose tissue,
and as such these concentrations are included in the adipose compartment of the model. Even
though this approximation is fairly reasonable given the available data, there is some concern
that the adipose compartment of this model also includes the lipid content of the liver to some
unknown extent. Removal of lipid volume from the liver would mathematically alter total
hepatic concentration and therefore would affect the estimated levels of the chemical available
for binding to proteins.
Distribution in the body is modeled to occur between hepatic and adipose/lipid
compartments, with the fraction of body burden in liver increasing according to a function that
parallels the induction of the binding protein CYP1A2. Elimination is modeled to occur through
hepatic metabolism (represented as a first-order process with rate constant K that decreases with
age) and through lipid-based partitioning of unmetabolized TCDD across the intestinal lumen
into the gut, which is also modeled as a first-order process. As the body burden increases, the
amount of TCDD in the liver increases nonlinearly, resulting in an increased overall elimination
rate.
3.3.4.3.1.2. Mathematical representation.
The CADM model describes the distribution to tissues (including liver and adipose
tissue) based on exchange from blood at time intervals of one month. The model is based on
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
quasi-steady-state-approximation, and thus it is also based on the consideration that the
intertissue processes reach their equilibrium values "quasi-instantaneously." In this regard,
absorption and internal distribution reflective of kinetics at the cellular level (e.g., diffusion,
receptor binding, and enzyme induction) likely occur on a relatively fast time scale (a few hours
to a few days). However, the overall body concentration (i.e., body burden) varies slowly with
time such that it remains virtually unchanged during short time intervals.
The CADM model does not differentiate between binding to AhR and CYP1A2, and it
lacks explicit descriptions of CYP1A2 induction, a key determinant of TCDD kinetics.
However, the empirical equation in the CADM model is based on five parameters (i.e., fmin, fmax,
K, Wa, and Wi; see Tables 3-4 and 3-5) that allow the successful description of the behavior of
TCDD in liver and adipose tissue (i.e., TCDD half-lives in each compartment increase with
decreasing body burden). This observation implies that the model adequately accounts for the
ensemble of the processes. Essentially, the CADM model describes the rate of change in tissue
concentrations of TCDD as a function of total body burden such that the global elimination rate
decreases with decreasing body burden or administered dose.
3.3.4.3.1.3. Parameter estimation.
The CADM model is characterized by its simplicity and fewer parameters compared to
physiologically-based models. Reflecting this simplicity, hepatic extraction is computed with a
unified empirical equation that accounts for all relevant processes (i.e., protein induction and
binding).
The key parameters (fmin. fmax, K, and ke) were all obtained by fitting to species-specific
pharmacokinetic data. The physiological parameters (such as tissue weights) used in the model
are within ranges documented in the literature. The fat content is described to vary as a function
of age, sex, and BMI. However, the BMI of the model is not allowed to change during an
individual simulation (which can range from 20 years to 70+ years) when in reality the
percentage of fat in humans changes over time. None of the TCDD-specific parameters were
estimated a priori or independent of the data set simulated by the model.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
3.3.4.3.1.4. Model performance and desree of evaluation.
The CADM model was not evaluated for its capabilities in predicting data sets not used
in its parameterization. In other words, one or more of the key input parameters (fhmin, fhmax, ke,
K) was or were obtained essentially by fitting to the species-specific pharmacokinetic data, such
that there was no "external" validation data set to which the model was applied. Despite the lack
of emphasis on the "external" validation aspect, the authors have demonstrated the ability of the
model to describe multiple data sets covering a range of doses and species.
The visual comparison of the simulated data to experimental values suggests that the
model could, to an approximate degree, correctly reproduce the whole set of data (e.g.,
pharmacokinetic [PK] profile over a range of dose and time) and not just part of the PK curve,
essentially with the use of a single set of equations and parameters.
The pharmacokinetic data sets for TCDD that were used to calibrate/evaluate the CADM
model included the following:
• Adipose tissue and liver concentrations of TCDD following a single oral dose of 1 |ig/kg
in monkeys (McNulty et al., 1982);
• Percent dose retained in liver for a total dose of 14 ng in hamsters (Van den Berg et al.,
1986);
• Elimination kinetics of TCDD in female Wistar rats following a single subcutaneous dose
of 300 ng/kg (data from Abraham et al., 1988);
• Liver and adipose tissue concentrations (terminal measurements) in Sprague-Dawley rats
given 1, 10 or 100 ng TCDD/kg bw during 2 years (Kociba et al., 1978); and
• Serum lipid concentrations of TCDD over a period of several years in 54 adults (29 men
and 25 women) from Seveso and in three Austrian patients (Aylward et al., 2005a).
For illustration purposes, Figure 3-9 shows model simulations of rat data from Carrier et
al. (1995a). Figure 3-2 (see Section 3.3.2.4) depicts the human data that were used by the
authors to support the concentration-dependent elimination concept; the model was
parameterized to fit approximately to these data (Aylward et al., 2005a).
The authors did not report any specialized analyses that quantitatively evaluated the
uncertainty, sensitivity, and/or variability of CADM model parameters and structure.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
3.3.4.3.1.5. Confidence in CADM model predictions of dose metrics.
The level of confidence associated with the predictability and reliability of absorbed dose
and body burden for oral exposures in humans (as well as several animal species) by this model
can be ranked as high (see Table 3-6). This model, however, does not account for the differential
solubility of TCDD in serum lipids and adipose tissue lipids, nor does it account for the
diffusion-limited uptake by adipose tissue. Due to these limitations, the confidence associated
with the predictions of the serum lipid concentration of TCDD is considered medium,
particularly when it is not documented that steady-state is reached during the critical toxicologic
studies and human exposures. Furthermore, the CADM model does not facilitate the
computation of TCDD concentrations in specific internal organs (other than liver and adipose
tissue). The reliability of this model for simulating the liver concentration (free, bound, or total)
of TCDD at low doses is considered to be low. This low confidence level is a result of the
uncertainty associated with the key parameter /hmin- This parameter needs to be re-calibrated for
each study/species/population to effectively represent the free fraction of TCDD in liver and the
amount of TCDD contained in the hepatic lipids and bound to the liver proteins (whose levels
might be reflective of background exposures of various sources; see Carrier et al., 1995a). The
uncertainty related to the numerical value of this parameter in animals and humans—particularly
at very low exposures—raises concern regarding the use of this model to predict TCDD
concentration (free, bound, or total) in liver as the dose metric for dose-response modeling.
Although the use of the parameter /hmax permits the prediction of the dose to liver at high doses,
it does not specifically facilitate the simulation of the amount bound to the protein or level of
induction in liver. Because the CADM model is not capable of simulating enzyme induction
based on biologically-relevant parameters, its reliability for predicting the concentration of
TCDD bound specifically to the AhR is not known. Finally, due to the lack of parameterization
or verification with kinetic data in pregnant, lactating, or developing animals or humans, the
CADM model is unlikely to be reliable in the current form for use in predicting potential dose
metrics in these subpopulations or study groups that might form the basis of points of departure
(PODs) for the assessment.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
3.3.4.3.2. PBPK model
3.3.4.3.2.1. Model structure.
Emond et al. (2004, 2006) simplified the eight-compartment rat model of Wang et al.
(1997) to a four-compartmental model (liver, fat, rest of body and placenta with fetal transfer)
(Emond et al., 2004), and later to a three-compartment adult model (liver, fat, rest of the body)
(Emond et al., 2006) (see Figures 3-10 and 3-11). Their rationale for simplification of the model
was based on evaluating, critiquing, and improving all earlier PBPK models by Wang et al.
(1997). In general, the main reason for the simplification was that extrapolation of a PBPK
model to humans with these many (i.e., eight compartments) compartments would be
problematic due to the limited availability of relevant human data for validation (Emond et al.,
2004). One major difference from earlier models, repeatedly emphasized by Emond et al. (2005,
2006), was their description (included in their simplified PBPK models) of the dose-dependent,
inducible elimination of TCDD. The rationale for including TCDD binding and induction of
CYP1A2 into the model was earlier described by Santostefano et al. (1998).
The most recent version of the rat and human PBPK models developed by Emond et al.
(2006) describes the organism as a set of three compartments corresponding to real physical
locations—liver, fat, and rest of the body—interconnected by systemic circulation (see
Figure 3-10). The liver compartment includes descriptions of CYP1A2 induction, which is
critical for simulating TCDD sequestration in liver and dose-dependent elimination of TCDD. In
this model, the oral absorption of TCDD from the GI tract accounts for both the lymphatic (70%)
and portal (30%) systems.
The biological relationship between TCDD "sequestration" by liver protein and its
"elimination" by the liver is not entirely clear. TCDD is metabolized slowly by unidentified
enzymes. CYP1A2 is known to metabolize TCDD based on studies in CYP1A2 KO mice
(Diliberto et al., 1997, 1999), in which the metabolic profile is different compared to wild-type
mice. However, since several metabolites appear in the feces of CYP1A2 knock out mice, it is
assumed that there are other enzymes involved in TCDD metabolism. TCDD binds to the AhR
and induces not only CYP1A2, but also CYP1A1, CYP1B1, and several UGTs and transporters
(Gasiwicz et al., 2008). Both hydroxylated and glucuronidated hydroxyl metabolites are found
in the feces of animals treated with TCDD (Hakk et al., 2009). Because the exact enzymes
involved with TCDD are unknown and yet the metabolism is induced by TCDD, an assumption
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
of increased the elimination rate of TCDD in proportion to the induction of CYP1A2 is made. In
the PBPK model, CYP1A2 is needed because TCDD binds to rat, mouse, and human CYP1A2
(Staskal et al., 2005; Diliberto et al., 1999). Thus CYP1A2 induction is necessary to describe
TCDD pharmacokinetics due to TCDD binding. Hence, CYP1A2 can be used as a marker of
Ah-receptor induction of "TCDD metabolizing enzymes." Other models use AhR occupancy as
a marker of induction of "TCDD metabolizing enzymes" (Andersen et al., 1997; Kohn et al.,
2001).
Figure 3-11 depicts the structure of the PBPK model developing rat (Emond et al., 2004).
3.3.4.3.2.2. Mathematical representation.
The key equations of the PBPK model of Emond et al. (2004) are reproduced in
Text Boxes 3-1 and 3-2, whereas those from Emond et al. (2005, 2006) are listed in Table 3-7.
The rate of change of TCDD in the various tissue compartments is modeled on the basis of
diffusion limitation considerations. Accordingly, mass balance equations are used to compute
the rate of change in the tissue (i.e., intracellular compartment) and tissue blood (i.e.,
extracellular compartment). The membrane transfer of TCDD is computed using a permeation
coefficient-surface area cross product (PA) for each tissue. Metabolism and binding of TCDD to
the AhR and inducible hepatic protein (CYP1A2) are described in the liver. The total mass in
the liver was then apportioned between free dioxin (Cif) and bound forms of TCDD (see Figure
3-12). The dose- and time-dependent induction of hepatic CYP1A2 in the liver is described per
Wang et al. (1997) and Santostefano et al. (1998). Accordingly, the amount of CYP1A2 in the
liver was computed as the time-integrated product of inducible production and a simple first-
order loss process (Wang et al., 1997):
dCYP
—j£~ = S(t)K0 - K2CA2t (Eq. 3-19)
at
In this expression, CYPm2 is the concentration of the enzyme (nmol/g), K2 is the rate constant for
the first order loss (liter/hour), Ca2i is the concentration of CYP1A2 in the liver (nM/hour), K{) is
the basal rate of production of CYP1A2 in the liver, and S(t) (unitless) is a multiplicative
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.32 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
stimulation factor for CYP1A2 production in the form of a Hill-type function (see
Section 3.3.2.3):
S(t) = 1 +
(pAh-TCDD ) '
(ICA2f+(C
Ah-TCDD
r
(Eq. 3-20)
where, S(t) is the stimulation function, lnA2 is the maximum fold of CYP1A2 synthesis rate over
the basal rate, Cah-tcdd is the amount of AhR occupied by TCDD, and IC,\2 is the Michaelis-
Menten constant of CYP1A2 induction (nM). The dose-dependent or variable elimination of
TCDD was described using the relationship:
KBILE LI =
CYP\A2induced-CYP\A2basal
CYP1A2
'basal
x Kelv
(Eq. 3-21)
where CYPlA2indUCed is the concentration of induced CYP1A2 (nmol/mL), CYPlA2basai is the
basal concentration of CYP1A2 (nmol/mL), and Kelv is the interspecies constant adjustment for
the elimination rate (hour-1).
There are various ways of formulating the dose-dependent elimination as a function of
the level of CYP1A2, and the above equation (used by the authors) can be viewed as one means
of describing this behavior quantitatively. The numerator in the equation above will always be
greater than zero when there is TCDD in the system (including TCDD derived from either
background exposures or defined external sources). Consequently, the rate of elimination will
correspond to a nonzero value for situations involving TCDD exposures. The above equation,
however, does not describe changes in elimination rate in direct proportionality with the
CYP1A2 levels; also, the Kelv value by itself does not reflect a scalable basal metabolic rate.
Rather, these two terms collectively describe the outcome related to the TCDD elimination
processes, based on fitting to observations in rats (Santostefano et al., 1998). The impact of
CYP1A2 induction and sequestration on binding and elimination of TCDD is simulated using the
Emond et al. (2004) model.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.33 DRAFT—DO NOT CITE OR QUOTE
-------
The gestational model consisted of a fetal compartment, and the transfer of TCDD
between the placental and fetal compartments was described as a diffusion-limited (rather than a
perfusion-limited) process (see Text Boxes 3-1 and 3-2).
Text Box 3-1.
Variation of Body Weight with Age: BWT (?) = BW initial x
1 im e * o/
( BWmother\J
' 0.41 x Tim e ^
1402.5 + Tim e
Cardiac Output: Oc(mL h) = Occ X 60
v 1,000 j
A factor of 60 corresponds to the conversion of minutes to hours, and 1,000 is the conversion of
body weight from g to kg.
Blood Compartment:
Cb(nmol ml.) =
((Of x Cft>) + {Ore x Creb) + (Oli x Clib) + (Opla x Cplab) + Lymph)) - (Cb x Clru)
Qc
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.34 DRAFT—DO NOT CITE OR QUOTE
-------
Text Box 3-2.
Placenta Tissue Compartment
(a) Tissue-blood subcompartment
dAplab ^nm()j _ Qpja(Qa _ Cplab) + PApla(Cplab - Cplafree)
dt
Cplab =
Aplab
Wplab
(b) Tissue cellular matrices
/ h) = PApIafCplab - Cplafree) - dAt>k'-fe> + d^<-P'"
dt dt dt
Cpla{nmol / mL) =
Apia
Wpla
Free TCDD Concentration in Placenta
Cplafree{nmol / mL) = Clpla -
(iCplafree x Ppla +
Plabmax x Cplafree
Kdpla + Cplafree
1
2
Dioxin Transfer from Placenta to Fetuses
dAPla _ fet ^nmQj / j^ = CJpla fet x Cpla
dt
Dioxin Transfer from Fetuses to Placenta
dAfelPla(imol/h) x
dt
Fetal Dioxin Concentration (Fetuses 510 Per Litter)
dAfet, , .,. dAPla fet dAfet Pla
-{nmol/ h) = - ~
dt
Cfet(fimol / h) =
dt
dt
Afet
Wfet
CfetV (nmol / mL) =
Cfet
Pfet
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.35 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
3.3.4.3.2.3. Parameter estimation.
Table 3-8 lists the numerical values of the adult rat and human PBPK models of Emond
et al. (2005, 2006). The values for key input parameters of the rat gestational model are
summarized in Table 3-8 as well as Figure 3-13.
The parameters for the rat model were obtained primarily from Wang et al. (1997) except
that the value of affinity constant for CYP1A2 was changed from 0.03 to 0.04 nmol/mL to get
better fit to experimental data (Emond et al., 2004) and the variable elimination parameter (Kelv)
was obtained by optimization of model fit to kinetic data from Santostefano et al. (1998) and
Wang et al. (1997) (Emond et al., 2005, 2006). Wang et al. (1997) used measured tissue weights
whereas the tissue blood flows and tissue blood weights were obtained from International Life
Sciences Institute (ILSI, 1994). The partition coefficients (which were similar to those of Leung
et al. [1988, 1990]), the permeability x area (PA) value for tissues, the dissociation constant for
binding to CYP1A2 (ICa2) and the Hill coefficient (h) were estimated using a two-stage process
of fitting to dose-response and time-course data on TCDD tissue distribution (Wang et al., 1997).
In the initial stage, the experimental data of arterial blood concentrations were used as input to
the individual compartment to estimate the parameters; then, with the values obtained during
stage one as initial estimates, those unknown parameters were re-estimated by solving the entire
model at once using an optimization route (Wang et al., 1997). The receptor concentrations and
dissociation constant of TCDD bound to AhR were obtained by fitting the model to TCDD tissue
concentration combining with enzyme data reported by Santostefano et al. (1998) whereas the
basal CYP1A2 in liver was based on literature data (Wang et al., 1997).
The parameters for the human PBPK model were primarily based on the rat model (Wang
et al., 1997; Emond et al., 2005, 2006). Specifically, the blood fraction in the tissues, the
tissue:blood partition coefficients, tissue permeability coefficient, the binding affinity of TCDD
to Ah and CYP, and the maximum binding capacity in the liver for AhR were all set equal to the
values used in the rat model. The species-specific Kelv was estimated by fitting to human data
(Emond et al., 2005).
For the gestational rat model, the parameters describing the growth of the placental and
fetal compartments as well as temporal change in blood flow during gestation were incorporated
based on existing data. Exponential equations for the growing compartments were used (see
Figure 3-13), except for adipose tissue for which a linear increment based on literature data was
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_36 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
specified. While physiological parameters for the pregnant rat were obtained from the literature,
all other input parameters were set equal to that of nonpregnant rat (obtained from Wang et al.
[1997]), see Tables 3-7 and 3-8. The current version of the rat gestational model contains
parameters for variable elimination from Emond et al. (2006; Table 3-8), and still provides the
essentially the same predictions as the original publication (Emond et al., 2004).
3.3.4.3.2.4. Model performance and degree of evaluation.
The PBPK model of Emond et al. (2004, 2005, 2006) had parameters estimated by fitting
to kinetic data, such that the resulting model consistently reproduced the kinetic data. The same
model structure with a single set of species-specific parameters could reproduce the kinetics of
TCDD following various doses and exposure scenarios not only in the rat but also in humans.
The simulations of the PBPK model of Emond et al. (2006) have been compared with two sets of
previously published rat data: blood pharmacokinetics following a single dose of 10 |ig/kg (the
dose corresponding to the mean effective dose for induction of CYP1A2) (Santostefano et al.,
1998) (see Figure 3-14); and hepatic TCDD concentrations during chronic exposure to 50, 100,
500, or 1,750 ng/kg (Walker et al., 1999) (see Figure 3-15). It is relevant to note that the PBPK
model of Emond et al. (2004, 2006) is essentially a reduced version of the Wang et al. (1997)
model, and it therefore provides simulations of liver and fat concentrations of TCDD that
deviated by not more than 10-15% of those of Wang et al. (1997). The nongestational model of
Emond et al. (2004) simulated the kinetic data in liver, fat, blood and rest of body of female
Sprague-Dawley rats given a single dose of 10 |ig TCDD/kg (data from Santostefano et al.,
1996) and in liver and fat of male Wistar rats treated with a loading dose of 25 rigkg followed by
a weekly maintenance dose of 5 ng TCDDAg by gavage (data from Krowke et al., 1989).
The gestational rat PBPK model simulated the following PK data sets (Emond et al.,
2004):
• TCDD concentration in blood, fat, liver, placenta, and fetus of female Long-Evans rats
given 1, 10, or 30 ng4g, 5 daysAveek, for 13 weeks prior to mating followed by daily
exposure through parturition (Hurst et al., 2000a);
• TCDD concentration in tissues (liver, fat), blood, placenta and fetus determined on
gestation day (GD) 16 and GD 21 following a single dose of 0.05, 0.8, or 1 (J,g/kg given
on GD 15 to pregnant Long Evans rat (Hurst et al., 2000b);
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.37 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
• Maternal and fetal tissue concentrations on GD 9, GD 16 and GD 21 after a single dose
of 1.15 jag TCDD/kg given to Long-Evans rats on GD 9 or GD 15 (Hurst et al., 1998);
and
• Fetal TCDD concentrations determined on GD 19 and GD 21 in rats exposed to 5.6 jag
TCDD/kg on GD 18 (Li et al., 2006)
Furthermore, the scaled rat model was shown to be capable of simulating human data
from the Austrian and Seveso subjects (see Figures 3-16 and 3-17). In this regard, it is useful to
note that the computational version of the PBPK model of Emond et al. (2005, 2006) also
contained the necessary equation to transform the model output of blood concentration into
serum lipid adjusted concentration of TCDD.
The human model of Emond et al. (2005) has advantages for improving the TCDD
dosimetry used in existing human epidemiological studies because its PBPK structure naturally
develops expectations for the redistribution of TCDD within the body (to stores in fat and liver)
relative to metabolism. However, because the dose-dependency of metabolic elimination in the
Emond et al. (2005) human model was not calibrated to human data, it is important to review the
expectations of this model using a database of human observations that is as extensive as
possible and a spread of internal TCDD concentrations that is as wide as possible. Thus,
presented below is a juxtaposition of expectations of elimination rates for the Emond model with
observations for two highly exposed Austrian patients and nine of 10 Ranch Hand veterans12
used for the original "validation" comparisons in the Emond et al. (2005).
Figure 3-18 shows the time course of the declines in TCDD serum concentrations in two
highly-exposed Austrian subjects compared with expectations from the Emond et al. (2005)
model. The comparison in Figures 3-17 and 3-18 indicates that the Emond model adequately
describes the rate of TCDD elimination for the more highly exposed Austrian patients, but
predicts a somewhat faster rate of decline than that observed for the less heavily exposed patient.
Figure 3-19 shows the results of combining the simulated and observed rates of loss for a
group of Austrian and Ranch Hand subjects evaluated by Emond et al. (2005), counting only one
data point per person. The X-axis in this figure is the TCDD serum concentration at the
12In preliminary comparisons, the simulation run for the 10th Ranch Hand veteran appeared anomalous and was
therefore excluded from this summary.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_38 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
midpoint of the observations for each subject. The error bars in the figure represent ±1 standard
error.
Table 3-9 presents the results of regression analyses of the observed rates of decline in
relation to the estimated TCDD serum levels at the midpoint of the observations for each subject
in the Ranch Hand study (see Figure 3-19). These results indicate that some appreciable dose
dependency of TCDD elimination is unequivocally supported. However, the central estimate of
the slope of the relationship between the log of the TCDD elimination rate and the log of the
TCDD level is only about 75% of that expected under the Emond et al. PBPK model
(i.e., 0.092-0.123 =0.748).
Based on Table 3-9 and Figure 3-19, the expected slope from the Emond model
(slope = 0.123) is about 2.8 standard errors above the regression line (slope = 0.092 with
standard errors value of 0.011) indicated by the observations, suggesting that the departure is
statistically significant. It can be seen that some appreciable dose dependency of TCDD
elimination is unequivocally supported, but it is indicated that the dependency is somewhat less
than presently incorporated into the Emond model. The difference might be slightly enhanced if
the modest effects of the special measures taken to accelerate removal of TCDD from the
Austrian subjects were subtracted from the observed elimination rates (i.e., oral administration of
Olestra for both patients and LDL-apheresis for the more highly exposed patient).
It would be desirable in the future to extend this juxtaposition to data for other intensively
studied subjects. Figure 4 in Aylward et al. (2005a) (reproduced here as Figure 3-2) shows a plot
of elimination rate versus initial level of lipid-corrected TCDD in serum for 36 people. This is
not the most desirable comparison for characterizing the relationship, however, because the rate
constant for loss should be related to the geometric mean or midpoint level in the decline for
each person (rather than to the initial level) in order to be most accurate in relating current
TCDD levels to elimination rates and to avoid possible "regression to the mean" type statistical
errors due to measurement imperfections. Overall, the conclusion from this analysis is that it is
not unreasonable to use the Emond model as it stands, but the model might be modestly
improved by adapting it to (1) include the two nondose-dependent pathways of elimination
documented in the Geusau papers (GI elimination via the feces and loss via the sloughing of skin
cells), and (2) reducing the extent of loss via the dose-dependent metabolism pathway from the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
liver (Geusau et al., 2002; Harrad et al., 2003) so that overall loss rates for the average
elimination rates from the Ranch Hand veterans is maintained.
A sensitivity analysis of inputs used to estimate inducible elimination rate for a single
oral dose of 0.001 to 10 |ig/kg in the rat indicated that the number of key parameters ranged from
seven at the low dose region to 12 at the high dose (see Figure 3-20) (Emond et al., 2006). The
sensitive parameters identified included the oral absorption parameters (KABS), volumes of liver
and adipose tissue (WLIO, WFO), adipose tissue:blood partition coefficient (PF), and the basal
CYP1A2 level (CYP1A2 1A2). At high doses, the most sensitive parameters also included those
related to the maximal induction of CYP1A2 and AhR binding capacity (see Figure 3-20)
(Emond et al., 2006).
The gestational rat model described in Emond et al. (2004), upon reparameterization,
could simulate the kinetics of TCDD in mice. The initial changes to the rat model parameters
included: rest of the body:blood partition coefficient (PRE), basal concentration (CYP1A21A2),
delay in induction time (CYP1A21TAU) and adipose tissue permeability coefficient (PAFF), in
accordance with Wang et al. (2000) (see Table 3-8). Subsequently, four parameters (adipose
tissue:blood partition coefficient, CYP1A2 affinity parameter, GI tract elimination transit
constant (hour-1) and the interspecies metabolic parameter Kelv (hour-1) were re-estimated based
on visually fit of model simulations to the PK data from Deliberto et al. (2001), following an oral
dose 150 ng TCDD/kg/day, 5 days/week for 17 weeks (see Table 3-7). The resulting mouse
model is capable of reproducing the kinetics of TCDD in the adult (see Figures 3-21 through
3-27), as well as, to a very limited extent, the kinetics during gestation (see Figure 3-28).
3.3.4.3.2.5. Confidence in PBPK model predictions of dose metrics.
The PBPK model facilitates prediction of absorbed dose, body burden, and blood
concentration of TCDD for oral exposures in adult humans and rats (adult and developing) with
high confidence (see Table 3-10). The model output of blood concentration can be normalized to
lipid content representative of the study group (species, sex, age, lifestage, and diet). However,
the PBPK model of Emond et al. (2004, 2005, 2006) does not simulate plasma and erythrocyte
TCDD concentrations separately, and it predicts tissue concentrations on the basis of
tissue:whole blood partition coefficients and not on the basis of serum lipid-normalized values.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-40 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
The reliability of this model for simulating the liver concentration of TCDD in rats is
considered to be high but it is considered to be medium for humans. Although empirical data on
bound or free concentrations were not used to evaluate model performance in humans, the
biological phenomena (consistent with available data) related to the hepatic sequestration,
enzyme induction, and dose-dependent elimination are described in the model. This is one of the
situations where PBPK models are uniquely useful; that is, they permit the prediction of system
behavior based on understanding of the mechanistic determinants, even though the required data
cannot be directly obtained in the system (e.g., bound concentrations in the liver of exposed
humans). For these dose measures (i.e., bound concentration and total liver concentration), the
level of confidence can be further improved or diminished by the outcome of sensitivity analysis.
In this regard, the results of a focused sensitivity analysis indicate that the most sensitive
parameters of the human model are among the most uncertain (i.e., those parameters for which
estimates were not obtained in humans) with respect to prediction of liver TCDD concentration,
contrary to the animal model (see Section 3.3.6).
With respect to the mouse model, however, the level of confidence is medium to low,
given that it has not been verified extensively with blood, body burden, or tissue concentration
time-course or dose-response data. However, the mouse PBPK model, based on the rat model
that has been evaluated with several PK data sets, has been shown to reproduce limited liver
kinetic data (see Figures 3-21 through 3-28; Boverhoff et al., 2005). The same model structure
has been used for simulating kinetics of TCDD in humans successfully. Overall, the adult mouse
model, given its biological basis combined with its ability to simulate TCDD kinetics in multiple
species, is considered to exhibit a medium level of confidence for simulating dose metrics for use
in high to low dose extrapolation and interspecies (mouse to human) extrapolation. Even though
similar considerations are applicable to gestational model in mice, the confidence level is
considered to be low since very limited comparison with empirical data has been conducted (see
Figure 3-28). Despite the uncertainty in these predictions, the scaled rat gestational model, given
its biological and mechanistic basis, might be of use in predicting dose metrics in these groups
that might form the basis of PODs in certain key studies.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_41 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
3.3.4.4. Applicability of PK Models to Derive Dose Metrics for Dose-Response Modeling of
TCDD: Confidence and Limitations
Both the CADM and PBPK models describe the kinetics of TCDD following oral
exposure to adult animals and humans by accounting for the key processes affecting kinetics,
including hepatic sequestration phenomena, induction, and nonlinearity in elimination, and
distribution in adipose tissue and liver. Both models can be used for estimating body burdens
and serum lipid adjusted concentrations of TCDD. However, there are several differences
between these two models. The PBPK model calculates the free and bound concentrations of
TCDD in the intracellular subcompartment of tissues. The total or receptor-bound
concentrations in liver are unambiguous and more easily interpretable with the PBPK model than
with the CADM model. In addition, the PBPK model computes bound and total concentrations
as a function of the free concentration in the intracellular compartment of the tissue. By contrast,
the CADM model simulates the total concentration based on empirical consideration of hepatic
processes. Consequently, the amount of TCDD bound to AhR or CYP1A2 cannot be simulated
with the CADM model. The CADM model computes only the total TCDD concentration in
liver, and describes TCDD elimination through partitioning from circulating lipids across the
lumen of the large intestine into the feces, while the PBPK model accounts for this process
empirically within its hepatic elimination constant. Elimination of TCDD via skin, a minor
process, is not described by either model. Thus, dose-response modeling based on body burden
of TCDD in adult animals and humans can be conducted with either of the models, provided the
duration of the experiment is at least one month, due to limitations in the CADM model. As
shown in Figure 3-29, the predicted slope and body burden over a large dose range are quite
comparable (generally within a factor of two).
Results of simulations of serum lipid concentrations or liver concentrations vary for the
two models to a larger extent (up to a factor of 7), particularly for simulations of short duration.
These differences reflect two characteristics of the PBPK model: first, quasi-steady-state is not
assumed in the PBPK model; second, the serum lipid composition used in the model is not the
same as the adipose tissue lipids. The CADM model does not account for differential solubility
of TCDD in serum lipids and adipose tissue lipids, nor does it account for the diffusion-limited
uptake by adipose tissue. Therefore, the PBPK model would appear to be superior to the CADM
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-42 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
model with respect to the ability to simulate serum lipid and tissue concentrations during
exposures that do not lead to the onset of steady-state condition in the exposed organism.
The CADM model is simple and based on fewer parameters than the PBPK model.
Because the CADM model is constructed by fitting to data, its performance is likely to be
reliable for the range of exposure doses, species, and life stages from which the parameter
estimates were obtained. On the other hand, the PBPK model structure and parameters are
biologically-based and can be adopted for each species and life stage. Accordingly, the PBPK
model has been adopted to simulate the kinetics of TCDD in the fetus and in pregnant rats, as
well as in adult humans and rats (Emond et al., 2004, 2005, 2006). The time step for calculation
and dosing in the CADM model corresponds to 1 month. This requirement represents a
constraint in terms of the use of this model to simulate a variety of dosing protocols used in
animal toxicity studies. This requirement, however, is not a constraint with the PBPK models.
So, simulating the body burden and serum lipid concentrations for a longer duration of exposure,
either model would appear to be useful; but the PBPK model would be the tool of choice for
simulating alternative dose metrics of TCDD (e.g., blood concentration, total tissue
concentration, bound concentration) for various exposure scenarios (including single dose
studies), routes and life stages in the species of relevance, to TCDD dose-response assessment,
particularly, mice, rats, and humans.
Two minor modifications, to enhance the biological basis, were made to the PBPK model
of Emond et al. (2006), before its use in the computation of dose metrics for TCDD. The first
one involved the recalculation of the volume of the rest of the body as follows:
WRE0 = (0.91 - (WLIB0 x WLI0 + WFB0 x WF0 + WLI0 + WF 0)/(l + WREB0)) (3-22)
where
WRE0 = weight of cellular component of rest of body compartment (as fraction of
body weight);
WLI0 = weight of cellular component of liver compartment (as fraction of body
weight);
WF0 = weight of cellular component of fat compartment (as fraction of body
weight);
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.43 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
WREBO = weight of the tissue blood component of the rest of body compartment (as
fraction of body weight);
WLIBO = weight of the tissue blood component of the liver compartment (as fraction
of body weight); and
WFBO = weight of the tissue blood component of the fat compartment (as fraction of
body weight).
In the original code, the weight of the rest of body compartment was calculated as the
difference between 91% of body weight and the sum total of the fractional volumes of blood,
liver tissue (intracellular component), and adipose tissue (intracellular component). The blood
compartment in the PBPK model is not explicitly characterized with a volume; as a result, the
total volume of the compartments is less than 91%. The recalculations shown above were used
to address this problem. Given the very low affinity of TCDD for blood and rest of the body,
reparameterizing the model resulted in less than a 1% change in output compared to the
published version of the PBPK model for chronic exposure scenarios (Emond et al., 2006).
The second minor modification related to the calculation of the rate of TCDD excreted
via urine. The original model code computed the rate of excretion by multiplying the urinary
clearance parameter with the concentration in the rest of the body compartment. Instead, the
code was modified to use the blood concentration in this equation. This resulted in the
re-estimation of the urinary clearance value in the rat and human models but it did not result in
any significant change in the fit and performance of the original model.
The revised parameter estimates of the rat, mouse, and human models are captured in
Table 3-8 with a footnote.
3.3.4.5. Recommended Dose Metrics for Key Studies
The selection of dose metrics for the dose-response modeling of key studies is largely the
result of (1) the relevance of a dose metric on the basis of current knowledge of TCDD's
mechanism of action for critical endpoints and (2) the feasibility and reliability of obtaining the
dose metric with available PK models. Secondarily, the goodness-of-fit of the dose-response
models (which reflects the relationship of the selected internal dose measures to the response)
can be used to inform selection of the most appropriate dose metric for use in deriving TCDD
toxicity values.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.44 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Body burden—even though this metric is based on mechanistic considerations—is a
somewhat distant measure of dose with respect to target tissue dose, and this metric represents
the "overall" average concentration of TCDD in the body. However, a benefit of body burden is
that this metric represents a dose measure for which the available PK models can provide highly
certain estimates. Thus, the overall confidence associated with the use of body burden in TCDD
assessment is categorized as medium.
The confidence in the ability of PK models to simulate blood concentration as a dose
metric is high, given that the models have been shown to consistently reproduce whole blood (or
serum lipid-normalized) TCDD concentration profiles in both humans and rats. Considering the
facts that the PBPK models simulate whole blood rather than the serum lipid-normalized
concentrations of TCDD and that the study-specific values of serum lipid content are not known
with certainty, it is preferable to rely on TCDD blood concentrations as the dose metric. The
blood concentrations, if intended, can be normalized on the basis of appropriate total lipid levels.
However, based on mechanistic considerations, the confidence in their use would be somewhat
lower for hepatic effects. This conclusion reflects the concern regarding the inconsistent
relationship between the two variables with increasing dose levels and the fraction of steady-
state attained at the time of observation. For other systemic effects related to tissue
concentrations, the confidence in the use of TCDD serum or blood concentration is high,
particularly for chronic exposures, given the absence of organ-specific nonlinear mechanisms.
The tissue concentration typically cannot be calculated as a reliable dose metric with either PK
model. One exception to this conclusion is the use of PBPK models to estimate levels in liver, a
metric that is highly relevant based on MOA considerations. Finally, the bound concentration
may be evaluated for receptor-mediated effects. This dose metric, of medium-low confidence,
can be obtained with PBPK models for high dose-low dose and interspecies extrapolations. The
alternative dose metrics for dose-response modeling of TCDD selected on the basis of MOA and
PK modeling considerations are summarized in Tables 3-11 and 3-12.
These measures of internal dose can be obtained as peak, average, integral (AUC), or
terminal values. For chronic exposures (e.g., exposures reflected by the results of a cancer
bioassay) in rodents, the terminal and average values would be fairly comparable. For less-than
lifetime exposures, however, the terminal and average values will differ, and therefore an overall
average or integrated value (AUC) would be more appropriate. Similarly, for developmental
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.45 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
exposures, these alternative dose metrics can be obtained with reference to the known or
hypothesized exposure window of susceptibility.
3.3.5. Uncertainty in Dose Estimates
3.3.5.1. Sources of Uncertainty in Dose Metric Predictions
3.3.5.1.1. Limitations of available PK data.
3.3.5.1.1.1. Animal data.
The available animal data relate to blood, liver, and adipose tissue concentrations for
certain exposure doses and scenarios. Although these data are informative regarding the dose-
and time-dependency of TCDD kinetics for the range covered by the specific studies (see
Section 3.3.2), they do not provide the peak, average, terminal, or lipid-normalized values of
dose metrics associated with the key studies selected for this assessment. The limited available
animal PK data are useful, however, in the evaluation of the pharmacokinetic models (see
Section 3.3.4).
3.3.5.1.1.2. Human data.
The human data on potential dose metrics are restricted to the serum lipid-adjusted
TCDD concentrations associated with mostly uncharacterized exposures (see Sections 3.3.2 and
3.3.3). While these data are useful in estimating half-lives in exposed human individuals, they
do not provide estimates of hepatic clearance or reflect target organ exposure. Some autopsy
data have been used to infer the partition coefficients; however, these data were collected
without quantification of the temporal nature of TCDD uptake (see Section 3.2). Despite the
limitations associated with the available human data, there has been some success in using these
data to infer the half-lives and elimination rates in humans using pharmacokinetic models
(Aylward et al., 2005b; Carrier et al., 2005a; Emond et al., 2006).
3.3.5.1.2. Uncertainties associated with model specification.
Uncertainty associated with model specification should be viewed as a function of the
specific application, such as interspecies extrapolation, intraspecies variability, or high dose to
low dose extrapolation. Because the use of pharmacokinetic models in this assessment is limited
to interspecies extrapolation and high dose to low dose extrapolation, it is essential to evaluate
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_46 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
the confidence in predicted dose metrics for these specific purposes. For interspecies
extrapolation, the PBPK and CADM models calculate differences in dose metric between an
average adult animal and an average adult human. Both models have a biologically and
mechanistically-relevant structure along with a set of parameters with reasonable biological
basis, and reproduce a variety of pharmacokinetic data on TCDD in both rodents and humans.
These models possess low uncertainty with respect to body burden, blood, and TCDD/serum
(lipid) concentration for the purpose of conducting rat to human extrapolation. However, for
other dose metrics, such as free, total, or bound hepatic concentrations, the uncertainty is higher
in the CADM model compared to the PBPK model due to model specification differences related
to the mechanisms of sequestration and induction in the liver (see Section 3.3.3).
For the purpose of high dose to low dose extrapolation in experimental animals,
confidence in both models is high with respect to a variety of dose metrics (see previous
discussion). The high confidence results from the use of the PBPK models to reproduce a
number of data sets covering a wide range of dose levels in rodents (rats, mice). This dose range
likely covers that of most of the key toxicological studies. Given that the TCDD levels during
and at the end of exposures were not measured in most of the key studies, use of the PBPK
models is preferred because these models account for dose-dependent elimination, induction, and
sequestration. Despite the empirical nature of the specification of these key processes in PBPK
models, they essentially reproduce the dose-dependent behavior in rodents, supporting their use
in deriving dose metrics for dose-response modeling of TCDD. Overall, the uncertainty
associated with the use of the dose metrics (identified in Table 3-10) is less than the uncertainty
associated with the use of administered dose of TCDD, for relating to the concentration within
tissues to produce an effect. The administered dose does not take into account interspecies
differences in the volume of distribution and clearance or the complex nonlinear processes
determining the internal dose.
The PBPK model of Emond et al. (2006) could benefit from further refinement and
validation, including a more explicit consideration of nondose-dependent elimination pathways.
As indicated in Section 4, there is some uncertainty associated with the way the elimination of
TCDD is described in the existing human PBPK model. The current model essentially treats all
TCDD elimination as related to dose dependent metabolism in the liver. In this regard, the
classical and more recent PK data on TCDD may be useful in further improving the confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.47 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
in their predictions. However, it is likely that there is nondose-dependent elimination of TCDD
via feces and, to a lesser extent skin; juxtaposition of available elimination rate data with the
PBPK model predictions suggests that the current PBPK model modestly overestimates the dose
dependency of overall TCDD elimination. (The central estimate of the slope of the relationship
between the log of the TCDD elimination rate and the log of the TCDD level is only about
three-fourths of that expected using the unmodified PBPK model). Emond et al. (2005)
acknowledge that the model did not describe the elimination of TCDD from the blood into the
intestines, but it indirectly accounted for this phenomenon with the use of the optimized
elimination rate.
3.3.5.1.3. Impact of human interindividual variability.
The sources and extent of human variability suggested by the available data are presented
in Section 3.3.3, although there is some discussion of the impact of individual differences in
body fat content. The CADM model facilitates the simulation of body burden and serum lipid
concentrations on the basis of BMI and tissue weights of people, and the PBPK model simulates
alternative dose metrics in the fetus and in pregnant animals in addition to adult animals and
humans. However, neither of these models has been parameterized for simulation of population
kinetics and distribution of TCDD dose metrics. Therefore, at the present time, a quantitative
evaluation of the impact of human variability on the dose metrics of TCDD is not feasible, and
dose metric-based replacement of the default interindividual factor has not been attempted.
3.3.5.2. Potential Magnitude and Sources of Uncertainty in Dose Metrics
3.3.5.2.1. Magnitude of uncertainty.
The usefulness of the CADM and PBPK models for conducting high dose to low dose
and interspecies extrapolations is determined by their reliability in predicting the desired dose
metrics. The confidence, or conversely the magnitude of uncertainty, in the model predictions of
dose metrics is dictated by the extent to which the model has been verified with empirical data
relevant to the dose metric, supplemented by sensitivity and uncertainty analyses. Analysis of
sensitivity or uncertainty has not been conducted with the CADM model. For the PBPK model,
Emond et al. (2006) published the initial results from sensitivity analyses of acute exposure
modeling (see Section 3.3.3). One of the objectives of a sensitivity analysis that is of highest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
relevance to this assessment is to identify the most critical model parameters with respect to the
model output (i.e., dose metric).
If the model simulations have only been compared to entities that do not correspond to
the moiety representing the dose metric, or if the comparisons have only been done for some but
not all relevant dose levels, routes, and species, then the reliability in the predictions of dose
metric can be an issue. The extent to which model results are uncertain will depend largely upon
the extent to which the dose metric is measurable (e.g., serum concentrations of TCDD) or
inferred (e.g., receptor-bound TCDD concentration).
With respect to body burden and blood concentration, extrapolation uncertainty is low,
and therefore the need for sensitivity and uncertainty analysis is less critical. For serum lipid-
based metric, the lipid content used for normalizing the animal and human blood concentrations
will have a direct impact on the outcome. Because the PBPK models have directly evaluated
blood or tissue concentration data of relevance to dose metric calculations in the species and life
stage of interest to the assessment, extrapolation uncertainty is low and therefore confidence in
these simulated dose metrics is high. For those dose metrics that are not directly measurable or
are less easily determined by available calibration methods (e.g., free liver concentration,
receptor bound concentration), sensitivity and uncertainty analyses are crucial to inform the
reliability of the PBPK model predictions
Sensitivity analysis for the PBPK model predictions of liver concentration of TCDD
indicated hepatic CYP1A2 concentration is the most sensitive parameter in the rat model
(Emond et al., 2006). In addition, the absorption parameters, basal concentration of CYP1A2,
and adipose tissue:blood partition coefficients were identified as highly sensitive model
parameters for simulations of human kinetics. These results indicate that the confidence in the
use of the rat PBPK model for high dose to low dose extrapolation is high. Confidence is low for
the purpose of rat to human extrapolation given that the values of these key human model
parameters are uncertain. A similar inference can be made with respect to the concentration of
bound TCDD in liver, suggesting that the use of bound concentration as a dose metric in the rat
is reliable for high dose to low dose extrapolation, and less reliable for extrapolating from rat to
human due to model sensitivity to parameters with high uncertainty.
With regard to the predictability of body burden, the absorption and excretion parameters
were among the sensitive parameters in the rat. Several other parameters were also identified as
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.49 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
being sensitive in humans. Despite the sensitivity to these parameters and the uncertainty
associated with individual parameter estimates, the overall confidence in the model predictions
of body burden appears to be high given the reproducibility of empirical data on tissue burdens
and blood concentrations of TCDD in various experiments by both models. Similar conclusions
can be drawn for serum lipid concentration of TCDD predicted by the PBPK model, except that
the assigned value of lipid content will have additional, direct impact on this dose metric.
Therefore, knowledge of the statistical representativeness of the lipid levels for rodents and
humans, intended for use in normalization, will be crucial. In this regard, variability of the total
lipid levels and the variability of the contribution of phospholipids and neutral lipids to the total
lipid pool across species, lifestage and study groups is to be expected (Poulin and Theil, 2001;
Bernert et al., 2007).
Both conceptual uncertainty and prediction uncertainty are relevant to dose metrics used
in the dose-response modeling. In the current context, conceptual uncertainty arises from the
assumed relevance of the dose metrics to the MOA and target organ toxicity of TCDD.
Prediction uncertainty reflects the lack of confidence in the model predictions of dose metrics.
Tables 3-13 and 3-14 provide a qualitative ranking of the importance and magnitude of each
dose metric with respect to these two sources of uncertainty. Overall, uncertainty is high for the
use of administered dose and absorbed dose at one end of the spectrum and receptor occupancy
at the other end of the spectrum. Based on the information presented in Table 3-13, for instances
involving the use of rat and human PBPK models, the overall uncertainty associated with the use
of body burden and serum concentrations are medium, whereas for hepatic effects, the use of
liver concentration would be considered to be medium/low. While using mouse PBPK model
along with the human model (see Table 3-14), the contribution of the prediction uncertainty to
the overall uncertainty increases due to the limited comparison of the mouse model simulations
with empirical data.
3.3.6. Use of the Emond PBPK Models for Dose Extrapolation from Rodents to Humans
EPA has selected the Emond et al. (2004, 2005, 2006) PBPK models, as modified by
EPA for this assessment, for establishing toxicokinetically-equivalent exposures in rodents and
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_50 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
humans.13 The 2003 Reassessment (U.S. EPA, 2003) presented a strong argument for using the
relevant tissue concentration as the effective dose metric. However, no models exist for
estimation of all relevant tissue concentrations. Therefore, EPA has decided to use the
concentration of TCDD in blood as a surrogate for tissue concentrations, assuming that tissue
concentrations are proportional to blood concentrations. Furthermore, because the RfD and
cancer slope factor are necessarily expressed in terms of average daily exposure, the blood
concentrations are expressed as averages over the relevant period of exposure for each endpoint.
Specifically, blood concentrations in the model simulations are averaged from the administration
of the first dose to the administration of the last dose plus one dosing interval unit in order to
capture the peaks and valleys for each administered dose. That is, for daily dosing, 24 hours of
TCDD elimination following the last dose is included in the average (the modeling time interval
is one hour); for a weekly dosing protocol, a full week is included. In addition, because of the
accumulation of TCDD in fat and the large differences in elimination kinetics between rodent
species and humans, exposure duration plays a much larger role in TK extrapolation across
species than for rapidly-eliminated compounds. Because of these factors, EPA is using discrete
exposure scenarios that relate human and rodent exposure durations. The use of discrete
exposure scenarios was introduced previously in Section 3.4.4.2 describing first-order kinetic
modeling and is further described in the following paragraphs. This section concludes with a
quantitative evaluation of the impact of exposure duration on the rodent-to-human TK
extrapolation from both the human and rodent "ends" of the process.
Figure 3-30 shows the TCDD blood concentration-time profile for continuous exposure
at 0.01 ng/kg-day, as predicted by the Emond human PBPK model, and the target TCDD
concentrations corresponding to the three discrete exposure scenarios used by EPA in this
document. The target concentrations are those that would be identified in the animal bioassay
studies that correspond to a particular POD (no-observed-adverse-effect level, lowest-observed-
adverse-effect level, or benchmark dose lower confidence bound) established for that bioassay.
That is, the target concentrations represent the toxicokinetically-equivalent internal exposure to
be translated into an equivalent human intake (or FLED).
13The models will be referred to hereafter as the "Emond human PBPK model" and the "Emond rodent PBPK
model," with variations when referring to individual species or components (e.g., gestational).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_51 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
For the lifetime exposure scenario, the HED is "matched" to the lifetime average TCDD
blood concentration from a lifetime animal bioassay result by determining the continuous daily
intake that would result in that average blood concentration for humans over 70 years.
For the gestational exposure scenario, the effective TCDD blood concentration (usually
the peak) determined for the particular POD in a particular developmental study is matched to
the average TCDD blood concentration over the gestational portion of the human gestational
exposure scenario. The HED is determined as the continuous daily intake, starting from birth
that would result in that average blood concentration over the 9-month gestational period for a
pregnancy beginning at 45 years of age. The choice of 45 years as the beginning age of
pregnancy is health protective of the population in that the daily exposure achieving the target
blood concentration is smaller than for earlier pregnancies.
For a less-than-lifetime exposure, the average TCDD blood concentration over the
exposure period in the animal bioassay associated with the POD is matched to the average over
the 5-year period that includes the peak concentration (58 years for an intake of 0.01 ng/kg-day).
The HED is determined as the continuous daily intake that would result in the target
concentration over peak 5-year period. The use of the peak is analogous to the approach in the
2003 Reassessment, where the terminal steady-state body burden played the same role. The
5-year average over the peak is taken to smooth out sharp peaks and more closely approximate a
plateau. The choice of peak is health protective because humans of any age must be protected
for short-term exposures, and the daily intake achieving a given TCDD blood concentration is
smallest when matched to the peak exposure as opposed to an average over shorter durations.
Thus, target concentrations for any exposure duration of less-than-lifetime must be averaged
backwards from the end of the lifetime scenario, rather than from the beginning. The only
exception would be if the short-term endpoints evaluated in the animal bioassay were associated
with a specific life stage (such as for the gestational scenario). Note that this scenario lumps all
exposures from 1 day to over 1 year in rodents into the same less-than-lifetime category.
Conceptually, duration-specific scenarios could be constructed by defining equivalent rodent and
human exposure durations. However, for the most part, defining duration equivalents across
species is a somewhat arbitrary exercise, not generally based on physiologic or toxicologic
processes, but relying primarily on fraction-of-lifetime conversions. EPA defines "lifetime"
exposure as 2 years and 70 years for rodents and humans, respectively. So, a half-lifetime
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_52 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
equivalence of 1 year in rodents and 35 years in humans is defined easily. Also, considering a
subchronic exposure to be 10-15% of lifetime, leads to an equivalence of 90 days in rodents and
7-10 years in humans. However, in the practical sense with respect to the Emond human PBPK
model predictions, the difference in the dose-to-target-concentration ratios are not significantly
different from the peak 5-year average scenario, differing by less than 5%.
The net effect of using three different scenarios for estimating the HED from rodent
exposures is that, for the same target concentration, the ratio of administered dose (to the rodent)
to HED will be larger for short-term exposures than for chronic exposures. Figure 3-31 is
similar to Figure 3-30, except that it shows the relationship of daily intake to a fixed target
TCDD blood concentration level. Figure 3-31 shows that, for human intakes of approximately
0.01 ng/kg-day, the difference in the defined scenarios is 40% or less, with a lifetime-scenario
daily intake of 0.014 ng/kg-day required to reach the same target concentration for a shorter-term
exposure of 0.01 ng/kg-day. The corresponding daily intake for the gestational scenario is
0.011 ng/kg-day. Because of the nonlinearities in the Emond human PBPK model, the
magnitude of the difference between the lifetime and less-than-lifetime exposure scenarios
increases at lower intake levels, but not to a substantial degree.
The differential effect of short- and long-term exposures is much more accentuated at the
rodent end of the exposure kinetic modeling. Analogous to the processes described in the
previous section for first-order body burden (see Section 3.4.2.2), the TCDD blood concentration
for single exposures is essentially the immediate absorbed fraction of the administered dose,
which will be somewhat lower than the administered dose, while for chronic exposure, the
TCDD blood concentration will reflect the long-term accumulation from daily exposure, which
will be very much larger than the administered dose (expressed as a daily intake). Table 3-15
shows the overall impact of TK modeling on the extrapolation of administered dose to HED,
comparing the Emond PBPK and first-order body burden models. For comparison purposes, the
administered dose is fixed at 1 ng/kg-day for all model runs. Large animal-to-human TK
extrapolation factors (TKef) are evident for short-term mouse studies, decreasing in magnitude
with increasing exposure duration. The only exception is the slightly lower extrapolation factor
for the mouse 1-day exposure, which is the result of the relatively short TCDD half-life (10 days)
in mice and the use of the peak TCDD blood concentration as representative of single exposures,
compared to the average TCDD blood concentration over the exposure period used for multiple
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.53 DRAFT—DO NOT CITE OR QUOTE
-------
1 exposures. The TKEfs are lower for rats because of the slower elimination of TCDD in rats
2 compared to mice. Also, because of the nonlinear kinetics inherent in the Emond PBPK model,
3 the span of the HED (13-fold for mice) across these exposure durations is greater than the span
4 of the lipid-adjusted serum concentration (LASC; 4-fold for mice). Because of the dose-
5 dependence of TCDD elimination in the Emond model, the TKEf becomes smaller with
6 decreasing intake. The result of this nonlinearity is that, although Table 3-15 shows much lower
7 TKefS for the Emond PBPK model than for the first-order body burden metric, at much lower
8 HED levels the two models give much closer predictions.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.54 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 3-1. Partition coefficients, tissue volumes, and volume of distribution
2 for TCDD in humans
3
Tissue
Tissue/blood
partition
coefficient
Tissue volume
(liters, for a
60 kg person)
Effective volume of
distribution (Vd—liters of
blood equivalent)
Percent
total Vd
Blood
1
3
3
0.25
Fat
100
11.4
1.140
94.19
Liver
6
1.56
9
0.77
Rest of the body
1.5
38.64
58
4.79
Total
54.6*
1.210
100.00
4
5 *The total tissue volume presented here represents only 91% of body weight because some of the weight and
6 volume of the body is occupied by bone and other structures where TCDD uptake and accumulation do not occur to
7 a significant extent.
8
9 Source: Wang et al. (1997), Emond et al. (2005, 2006).
10
11
12 Table 3-2. Blood flows, permeability factors and resulting half lives (t1^) for
13 perfusion losses for humans as represented by the TCDD PBPK model of
14 Emond et al. (2005, 2006)
15
Tissue
Permeability (fraction of
compartment blood flow)
Rate constant for
compartmental
elimination (hour1)
ty2 (hrs)
Fat
0.12
0.0049
143
Liver
0.03
0.77
0.90
Rest of the body
0.35
3.84
0.18
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.55 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 3-3. Toxicokinetic conversion factors for calculating human equivalent
2 doses from rodent bioassays
3
Mouse
Rat (Wistar)
Rat (other)
Guinea pig
Half-life (days)3
10
20
25
40
Exposure
duration (days)
Conversion factor (CF)b BBA(tA):dA given in parentheses
1
3882 (0.77)
3815 (0.79)
3802 (0.79)
3783 (0.79)
7
1107 (2.71)
1020 (2.94)
1004 (2.99)
979 (3.07)
14
681 (4.41)
587 (5.11)
569 (5.27)
543 (5.53)
28
453 (6.62)
350 (8.56)
331 (9.06)
303 (9.90)
90
307 (9.76)
186(16.1)
163 (18.4)
130(23.0)
180
282(10.6)
154 (19.5)
129(23.2)
93 (32.1)
365
270(11.1)
141 (21.3)
115(26.0)
77 (38.9)
730
226(11.3)
115 (22.2)
93 (27.4)
60 (42.5)
4
5 aHalf-life for humans = 2,593 days (7.1 years).
6 hdH = dJCF; BBH(tH):dH = 2,185 (1-180 days), 2,202 (365 days), 2,555 (730 days).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_56 DRAFT—DO NOT CITE OR QUOTE
-------
Table 3-4. Equations used in the CADM PBPK model*
Parameter
Equation
Hepatic
Concentration
(ng/kg)
0 (f - f \*r
~-» s^body *j, / r \J max J mm' body
hepatic jrr min 1 jy ^
WI K+ body
Fat
Concentration
(ng/kg)
O (f — f
x-r s^body & /i / -f max J min / body
appose ~ Wq K l/mm 1 K + Cbody
Hepatic
Elimination
Exr_hepatic = k, .«(1 ¦- + (f-~ ))
body
Excretion via
gut of
Unchanged
TCDD
(Exsorption)
Exr _ gut = ka* Qa
Change of
TCDD due to
bodyweight
change
ChangeTCDD_BW - Qtajy * + dt)-BW(t))
Amount in
body as a
function of
Qbody (t + dt) - Qbody (0 = Exr hepatic + Exr gut + ChangeTCDD BW
time
Adipose tissue
growth
1.2 * BMI + (0.23 * Age) -10.8 * sex
100
Change of
hepatic
elimination
constant with
age
ke =ke0~keSlope* A§e
*For abbreviations and parameters, see Table 3-5.
Source: Aylward et al. (2005b).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.57 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 3-5. Parameters of the CADM model
2
Parameter
Value
Units
Comments/sources
fhmin
0.01
unitless
Minimum body burden fraction in liver
fhmax
0.7
unitless
Maximum body burden fraction in liver
K
100
ng/kg
Body burden at half-maximum of fraction
liver
ke
Calculated
per year
ke = ke0 - ke siope*(age) with enforced
minimum of ke min
keo
0.85
per year
CADM-mean hepatic elimination base rate at
age 0
ke slope
0.011
per year
Change in ke per year of age
\c
mm
0.2
per year
Minimum hepatic elimination rate
wa (adipose weight fraction)
Calculated
unitless
wa = [(1.2*BMI)+0.23*Age-10.8*sex]/100
wh (liver body weight fraction)
0.03
unitless
Assumed constant
k:i (adipose clearance factor)
0.0025
per month
Passive elimination rate from intestinal tract
Monthly dose
0.15507069
ng
per month
Estimated absorption fraction
0.97
unitless
From Moser and McLaghlan (2001)
Body weight
70
kg
Standard male weight
Sex
1
unitless
1 = male; 0 = female
Time of administration
840
months
Initial Cbody
0.2
ng/kg
Estimated background young adults UMDES
sampling
Absorbed monthly dose 1
0.150418569
ng
per month
3
4 Note: The values of fhmin, fhmax, and K were obtained by best fit of the model simulations to the experimental data
5 with the method of least squares (Carrier et al., 2005a).
6
7 Source: Aylward et al. (2005b).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_58 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 3-6. Confidence in the CADM model simulations of TCDD dose
2 metrics
3
Dose metric
Level of confidence
Administered dose
N/A
Absorbed dose
H
Body burden
H
Serum lipid concentration
M
Total tissue (liver) concentration
L
Receptor occupancy (bound concentration)
N/A
4
5 H = high, M = medium, L = low, NA = not applicable.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3.59 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 3-7. Equations used in the TCDD PBPK model of Emond et al. (2006)
2
Aspect
Equation
Body weight
growth with age
J>ur t \ J>ur ™ ( OAlxtime }
BW, (g) = BW TOx
"me ~ ^1402.5 + time J
Cardiac output
( BW Y'75
Qc(mL / hr) = QCCAR x 60
A factor of 60 corresponds to the conversion of minutes to hours, and 1,000 is conversion of BW
from grams to kilograms.
Blood
compartment
\{Qf x Cft>) + (Ore x Creb) + (01 i xClib) + lymph] (CbxCLURI)
\^u\mnui / itils) —
Qc Qc
Tissue compartment (fat, rest of the body)
Tissue blood
subcompartment
dAtb (nmollmL) = Qt(Ca Ctb) PAt\ctb
Ath
Ctb(nmol / mL) =
Wtb
Tissue cellular
matrices
~~~(nmol/mL) = PAt^Ctb -
At
CtinmollmV) - —
Wt
Liver tissue compartment
Tissue blood
subcompartment
dAUb ^nmoi / mj_^ _ QH(ca _ Clib) - PALI (Clib - Clifree) + input oral
dt
Clib(nmol / mL) =
WLIB
Tissue cellular
matrices
(nmol / mL) = PALI (Clib - Clifree) - (KBIIE II x Clifree x WII)
dt
Ali
Cli(nmol / mL) =
Wli
Free TCDD
concentration in
liver
[ (IIBMAX x Clifree \ f CYPXA2 x Clifree \\
Clifreeinmol / ml) = Cli - Clifree x PII + +
^ KDII +Clifree J { KDLI1A2 + Clifree J
Concentration
bound to AhR
in hepatic tissue
, , rx LIBMAX x Clifree
CtAmbound(nm°llmL)= nvf
KDLI + Clifree
All other induction processes and equations have been described and presented by Wang et al.
(1997).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_60 DRAFT—DO NOT CITE OR QUOTE
-------
1
Table 3-7. Equations used in the TCDD PBPK model of Emond et al. (2006)
(continued)
Aspect
Equation
Gastrointestinal absorption and distribution of TCDD to the portal lymphatic circulation
Amount of
TCDD
remaining in
lumen cavity
dLu™en (nmoi / /;/-) = [(KST + KABS) x lumen J + intake
Lumen in the amount of TCDD remaining in the GI tract (nmol); intake is the rate of intake of
TCDD during a subchronic exposure (nmol/hr).
Amount of
TCDD
eliminated in the
feces
dFeces nmQ^ x lumen
dt
Absorption rate
of TCDD to the
blood via the
lymphatic
circulation
(jimoi / _ KABS x lumen x 0.7
Absorption rate
of TCDD by the
liver via portal
circulation
dPortal . , ., . „ , ,
(nmol / nr) = KABS x lumen x 0.3
dt
2
3 Note: Key parameters and abbreviations are defined in Table 3-10.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3_6l DRAFT—DO NOT CITE OR QUOTE
-------
Table 3-8. Parameters of the PBPK model for TCDD
Parameter
Description
Symbol
Parameter values
Human
nongestational3
Human
gestational"
Mouse
nongestational
Mouse
gestational
Rat
nongestational
Rat
gestational
Body weight (g)
BW
Calculated
Calculated
23-28b
23-28
125-250b
85-190b
Cardiac output (mL/hour/kg)
QCCAR
15.36cd
Calculated
275°
275°
311.4e
311,4e
Tissue (intracellular) volumes (fraction of BW)
Liver
WLIO
Calculated
Calculated
0.0549f
0.0549f
0.036e
0.036e
Fat
WFO
Calculated
Calculated
0.069e
Calculated
0.069e
Calculated
Tissue blood volumes
Liver (fraction of WLIO)
WLIBO
0.266e
0.266e
0.266e
0.266e
0.266e
0.266e
Fat (fraction of WFO)
WFBO
0.05e
0.05e
0.05e
0.05 e
0.05e
0.05e
Rest of body (fraction of WREO)
WREBO
0.03e
0.03e
0.03e
0.03 e
0.03e
0.03e
Placenta tissue fraction of tissue blood
weight (unitless)
WPLABO
N/A
0.5s
N/A
0.5e
N/A
0.5e
Tissue blood flow (fraction of cardiac output)
Liver
QLIF
0.26°
0.26°
0.16 lf
0.161f
0.183e
0.183e
Fat
QFF
0.05°
0.05°
0.07h
0.07h
0.069e
0.069e
Placenta
QPLAF
N/A
Calculated
N/A
Calculated
N/A
Calculated
Tissue permeability (fraction of tissue blood flow)
Liver
PALIF
0.35e
0.35e
0.35e
0.35e
0.35e
0.35e
Fat
PAFF
0.121
0.121
0.121
0.121
0.091e
0.091e
Placenta diffusional permeability fraction
(unitless)
PAPLAF
N/A
0.3s
N/A
0.03s
N/A
0.3s
Rest of body
PAREF
0.03e
0.03e
0.03e
0.03e
0.0298e
0.0298e
-------
0
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
^s
^s
TO
TO'
*
^5
«
0
LtJ £j
1 TO
On ^
uj 0
S
1
a,
2 §-
7? ^
J> ^5
Tl S
H ©
3 §
2 §
O ^
H r?
<*^
hh Oq
H to
W|
O £?
o
H
W
Table 3-8. Parameters of the PBPK model for TCDD (continued)
Parameter
Description
Symbol
Parameter values
Human
nongestational3
Human
gestational"
Mouse
nongestational
Mouse
gestational
Rat
nongestational
Rat
gestational
Partition coefficient
Liver
PLI
6e
e
e
e
e
e
Fetus/blood partition coefficient (unitless)
PFETUS
N/A
4J
N/A
4J
N/A
4J
Placenta/blood partition coefficient (unitless)
PPLA
N/A
1.5J
N/A
3s
N/A
1.5J
Fat
PF
100&
ioos
4006
4006
1006
100e
Rest of body
PRE
1.5e
1.5e
k
k
1.5e
1.5e
Metabolism constants
Urinary clearance elimination (mL/hour)
CLURI
4.17E-081
4.17E-081
0.091
0.091
0.01J
0.01J
Clearance - transfer from mother to fetus
(mL/hour)
CLPLAFET
N/A
16e 3
N/^3
0.171
N/A
0.171
Liver (biliary elimination and metabolism;
hour"1)
KBILELI
Inducible
Inducible
Inducible
Inducible
Inducible
Inducible
Interspecies constant (hour1)
Kelv
0.00111
0.00111
0.41
0.41
0.15e
0.15e
AhR
Affinity constant in liver (nmol/mL)
KDLI
o.r
o.r
o.ooor
o.ooor
o.ooor
o.ooor
Binding capacity in liver (nmol/mL)
LIBMAX
0.35e
0.35e
0.00035e
0.00035e
0.00035e
0.00035e
Placenta binding capacity (nmol/mL)
PLABMAX
N/A
0.2J
N/A
0.0002J
N/A
0.00021
Affinity constant protein (AhR) in placenta
(nmol/mL)
KDPLA
N/A
0.1J
N/A
0.0001J
N/A
0.0001J
-------
0
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
^s
^s
TO
TO'
*
^5
«
0
LtJ £j
1 TO
On ^
¦J^ O
s
1
a,
2 §-
7? ^
J> ^5
Tl S
H ©
3 §
2 §
O ^
H r?
<*^
hh Oq
H to
W|
O £?
o
H
W
Table 3-8. Parameters of the PBPK model for TCDD (continued)
Parameter
Description
Symbol
Parameter values
Human
nongestationala
Human
gestational"
Mouse
nongestational
Mouse
gestational
Rat
nongestational
Rat
gestational
CYP1A2 induction parameters
Dissociation constant CYP1A2 (nmol/mL)
KDLI2
401
401
0.02'
0.02'
0.04J
0.04J
Degradation process CYP1A2 (nmol/mL)
CYP1A210UTZ
l,600e
l,600e
1.6e
1.6e
1.6e
1.6e
Dissociation constant during induction
(nmol/mL)
CYP1A21EC50
130e
130e
0.13s
0.13e
0.13s
0.13s
Basal concentration of CYP1A2 (nmol/mL)
CYP1A21A2
l,600e
l,600e
1.5k
1.5k
1.6e
1.6e
First-order rate of degradation (hour1)
CYP1A21KOUT
o.r
o.r
o.r
o.r
o.r
o.r
Time delay before induction process (hour)
CYP1A21TAU
0.25e
0.25e
1.5k
1.5k
0.25e
0.25e
Maximal induction of CYP1A2 (unitless)
CYP1A21EMAX
9,300'
9,300'
600e
600e
600e
600e
Other constants
Oral absorption constant (hour1)
KABS
0.061
0.06'
0.48'
0.48'
0.48e
0.48e
Gastric nonabsorption constant (hour1)
KST
0.01m
0.01m
0.30'
0.30'
0.36e
0.36e
aUnits for human nongestational parameters are L rather than mL and kg rather than g where applicable.
bBody weight varies by study (Emond et al., 2004).
°Krishnan (2008).
dUnits are L/kg/hr.
eWang et al. (1997).
fILSI (1994).
8Fixed.
hLeung et al. (1990).
'Optimized.
JEmond et al. (2004).
kWang et al. (2000).
'Lawrence and Gobas (1997).
""Calculated to estimate 87% bioavailability of TCDD in humans (Poigerand Schlatter, 1986).
-------
1 Table 3-9. Regression analysis results for the relationship between logio
2 serum TCDD at the midpoint of observations and the logio of the rate
3 constant for decline of TCDD levels using Ranch Hand data
4
Item
Aspect
Value
Summary of fit
RSquare
0.894
RsquareAdj
0.871
Root mean square error
0.044
Mean responses
0.130
Observations (or sum weights)
11
Parameter estimates
Intercept
Estimate
-0.054
Standard deviation
0.026
t ratio
-2.07
Prob>t
0.0679
Log (TCDDpg/g)
Estimate
0.092
Standard error
0.011
t ratio
8.28
Prob>t
<0.0001
5
6
7 Table 3-10. Confidence in the PBPK model simulations of TCDD dose
8 metrics
9
Dose metric
Human model
Rat model
Mouse model
Administered dose
N/A
N/A
N/A
Absorbed dose
H
H
M
Body burden
H
H
M
Serum (blood)concentration
H
H
M
Total liver concentration
M/L
H
M
Receptor occupancy (bound concentration)
L
L
L
10
11 H = high, M = medium, L = low.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-65 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 3-11. Overall confidence associated with alternative dose metrics for
2 cancer and noncancer dose-response modeling for TCDD using rat PBPK
3 model
4
End point
Body
burden
Blood or serum
concentration
Liver
concentration
Bound
concentration in
liver
Liver effects
M
II
M 1.
Nonhepatic effects
M
II
M 1.
5
6 H = high, M = medium, L = low.
7
8
9 Table 3-12. Overall confidence associated with alternative dose metrics for
10 cancer and noncancer dose-response modeling for TCDD using mouse PBPK
11 model
12
End point
Body
burden
Blood or serum
concentration
Liver
concentration
Bound
concentration in
liver
Liver effects
M
M
1.
Nonhepatic effects
M
M
1.
13
14 H = high, M = medium, L = low.
15
16
17 Table 3-13. Contributors to the overall uncertainty in the selection and use
18 of dose metrics in the dose-response modeling of TCDD based on rat and
19 human PBPK models
20
Dose metric
Conceptual uncertainty
Prediction uncertainty
Administered dose
H
NA
Absorbed dose
H
L
Body burden
M
L
Blood or serum concentration
M
L
Tissue concentration
L
M
Receptor occupancy
L(?)
H
21
22 H = high, M = medium, L = low, NA = not applicable, ? = if relevant to MOA of response.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-66 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 3-14. Contributors to the overall uncertainty in the selection and use
2 of dose metrics in the dose-response modeling of TCDD based on mouse and
3 human PBPK models
4
Dose metric
Conceptual uncertainty
Prediction uncertainty
Administered dose
H
NA
Absorbed dose
H
L
Body burden
M
M
Blood or serum concentration
M
M
Tissue concentration
L
MH
Receptor occupancy
L(?)
H
5
6 H = high, M = medium, L = low, NA = not applicable, ? = if relevant to MOA of response.
7
8
9 Table 3-15. Impact of toxicokinetic modeling on the extrapolation of
10 administered dose to HED, comparing the Emond PBPK and first-order
11 body burden models
12
Exposure
duration (days)
lst-order BB
Emond PBPK
HED
(ng/kg-day)
TKEF
LASC
(ng/kg)
HED
(ng/kg-day)
TKEF
Mouse
1
2.57E-4
3,882
75.5
9.49E-4
1,054
14
1.47E-3
681
64.4
8.17E-4
1,224
90
3.25E-3
307
173
3.83E-3
261
365
3.70E-3
270
248
6.66E-3
150
730
4.43E-3
226
263
1.08E-2
93
Rat
1
2.63E-4
3,802
110
1.87E-3
535
14
1.76E-3
569
208
5.22E-3
192
90
6.13E-3
163
599
2.81E-2
36
365
8.68E-3
115
811
4.52E-2
22
730
1.07E-2
93
853
6.47E-2
15
13
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-67 DRAFT—DO NOT CITE OR QUOTE
-------
4
D 7 Day Liver/Fat
® 14 Day Liver/Fat
A 21 Day Liver/Fat
35 Day Liver/Fat
0
~
Dose jug/Kg
Figure 3-1. Liver/fat concentration ratios in relation to TCDD dose at
various times after oral administration of TCDD to mice.
Source: Dilberto et al. (1995).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-68 DRAFT—DO NOT CITE OR QUOTE
-------
C.25'
111
4,000
INITIAL SERUM LIPID TCDD LEVEL
(ppt)
1
2 Figure 3-2. First-order elimination rate fits to 36 sets of serial TCDD
3 sampling data from Seveso patients as function of initial serum lipid TCDD.
4
5 Source: Aylward et al. (2005).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-69 DRAFT—DO NOT CITE OR QUOTE
-------
1
0.05
u
es
4>
"«
u
o>
u.
u
es
•-
U.
0.04 "
0.03 -
0.02
y = 0.09735 - 0.00282x RA2 = 0.752
% Body Fat
2
3 Figure 3-3. Observed relationship of fecal 2,3,7,8-TCDD clearance and
4 estimated percent body fat.
5
6 Source: Rohde et al. (1999).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-70 DRAFT—DO NOT CITE OR QUOTE
-------
1
11.0
y = - 1.89 + 0.314x RA2 = 0.998
CAj
u
C5
C5
a
S3
C5
S3
W
—
—
u
H
10.0
Seveso Females
Ranch Hand Males
Seveso Males
Error bars are ± 1 standard error
% Body Fat
2
3 Figure 3-4. Unweighted empirical relationship between percent body fat
4 estimated from body mass index and TCDD elimination half-life—combined
5 Ranch Hand and Seveso observation.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-71 DRAFT—DO NOT CITE OR QUOTE
-------
01
u
c
S
0i
"5
m
»
Functional biomarkers
Receptor occupancy
Total tissue concentration
Blood or scrum concentration
Body burden
Absorbed dose
Intake
1
2 Figure 3-5. Relevance of candidate dose metrics for dose-response modeling,
3 based on mode of action and target organ toxicity of TCDD.
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-72 DRAFT—DO NOT CITE OR QUOTE
-------
0 5?
>5*
§•
to
s
a
to
2
>;*
ss»
2r
%
>4
TO
jg
TO'
86
^s
5S
§
0
UJ £j
1 TO
--0 S"
feS o
1
Su
§ §•
c w
ln a
H ©
. . ^
TO
O
JS
!z! §
O 3~
H
HH OO
H TO
W|
O g?
o
H
W
o
o
Experimental Appliedj^)^
Body BurdenRat{t) = BB( 0)e ""' +
kt , rf(l-0/fl
Human
du = d
'1/2,1
(1
k jt A
)
h,iH (1" )
Estimated »-
Exposure
BurdenRat (t)
A
V
Body BurdenHuman{t)
Figure 3-6. Process of estimating a human-equivalent TCDD lifetime average daily oral exposure (dH) from an
experimental animal average daily oral exposure (d\) based on the body-burden dose metric. The arrows
represent mathematical conversions based on toxicokinetic modeling. BBa (TWA animal body burden) and BBH (TWA
human body burden) are assumed to be toxicokinetically equivalent. See text for further explanation.
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
o
o
o
ro
o
o
in
CM
o
o
o
CM
£=
"O
=3
JD
>>
TD
O
m
o
o
m
o
o
o
" target body burden
" chronic exposure (BB:d = 2555)
half-chronic exposure (BB:d = 2202)
shorter exposures (BB:d = 2185)
o
o -
m
o -
0
5000
10000
15000
20000
25000
Exposure days
Figure 3-7. Human body burden time profiles for achieving a target body
burden for different exposure duration scenarios. BB:d is BBH(tH):dH in
Figure 3-6. The curve depicted using the solid line illustrates the increase in the
human body burden over time for a hypothetical human administered a daily
TCDD dose where the time-weighted average human body burden estimate over
the lifetime is equal to the target body burden attained in a rodent bioassay. When
compared to shorter durations (dashed lines), a higher average daily TCDD dose
is required to yield a time-weighted average human body burden over a lifetime
that is equal to the target body burden attained in a rodent bioassay. The half-
chronic exposure scenario (depicted using a dashed line) is equivalent to a 1-year
exposure in rodents. When compared to a chronic BBH, a lower value of dH is
needed to attain the target body burden in a rodent bioassay when the time-
weighted average is over the last 35 years of life; the dose to plateau ratio is also
smaller (i.e., dH.c < dH.sc to attain the target body burden in a rodent bioassay).
The shorter exposure scenario is equivalent to most other shorter rodent exposure
durations, from 1 day to subchronic, which are indistinguishable with respect to
the BB:d ratio (subchronic shown).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-74 DRAFT—DO NOT CITE OR QUOTE
-------
DISTRIBUTION ELIMINATION
Tissue-Specific
Concentration-Dependant
Hepatic metabolism
with first-order rate
^ constant^
ABSORPTION
Fecal excretion
with the first-order
^ rate constant k,,^-
TCDD
Blood
Adipose Burden
a»=a(o*[i-/,(c,)]
Liver Burden
Qh(t) = Qb(t)*fh(Cb)
Figure 3-8. Schematic of the CADM structure.
Source: Aylward et al. (2005).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-75 DRAFT—DO NOT CITE OR QUOTE
-------
-tr
10
100
cb
1000
1 Figure 3-9. Comparison of observed and simulated fractions of the body
2 burden contained in the liver and adipose tissues in rats. fraction contained
3 in liver (observation) (~X./h-sim, fraction contained in liver (simulation) (—);/at,
4 fraction contained in the adipose tissue (observation) (0);/at-sim, fraction contained
5 in the adipose tissue (simulation) (—); and Cb. body concentration in ng TCDD/kg
6 body wt.
7
8 Source: Carrier et al. (1995a); data from Abraham et al. (1988) measured 7 days
9 after dosing.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-76 DRAFT—DO NOT CITE OR QUOTE
-------
Blood systemic circulation
Urinary
excretion
GI tract
elimination
Oral
absorption
Liver
Cellular matrices
Cellular matrices
Blood tissue
of body
Cellular matrices
Cellular matrices
Blood tissue
Fat
Cellular matrices
Cellular matrices
Blood tissue
1
2
3 Figure 3-10. Conceptual representation of PBPK model for rat exposed to
4 TCDD.
5
6 Source: Emond et al. (2006).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-77 DRAFT—DO NOT CITE OR QUOTE
-------
Urinary
excretion
Elimination Gl tract
Oral absorption
Portal vein
Liver (AhR and CYP1A2 induction)
Rest of body
Placenta (AhR)
Fetus
Figure 3-11. Conceptual representation of PBPK model for rat
developmental exposure to TCDD.
Source: Emond et al. (2004).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-78 DRAFT—DO NOT CITE OR QUOTE
-------
Tissue blood
Plasma
proteins
-AB
Bb
Bf
TB
PA
Water
fraction
Lipid fraction
Ks
Lip
Kn
TCDD-Ah
Tnb
Nonspecific
bound
CYP1A2-TCDD
Tissue
Figure 3-12. TCDD distribution in the liver tissue.
Source: Wang et al. (1997).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-79 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
10 1S
lime fctay#
10 15
Tinw kljys}
10 IS
Tint* uljyia
10 15
Time icljya
5
; ooa
•n
% 0 06
i
6
- o.w
z
I 002
Figure 3-13. Growth rates for physiological changes occurring during
gestation, (a) Placental growth during gestation (calculated for n = 10 placenta).
Experimental data from Sikov (1970). (b) Blood flow rate in Placental
compartment during gestation. Experimental data from Buelke-Sam et al.
(1982a, b). (c) Fat fraction of body weight during gestation. Experimental data
came from Fisher et al. (1989), and (d) Fetal growth during gestation.
Experimental data obtained from Sikov (1970).
A
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-80 DRAFT—DO NOT CITE OR QUOTE
-------
Fixed elimination rate
EXPEL
ffl O.OOI -
400 60CI
Time (for)
1.000
o
£
— 0.1-
c
-------
1
2
100.00
g- 10.00
o
~ 1.00
¦
o
g 0.10
J—
0.01
0 5 10 15 20 25 30 35
Time (week)
3 Figure 3-15. PBPK model simulation of hepatic TCDD concentration (ppb)
4 during chronic exposure to TCDD at 50,150, 500,1,750 ng TCDD/BW using
5 the inducible elimination rate model compared with the experimental data
6 measured at the end of exposure.
7
8 Source: Emond et al. (2006).
1,750 ng TCDD/kg BW
^ 500 ng TCDD/kg BW ~
150 ng TCDD/kg BW ~
50 ng TCDD/kg BW
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-82 DRAFT—DO NOT CITE OR QUOTE
-------
1
10000
1000
100
10
1
0
10
20
Time (year)
30
40
10000
CBPPTRH
¦ V46286
1000
100
10
1
0
5
10
15
Time (year)
20
25
30
10000
CBPPTRH
¦ V33457
1000
100
10
1
0
10
20
Time (Year)
30
40
Q
-CBPPTRH
V36564 -
D
20
Time (year)
10000
CBPPTRH
1000
¦ V30208
100
10
1
0
5
10
15 20
Time (year)
25
30
35
10000
CBPPTRH
¦ V52064
1000
100
10
1
0
5
10
15
20
25
Time (year)
10000
CBPPTRH
¦ V49044
1000
100
10
1
0
5
10
15 20
Time (year)
25
30
35
100000
.CBPPTRH
10000
¦ V30172
1000
100
10
0
10
20
30
40
Time (year)
000000
CBPPTRH
100000
V31731
10000
m
20
Time (year)
1000000
100000
; 10000
1000
100
10
1
m
-CBPPTRH
V30991
20
Time (year)
30
40
2 Figure 3-16. Model predictions of TCDD blood concentration in 10 veterans
3 (A-J) from Ranch Hand Cohort.
4 Source: Emond et al. (2005).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-83 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
0
1
1,000,000
"O
J 10,000
-Q
o
Q
U
I-
1,000
0 100 200 300 400 500 600 700 800 900 1,000 1,100 1,200
J^^^^^^Bloo^TTodehjrediction^p^l^jj^^^^^Patien^^Vienn^vQmef^^^^^lj
1,000,000
100,000
1,000 —i—i—i—j—i—i—i—i—i—i—i—j—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—¦—¦—i—i—i—i—¦—j—i—¦—¦—i—i—r—«—j—r
0 100 200 300 400 500 600 700 800 900 1,000 1,100 1,200
Blood model predictions (pt #2) @ Patient 2 Vienna women
Figure 3-17. Time course of TCDD in blood (pg/g lipid adjusted) for two
highly exposed Austrian women (patients 1 and 2). Symbols represent
measured concentrations, and lines represent model predictions. These data were
used as part of the model evaluation (Geusau et aL 2002).
Source: Emond et al. (2005).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-84 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
12
y = 11.8 - 0.48x R"2 = 0.999 ~ ln(ptl Em Mod Sim pg/g TCDD)
y = 11.8 - 0.45x R"2 = 0.772 ¦ ln(ptl Obs pg/g TCDD)
y = 10.5 - 0.43x R"2 = 1.000 01n(pt2 Em Mod Sim pg/g TCDD)
y = 10.0 - 0.24x R"2 = 0.612 •ln(pt2 Obs pg/g TCDD)
Years After Exposure
3
4 Figure 3-18. Observed vs. Emond et al. (2005) model simulated serum
5 TCDD concentrations (pg/g lipid) over time (In = natural log) in two
6 Austrian women. Data from Geusau et al. (2002).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-85 DRAFT—DO NOT CITE OR QUOTE
-------
Emond Mod Sim Ln(Decline/Yr) y = -0.101 + 0.123x RA2 = 0.995
• Observed Ln(Decline/Yr) y= - 0.054 + 0.092x RA2 = 0.884
Log(TCDD pg/g at Midpoint Obs)
1
2 Figure 3-19. Comparison of the dose dependency of TCDD elimination in the
3 Emond model vs. observations of nine Ranch Hand veterans and two highly
4 exposed Austrian patients. Circles are observed data.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-86 DRAFT—DO NOT CITE OR QUOTE
-------
to
to
I
s
Q_
WHO
WFO
PF
Ksr
KABS
CYP1A2JOUTZ
CYP1A2JA2
~
~~r
s
~~r
-6
—T
-i
~r
-2
T
0 2
Percent of change
n
10
3
4
5
6
7
0-
WLIO
WFO
PF
KST
KABS
KD LI
LIB MAX
KDLI2
CTF1A2_10UTZ
CYP1A2JEHAX
CYP1A2JEC50
CYP1A2_1A2
a
-20
-15
~i 1 1 1 r~
-10 -5 0 5 10 15
Percent of change
n
20
Figure 3-20. Sensitivity analysis was performed on the inducible elimination
rate. The analysis was performed at 0.001 |ig/kg (A) and at 10 |ig/kg (B).
Source: Emond et al. (2006).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-87 DRAFT—DO NOT CITE OR QUOTE
-------
jll kjLUJujrijJiii uui iiummii uiuu.iiiJi^-5
0 ID 23 id 10 «& 6ii 70 80 30 LOO Lift I Hi
v — cb {po^a)
v M Cb motuftd
v -—- CI" (MO) 5.™
v Cli (pool1 measured
1
2
3
4
5
6
7
C
10,000
£,003
JOO
10
1
¦ Cf (po/dl Simulated I
| Cf foa^a! |
2d
Figure 3-21. Experimental data (symbols) and model simulations (solid lines)
of (A) blood, (B) liver and (C) adipose tissue concentrations of TCDD after
oral exposure to 150 ng/kg-day, 5 days/week for 17 weeks in mice. Y-axis
represents concentration in pg/g and X-axis represents time in days.
Source: Experimental data were obtained form Diliberto et al (2001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-88 DRAFT—DO NOT CITE OR QUOTE
-------
Q.
Q.
C
o
c
d)
o
c
o
o
d)
>
10000000
1000000
100000
10000
1000
100
10
~ Measured
~ Simulated
¦
,ri
0.001 0.01
0.1 1 10
Dose (ug/kg)
100
300
1
2 Figure 3-22. Comparison of PBPK model simulations with experimental
3 data on liver concentrations in mice administered a single oral dose of
4 0.001-300 jig TCDD/kg. The simulations and experimental data were obtained
5 24 hour post-exposure.
6
7 Source: Data obtained from Boverhoff et al. (2005).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-89 DRAFT—DO NOT CITE OR QUOTE
-------
100000
10000
1000
100
10
1
cb sim
¦ cb measured
cli sim
¦ cli measured
0
cf sim
¦ cf measured
0
0.
10.0
Dose (ng/kg)
100.0
1000.0
1
2 Figure 3-23. Comparison of model simulations (solid lines) with
3 experimental data (symbols) on the effect of dose on blood (cb), liver (cli) and
4 fat (cf) concentrations following repetitive exposure to 0.1-450 ng TCDD/kg,
5 5 days/week for 13 weeks in mice.
6
7 Source: Data obtained from Diliberto et al. (2001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-90 DRAFT—DO NOT CITE OR QUOTE
-------
A
— C& (cs/fl) |
w m Cb rrveavured 1
UJJ-LmU- .mil ^iin
B
10 20 3d 40 50 60 70 SO *f> 100 L10 L20
ff
¦
¦
¦
¦
• Cli (m) S«rri I
| Cli {ugy '• ri-.eoiurei! I
0 10 20 30 40 Sfl €0 ?0 SO $0 SOO ltfl 120
1
2
3
4
5
6
7
C
- Cf (sew s>mgi#t*d I
[ Cf (pff/tf) iric.aii.ur fed |
o io 20 ao 40 za so ?n so *a soo no 120
Figure 3-24. Comparison of experimental data (symbols) and model
predictions (solid lines) of (A) blood, (B) liver and (C) adipose tissue
concentrations of TCDD after oral exposure to 1.5 ng/kg-day, 5 days/week
for 17 weeks in mice. Y-axis represents concentration in pg/g and X-axis
represents time in days.
Source: Experimental data were obtained form Diliberto et al. (2001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-91 DRAFT—DO NOT CITE OR QUOTE
-------
A
,111 illli lit
Jin,
4
• m mi cm ¦ I am tjtm im : «w )¦
I
D
M W IH U9M UBI I.4N l.«M l.«* Z.tti 2.H4
I M 4« M M IM LJv> L«H l.Mt UC* IMft
1
2 Figure 3-25. Comparison of experimental data (symbols) and model
3 predictions (solid lines) of (A) blood concentration, (B) liver concentration,
4 (C) adipose tissue concentration (D) feces excretion (% dose) and (E) urinary
5 elimination (% dose) of TCDD after oral exposure to 1.5 ng/kg-day,
6 5 days/week for 13 weeks in mice. Y-axis represents concentration in pg/g and
7 X-axis represents time in days.
8 Source: Experimental data were obtained form Diliberto et al. (2001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-92 DRAFT—DO NOT CITE OR QUOTE
-------
1
1
1
1
1
i
u
-¦
ill
il
M
11
J
Ml
¦
u
V
6 Hi ,3« l,4M
tm J»t
1
2
3
4
5
6
7
Figure 3-26. Comparison of experimental data (symbols) and model
predictions (solid lines) of (A) blood concentration, (B) liver concentration,
(C) adipose tissue concentration (D) feces excretion (% dose) and (E) urinary
elimination (% dose) of TCDD after oral exposure to 150 ng/kg-day,
5 days/week for 13 weeks in mice. Y-axis represents concentration in pg/g and
X-axis represents time in days.
Source: Experimental data were obtained form Diliberto et al. (2001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-93 DRAFT—DO NOT CITE OR QUOTE
-------
c
D
Figure 3-27. PBPK model simulations (solid lines) vs. experimental data
(symbols) on the distribution of TCDD after a single acute oral exposure to
A-B) 0.1, C-D) 1.0 and F.-F) 10 ^g of TCDD/kg of body weight in mice.
Liver and adipose concentration for each dose was measured after 72 hours.
Y-axis represents the concentration in tissues (ng/g); insets A, C, and E represent
liver tissue, whereas B, D, and F correspond to adipose tissue. X-axis represents
the time in hours.
Source: experimental data were obtained from Santostefano et al. (1996).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-94 DRAFT—DO NOT CITE OR QUOTE
-------
3.2
y
2-fl
it.
2-4
2&
2
1.1
1.6
1,4
1J
I
0 9
fl.fi
ft,*
4.2
¦
\
j%r
m
a
k.
¦
——
M-
111 262 2*4 2M 2M 2"HS 242 244 2M 23S 3.5ft 362 *04 3M 36fc 2l6 Jl2 3S4 3lfi ili 3
f — Cfr (flft'fl) j
mf ¦ <" vp-pn.TfrirAl I
* — cli (ivfl/g) j
* ¦ tKptr'mtntfel I
280 2&2 2B4 2Bfe .2 SB 25«0 292 234 29fc 2?B 3W 302 304 30& 30*8 310 312 31* 31fc 31B 320
1
2
3
4
5
6
7
v —¦ cf {fi&'g) j
v ¦ E *,|»-cnrnenlal I
230 253 284 2-B« 2OT 2W 212 J?4 2?& 2?8 300 302 304 306 3 OS 310 312 314 316 313 320
Figure 3-28 PBPK model simulation (solid lines) vs. experimental data
(symbols) on the distribution of TCDD after a single dose of 24 jig/kgBW on
GD 12 in mice. Concentrations expressed as ng TCDD/g tissue. (A) maternal
blood, (B) maternal liver and (C) maternal adipose tissue. Y-axis represents the
tissue concentration whereas X-axis represents the time in hours.
Source: Experimental data were obtained from (Abbott et al.. 1996).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-95 DRAFT—DO NOT CITE OR QUOTE
-------
CADM (human)
Emond (human)
CADM (rat)
Emond (rat)
o>
D>
D)
>
T>
O
m
v
Q.
C
V
T>
m
>
T>
o
m
1
1
1
1
1
1
1
1
.E+07
.E+06
.E+05
.E+04
.E+03
.E+02
.E+01
.E+00
100
Intake (ng/kg-day)
1000
10000
1
2 Figure 3-29. Comparison of the near-steady-state body burden simulated
3 with CADM and Emond models for a daily dose ranging from 1 to
4 10,000 ng/kg-day in rats and humans. The rat model was run for 13 weeks and
5 the human model was run from age 20 to 30. The time-averaged concentration
6 was used for each.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 3-96 DRAFT—DO NOT CITE OR QUOTE
-------
1—'
1 .
o
>;*
&
O
Cl
s;
3
^
a
D)
>;*
O)
£=
C
5f
O
U—<
to
-i—'
c
CD
o
c
o
o
¦
~o
o
*
o
_Q
s;
Q
Q
O
o
1-
LtJ >l
1 ^
« o
3
s
3 §-
^ >3
nr\ S
H ©
o 8
o I
H 3
HH (>Q
H
W|
o ^
o
H
W
o
o
CM
O
in
peak 5-year average concentration
gestational average concentration
lifetime average concentration
o
o
o
in
Concentration-time profile
Less-than-lifetime scenario
Gestational scenario
Lifetime scenario
20
—r~
60
Year
Figure 3-30. TCDD serum concentration-time profile for lifetime, less-than-lifetime and gestational exposure
scenarios, with target concentrations shown for each; profiles generated with Emond human PBPK model.
-------
Q)
Q)
C
C
o
c
CD
O
c
o
o
"a
o
o
_Q
O
Q
O
o
LO
CM
O
O
CM
O
LO
O
O
O
LO
Lifetime scenario
Less-than-lifetime scenario
Gestational scenario
Target concentration
~r~
40
T
60
20
Year
Figure 3-31. TCDD serum concentration-time profile for lifetime, less-than-lifetime and gestational exposure
scenarios, showing continuous intake levels to fixed target concentration; profiles generated with Emond human
PBPK model.
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
4. CHRONIC ORAL REFERENCE DOSE
This section presents U.S. Environmental Protection Agency (EPA)'s response to the
National Academy of Sciences (NAS) recommendations that EPA more explicitly discuss the
modeling of noncancer endpoints and develop a reference dose (RfD) to address noncancer
effects associated with oral 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) exposures. Section 2
details the selection of the animal studies with the lowest TCDD doses associated with the
development of adverse noncancer effects and the selection of relevant epidemiologic studies of
adverse noncancer health effects. Section 3 discusses the kinetic modeling and estimation of
human equivalent daily oral doses used in TCDD RfD development in this section. This section
discusses the modeling of noncancer health effect data associated with TCDD exposure and the
derivation of an RfD. Specifically, Section 4.1 summarizes the NAS comments on TCDD dose-
response modeling and EPA's response, including justification of selected noncancer effects and
statistical characterization of modeling results. Section 4.2 presents the TCDD dose-response
modeling undertaken for identification of candidate points of departure (PODs) for derivation of
an RfD. Finally, in Section 4.3, EPA develops an RfD for TCDD.
4.1. NAS COMMENTS AND EPA'S RESPONSE ON IDENTIFYING NONCANCER
EFFECTS OBSERVED AT LOWEST DOSES
The NAS recommended that EPA identify the noncancer effects associated with low dose
TCDD exposures and discuss its strategy for identifying and selecting PODs for noncancer
endpoints, including biological significance of the effects.
With respect to noncancer end points, the committee notes that EPA does not use
a rigorous approach for evaluating evidence from studies. . . (NAS, 2006a, p. 47).
The Reassessment should describe clearly the following aspects:
1. The effects seen at the lowest body burdens that are the primary focus for any
risk assessment—the "critical effects."
2. The modeling strategy used for each noncancer effect, paying particular
attention to the critical effects, and the selection of a point of comparison based
on the biological significance of the effect; if the ED0i is retained, then the
biological significance of the response should be defined and the precision of
the estimate given... (NAS, 2006a, p. 187).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
In this document, EPA has developed a strategy for identifying the noncancer data sets
and PODs that represent the most sensitive and biologically relevant endpoints for derivation of
an RfD for TCDD. EPA began this process by using the animal bioassays and human
epidemiologic studies that met its study inclusion criteria as sources of these data sets. For all
epidemiologic studies that were identified as suitable for further quantitative dose-response
analyses in Section 2.4.3, EPA has chosen to identify PODs (i.e., estimates of a no-observed-
adverse-effect level [NOAEL], lowest-observed-adverse-effect level [LOAEL], or benchmark
dose lower confidence bound [BMDL] when possible). Figure 4-1 shows EPA's process to
select and identify candidate PODs from these key epidemiologic studies. EPA first evaluated
the dose-response information in the study to determine whether it provided an estimate of
TCDD dose and an observed noncancer effect that was relevant for RfD derivation. If such data
were available, then EPA identified a NOAEL or LOAEL as a candidate POD. For each of
these, EPA applied a human kinetic model to estimate the continuous oral daily intake
(ng/kg-day) associated with the POD that could be used in the derivation of an RfD (see Section
4.2). If all of this information was available, then the result was included as a candidate POD.
Figure 4-2 summarizes the strategy employed for identifying and selecting candidate
PODs from the key animal bioassays identified in Section 2.4.3 for use in noncancer dose-
response analysis of TCDD. For each noncancer endpoint, EPA first evaluated the toxicologic
relevance of each endpoint, rejecting those judged not to be relevant for RfD derivation. Next,
initial PODs (NOAELs, LOAELs, and BMDLs) based on the first-order body burden metric (see
Section 3.3.4.2) and expressed as human-equivalent doses (HEDs) were determined for all
relevant endpoints (summarized in Table 4-3). Because there were very few NOAELs and
BMDL modeling was largely unsuccessful due to data limitations, the next stage of evaluation
was carried out using LOAELs only. Endpoints not observed at the LOAEL (i.e., reported at
higher doses) with BMDLs greater than the LOAEL were eliminated from further analysis, as
they would not be considered as candidates for the final POD on either a BMDL or
NOAEL/LOAEL basis (i.e. the POD would be higher than the PODs of other relevant
endpoints). In addition, all endpoints with HED estimates based on LOAELs (LOAELreds)
beyond a 100-fold range of the lowest identified LOAELred were eliminated from further
consideration, as they would not be potential POD candidates either (i.e., the POD would be
higher than the PODs of other relevant endpoints). For the remaining endpoints, EPA then
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
determined final potential PODs (NOAELs, LOAELs and BMDLs) based on TCDD blood
concentrations obtained from the Emond rodent physiologically based pharmacokinetic (PBPK)
models. HEDs were then estimated for each of these PODs using the Emond human PBPK
model. From these HEDs, a PODred was selected14 for each study as the basis for the candidate
RfD, to which appropriate uncertainty factors (UFs) were applied following EPA guidelines.
The resulting candidate RfDs were then considered in the final selection process for the RfD.
Other endpoints occurring at slightly higher doses representing additional effects associated with
TCDD exposure (beyond the 100-fold LOAEL range) were evaluated, modeled, and included in
the final candidate RfD array15 to examine endpoints not evaluated by studies with lower PODs.
In addition, BMD modeling based on administered dose was performed on all endpoints for
comparison purposes. The final array of selected endpoints is shown in Table 4-4 (summary of
BMD analysis) and Table 4-5 (candidate RfDs).
The NAS recommended that EPA better justify the selection of response levels for
endpoints used to develop risk estimates. The NAS commented on EPA's decision to estimate
an EDoi (effective dose eliciting x percent response) for noncancer bioassay/data set
combinations as a comparative tool across studies, suggesting that EPA identify and evaluate the
levels of change associated with adverse effects to define the benchmark response (BMR) level
for continuous noncancer endpoints.
The committee notes that the choice of the 1% response level as the POD
substantially affects ... the noncancer analyses.... The committee recommends
that the Reassessment use levels of change that represent clinical adverse effects
to define the BMR level for noncancer continuous end points as the basis for an
appropriate POD in the assessment of noncancer effects (NAS, 2006a, p. 72).
The committee concludes that EPA did not adequately justify the use of the 1%
response level (the ED0i) as the POD for analyzing epidemiological or animal
bioassay data for ... noncancer effects (NAS, 2006a, p. 18).
In the 2003 Reassessment (U.S. EPA, 2003), EPA was not attempting to derive an RfD
when it conducted TCDD dose-response modeling. The 2003 Reassessment developed ED0i
14In the standard order of consideration: BMDL, NOAEL, and LOAEL.
15However, studies with a lowest dose tested greater than 30 ng/kg-day were not included in the expanded
evaluation.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
estimates for noncancer effects in an attempt to compare disparate endpoints on a consistent
response scale. Importantly, the 2003 Reassessment defined the ED0i as 1% of the maximal
response for a given endpoint, not as a 1% change from control. Because RfD derivation is one
goal of this document, the noncancer modeling effort undertaken here differs substantially from
the modeling in the 2003 Reassessment.
The NAS committee was concerned with the statistical power to determine the shape of
the dose-response curve at doses far below observed dose-response information. EPA agrees
that the shape of the dose-response curve in the low-dose region cannot be determined
confidently when based on higher-dose information. An observed response above background
near (or below) the BMR level is needed for discrimination of the shape of the curve and for
accurate estimation of an EDX or BMDL. Although many of the ED0iS presented in the 2003
Reassessment were near the lowest dose tested, responses at the lowest doses were often high
and much greater than a 1% response (i.e., 1% of the maximum response). The lack of an
observed response near the BMR level is often a problem in interpretation of BMD modeling
results.
In this document, EPA has used a 10% BMR for dichotomous data for all endpoints
except for developmental study designs that incorporate litter effects, for which a 5% BMR is
used (U.S. EPA, 2000a). For continuous endpoints in this document, EPA has used a BMR of
1 standard deviation from the control mean whenever a specific toxicologically-relevant BMR
could not be defined. For the vast majority of continuous endpoints, EPA could not establish
unambiguous levels of change representative of adversity, which EPA defines as "a biochemical
change, functional impairment, or pathologic lesion that affects the performance of the whole
organism, or reduces an organism's ability to respond to an additional environmental challenge"
(U.S. EPA, 2009b). For body and organ weight change, EPA has previously established a BMR
of 10% change, which also is used in this document.
The NAS commented on EPA's development of ED0i estimates for numerous study/data
set combinations in the 2003 Reassessment, suggesting that EPA had not appropriately
characterized the statistical confidence around such model predictions in the low-response region
of the model.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
It is critical that the model used for determining a POD fits the data well,
especially at the lower end of the observed responses. Whenever feasible,
mechanistic and statistical information should be used to estimate the shape of the
dose-response curve at lower doses. At a minimum, EPA should use rigorous
statistical methods to assess model fit and to control and reduce the uncertainty of
the POD caused by a poorly fitted model. The overall quality of the study design
is also a critical element in deciding which data sets to use for quantitative
modeling (NAS, 2006a, p. 18).
EPA should ... assess goodness-of-fit of dose-response models for data sets and
provide both upper and lower bounds on central estimates for all statistical
estimates. When quantitation is not possible, EPA should clearly state it and
explain what would be required to achieve quantitation (NAS, 2006a, p. 10).
The NAS also commented that EPA report information describing the adequacy of dose-
response model fits, particularly in the low response region. For those cases where biostatistical
modeling was not possible, NAS recommended that EPA identify the reasons.
The Reassessment should also explicitly address the importance of statistical
assessment of model fit at the lower end and the difficulties in such assessments,
particularly when using summary data from the literature instead of the raw data,
although estimates of the impacts of different choices of models would provide
valuable information about the role of this uncertainty in driving the risk estimates
(NAS, 2006a, p. 73).
To address this concern, in this document EPA has reported the standard suite of
goodness-of-fit measures from the benchmark dose modeling software (BMDS 2.1). These
include chi-square ^-values, Akaike's Information Criterion (AIC), scaled residuals at each dose
level and plots of the fitted models. In some cases, when restricted parameters hit a bound, EPA
used likelihood ratio tests to evaluate whether the improvement in fit afforded by estimating
additional parameters could be justified. Goodness-of-fit measures are reported for all key data
sets in Appendix E.
4.2. NONCANCER DOSE-RESPONSE ASSESSMENT OF TCDD
This section describes EPA's current effort to conduct an evaluation of TCDD dose-
response for the noncancer endpoints from studies that met the study inclusion criteria.
Discussions include benchmark dose modeling procedures, kinetic modeling, and POD
candidates for derivation of the RfD. Section 4.2.1 describes how EPA has used physiologically-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
based pharmacokinetic (PBPK) modeling to estimate effective internal exposures as an
alternative to using administered doses or body burdens based on first-order kinetics.
Section 4.2.2 details the dose-response analysis of the epidemiologic data, with supporting
information on kinetic modeling in Appendix D. Section 4.2.3 details the dose-response analysis
for the animal bioassay data; Appendix E provides the BMDS input tables (see Section E. 1) and
output for all modeling, including blood concentrations (see Section E.2) and administered dose
(see Section E.3).
4.2.1.1. Determination of Toxicologically Relevant Endpoints
The NAS committee commented on the low dose model predictions and the need to
discuss the biological significance of the noncancer health effects modeled in the 2003
Reassessment. In selecting POD candidates from the animal bioassays for derivation of the
candidate RfDs, EPA had to consider the toxicological relevance of the identified endpoint(s)
from any given study. Some endpoints/effects may be sensitive, but lack general toxicological
significance due to not being clearly adverse (defined in EPA's Integrated Risk Information
System glossary as "a biochemical change, functional impairment, or pathologic lesion that
affects the performance of the whole organism, or reduces an organism's ability to respond to an
additional environmental challenge" (U.S. EPA, 2009)), being an adaptive response or not being
clearly linked to downstream functional or pathological alterations. It is standard EPA RfD
derivation practice not to base a reference value on endpoints that are not adverse or not
precursors to an adverse effect. Studies meeting the study selection criteria with endpoints that
were not considered for derivation of a candidate RfD (because they were not considered to be
toxicologically relevant noncancer effects) are: Kitchin and Woods (1979), Hassoun et al. (1998,
2000, 2002, 2003), Burleson et al. (1996), Kuchiiwa et al. (2002), Mally and Chipman (2002),
Vanden Heuvel et al. (1994), Devito et al. (1994), Lucier et al. (1986), Sugita-Konishi et al.
(2003), and Sewall et al. (1993). Appendix G identifies the endpoints from these studies that
were not considered to be toxicologically relevant (e.g., cytochrome P450 induction, oxidative
stress measures, gap junction disruption, mRNA induction, brain serotonin levels) and provides
the rationales for the toxicological relevance decisions on the endpoints. Note that for many of
these studies, other endpoints were examined that are toxicologically relevant and were
considered in the RfD derivation process.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
4.2.2. Use of Toxicokinetic Modeling for TCDD Dose-Response Assessment
Given that TCDD accumulates in fat with continuous exposure and is eliminated slowly
from the body, but at very different rates across species, EPA has determined that the standard
UF approach or allometric scaling of body weight for interspecies extrapolation is not
appropriate. Therefore, EPA has decided to use toxicokinetic modeling to estimate an effective
internal dose for equivalence across species. The toxicokinetic models chosen by EPA are the
rodent and human PBPK models described by Emond et al. (2004, 2006) and modified by EPA
for this assessment as described in Section 3.3.4 (hereafter referred to as the "Emond [rodent or
human] PBPK model"). Both the rodent and human models have a gestational component,
which allow for more relevant exposure comparisons between general adult exposures and the
numerous gestational exposure studies. Ideally, a relevant tissue concentration for each effect
would be estimated. However, no models exist for estimation of all relevant tissue
concentrations. Therefore, EPA has decided to use the concentration of TCDD in blood as a
surrogate for tissue concentrations, assuming that tissue concentrations are proportional to blood
concentrations.16 Furthermore, because the RfD is necessarily expressed in terms of average
daily exposure, the blood concentrations are expressed as averages over the relevant period of
exposure for each endpoint. For the animal bioassay studies, the relevant period of exposure is
the duration of dosing, starting at the age of the animals at the beginning of the study. For
humans, the relevant period of exposure is generally lifetime, which is defined as 70 years by
convention. However, EPA varied the averaging time for the equivalent human blood
concentrations to correspond to the test-animal exposure duration in the following manner.
• For correspondence with animal chronic exposures,17 the human-equivalent
TCDD blood concentration is assumed to be the 70-year average.
• For correspondence with animal gestational exposures, the human-equivalent
TCDD blood concentration is assumed to be the average over 45 years for a
female, beginning at birth, plus 9 months of gestational exposure. The choice of
45 years to beginning of pregnancy is health protective of the population in that
16As virtually all TCDD is found in the adipose fraction of tissues, or bound to specific proteins, an ideal better
approach would be to account for the fat fraction of each tissue and protein binding; EPA has not found sufficient
data to implement this approach.
"Assumed to be >75% of nominal lifetime, or about 550 days in rodents.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
the daily exposure achieving the target blood concentration is smaller than for
shorter averaging times.18
• For correspondence with any other animal exposure duration, the human-
equivalent TCDD blood concentration is assumed to be the average over the
equivalent human exposure duration calculated backward from the peak exposure
plateau at or near the end of the 70-year scenario. The average is determined
from the terminal end of the human exposure period because the daily exposure
achieving the target blood concentration is smaller than for the same exposure
period beginning at birth and is health protective for effects occurring after
shorter-term exposure. The determination of equivalent exposure durations across
species is problematic and somewhat arbitrary, so EPA uses the average peak
blood concentration as the human equivalent for all less-than-chronic animal
exposures (other than gestational).19 For the first-order kinetics model, the
average peak exposure is close to the theoretical steady-state asymptote (see
Section 3.3.4.2). However, for the Emond human PBPK model used by EPA in
this assessment, the timing of the peak exposure is dose-dependent and tends to
decline after 60 years in some cases. Therefore, the 5-year average TCDD blood
concentration that includes the peak ("5-year peak") is used as the relevant dose-
metric for the PBPK model applications.
4.2.3. Noncancer Dose-Response Assessment of Epidemiological Data
The following four epidemiologic studies describing noncancer endpoints were identified
in Section 2.4.3 as studies to be evaluated for development of PODs for derivation of candidate
RfDs: Baccarelli et al. (2008), Mocarelli et al. (2008), Alaluusua et al. (2004) and Eskenazi et al.
(2002). Each of these studies described effects observed in the Seveso cohort (see study
summaries in Section 2.4.1 and Table 2-5). Each study provided individual-level human
exposure measures and an exposure window over which kinetic models could be used to quantify
TCDD exposures for dose-response assessment. EPA used kinetic information to estimate
group-mean daily TCDD intake rates for the exposure groups presented in these studies (see
Appendix D for details). EPA focused on identifying NOAELs and LOAELs for these studies;
EPA did not conduct Benchmark Dose modeling because the covariates identified by the study
authors could not be incorporated by modeling the grouped response data. EPA's development
of PODs for these studies is described in this section and shown in Table 4-1.
18See Section 3.3.4.2 for a discussion of this issue.
19By comparison to a half-lifetime equivalent (1 year in rodents, 35 years in humans), the ratio of body burden
(lst-order kinetic model) to oral intake does not differ significantly from the average-peak scenario; all shorter-term
scenarios differ even less (see Section 3.3.4.2). These relationships, with respect to the 5-year peak, hold for the
PBPK model results, as well (see Section 3).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
4.2.3.1. Baccaretti et al. (2008)
For Baccarelli et al. (2008), EPA was able to define a LOAEL as the group mean of
39 ppt TCDD in neonatal plasma for thyroid stimulating hormone (TSH) values above 5 |iU/mL.
Baccarelli et al. (2008) did not estimate the equivalent oral intake associated with TCDD serum
concentrations and gave only neonatal serum TCDD concentrations for the groups above and
below the 5 |iU/mL standard. The study authors, however, developed a regression model
relating the level of TSH in 3-day-old neonates to TCDD concentrations in maternal plasma at
birth (given as lipid-adjusted serum concentrations, LASC). The authors extrapolated maternal
plasma concentrations from previous measurements using simple first-order pharmacokinetic
model. Because there is limited information regarding the relationship between maternal and
neonatal serum TCDD levels, EPA determined that there was too much uncertainty in estimating
maternal intake from neonatal TCDD serum concentrations, directly. Therefore, EPA
determined the maternal intake at the LOAEL from the maternal serum-TCDD/TSH regression
model by finding the maternal TCDD LASC at which neonatal TSH exceeded 5 |iU/mL. EPA
then used the Emond PBPK model under the human gestational scenario (see Section 4.2.1) to
estimate the continuous daily TCDD intake that would result in a TCDD LASC corresponding to
a neonatal TSH of 5 |iU/mL at the end of gestation, with the resulting maternal intake established
as the LOAEL (0.024 ng/kg-day), shown in Table 4-1 as a candidate POD for derivation of
candidate RfDs. The results of the PBPK modeling are shown in Appendix D.
4.2.3.2. Mocarelli et al. (2008)
Mocarelli et al. (2008) reported decreased sperm concentrations (20%), decreased motile
sperm counts (11%), and decreased serum estradiol (23%) in men who were 1-9 years old in
1976 at the time of the accident (initial TCDD exposure event). Men who were 10-17 years old
in 1976 were not affected, with the possible exception of reduced serum estradiol. Serum
(LASC) TCDD levels were measured within one year of the initial exposure. Serum TCDD
levels and corresponding responses were reported by quartile, with a reference group of less-
exposed individuals assigned a TCDD LASC value of 15 ppt (which was the mean of individuals
outside the contaminated area). The lowest exposed group mean was 68 ppt (1st quartile).
Because effects were detected only among boys under the age of 10, EPA assumes there is a
maximum 10-year critical exposure window for elicitation of these effects. However, for the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
exposure profile, with a high initial pulse followed by an extended period of elimination with
only background exposure, the estimation of an average exposure resulting in the effect is
problematic. Therefore, EPA implemented a procedure for the estimation of the continuous
daily TCDD intake associated with the LOAEL in the Mocarelli et al. (2008) study using a
5-step process.
1. Using the Emond human PBPK model, the initial (peak) blood TCDD concentrations
associated with the accident were back-calculated based on the time that had elapsed
between the explosion and the serum collection. As serum measurements were taken
within 1 year after the event, a lag time of 0.5 years was assumed.
2. The oral exposure associated with the peak blood TCDD concentration (peak exposure)
was calculated using the Emond PBPK model.
3. Starting with the peak exposure and accounting for background TCDD intake, the
average daily blood TCDD concentration experienced by a representative individual in
the susceptible population (boys under 10 years old) was estimated using the Emond
PBPK model. Assuming a random distribution of ages at the time of the event, the
average age of the exposed male children would be 5 years. Consequently, a critical
exposure window for the cohort was estimated to be, on average, 5 years (i.e., a boy aged
5 years would remain in this exposure window for 5 more years until he was 10 years of
age).
4. Using the Emond PBPK model, the average daily TCDD intake rate needed to attain the
5-year average blood TCDD concentration in a boy 10 years old was calculated.
5. The LOAEL POD was calculated as the average of the peak exposure (0.032 ng/kg-day)
and the 5-year average exposure (0.0080 ng/kg-day), resulting in LOAEL of
0.020 ng/kg-day, shown in Table 4-1 as a candidate POD for derivation of a candidate
RfD. However, neither of the extremes was used because (1) the peak exposure does not
account for the continuing internal exposure from TCDD given its slow elimination, and
(2) the 5-year average does not reflect the influence of the much higher peak exposure,
which may be a significant factor in TCDD toxicity (Kim et al., 2003).
The results of the modeling are shown in Appendix D.
4.2.3.3. Alaluusua et al (2004)
For Alaluusua et al. (2004), the approach for estimation of daily TCDD intake is virtually
identical to the approach used for the Mocarelli et al. (2008) data. Alaluusua et al. (2004)
reported dental effects in male and female adults who were less than 5 years of age at the time of
the initial exposure (1976). For the 75 boys and girls who were less than 5 years old at the time
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
of the accident, 25 (33%) were subsequently diagnosed with some form of dental enamel defect.
For the 38 individuals who were older than 5, only 2 (5.3%) suffered dental enamel defects at a
later date. A window of susceptibility of approximately 5 years is established. Serum
measurements for this cohort were taken within a year of the accident. Serum TCDD levels and
corresponding responses were reported by tertile, with a reference group of less-exposed
individuals assigned a TCDD LASC value of 15 ppt (ng/kg); the tertile group means were 130,
383, and 1830 ppt. The incidence of dental effects for the reference group was 26% (10/39).
The incidence of dental effects in the 1st, 2nd and 3rd tertile exposure groups was 10% (1/10),
45%) (5/11) and 60% (9/15), respectively. EPA judged that the NOAEL and LOAEL were 130
and 383 ppt TCDD in serum. Following the same procedure used for the Mocarelli et al. (2008)
study (see Section 4.2.3.2), EPA estimated the continuous daily human TCDD intake associated
with each of the tertiles for both peak and average exposure across the critical exposure window,
assuming that the average age of the susceptible cohort at the time of the accident was 2.5 years.
Separate estimates for boys and girls were developed based on both the peak intake and average
intake across the critical exposure window (see Appendix D for details of the PBPK modeling
are shown in Appendix D). The estimated averaged daily intakes for the tertiles, averaged for
boys and girls, are 0.20, 1.7 and 30 ng/kg-day for the peak exposure and 0.033, 0.15 and
1.5 ng/kg-day for the critical exposure window average. A study NOAEL at the second tertile of
0.12 ng/kg-day was identified as a candidate POD for derivation of a candidate RfD in Table 4-1.
4.2.3.4. Eskenazi et al. (2002)
The approach used to estimate daily TCDD intake in Eskenazi et al. (2002) combines the
approaches EPA used for Baccarelli et al (2008), Mocarelli et al. (2008) and Alaluusua et al.
(2004). Eskenazi et al. (2002) reported menstrual effects in female adults who were
premenarcheal at the time of the initial exposure (1976). In Rigon et al. (2009), the median age
at menarche was shown to be 12.4 in Italian females with intergenerational decreases in age at
menarche. Thus, EPA established a window of susceptibility of approximately 13 years for this
analysis. The average age of the premenarcheal girls at the time of the initial exposure in 1976
was 6.8 years, establishing an average critical-window exposure duration of 6.2 years for this
cohort. Serum samples were collected within a year of the accident from this cohort. However,
serum TCDD levels and corresponding responses were not reported by percentile and no internal
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-11 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
reference group was identified. As for Baccarelli et al (2008), Eskenazi et al. (2002) developed a
regression model relating menstrual cycle length to plasma TCDD concentrations (LASC)
measured in 1976. The model estimated that menstrual cycle length was increased 0.93 days for
each 10-fold increase in TCDD LASC, with a 95% confidence interval of-0.01 to 1.86 days.
EPA judged a 1-day increase in menstrual cycle length to be adverse; a normal menstrual cycle
length is 28 days. EPA then determined the 1976 TCDD serum level corresponding to a 29-day
menstrual cycle length in the exposed cohort from the regression model developed by Eskenazi
et al. (2002). Using this serum level, the peak initial exposure and average exposure over the
6.2 year window were calculated using the Emond human PBPK model, in the same manner as
for Mocarelli et al. (2008) and Alaluusua et al. (2004). The resulting peak TCDD intake is
3.2 ng/kg-day. The average exposure experienced by this cohort over the critical exposure
window is estimated to be 0.12 ng/kg-day. The average of these two estimates is 1.64 ng/kg-day,
which is designated as a LOAEL and shown in Table 4-1. Because the LOAEL is almost
2 orders of magnitude higher than the LOAELs for Baccarelli et al (2008) and Mocarelli et al.
(2008), it was not considered further as a candidate POD for derivation of the RfD. The results
of the PBPK modeling are shown in Appendix D.
4.2.4. Noncancer Dose-Response Assessment of Animal Bioassay Data
EPA followed the strategy illustrated in Figure 4-2 to evaluate the animal bioassay data
for TCDD dose-response. For the administered average daily doses (ng/kg-day) in each animal
bioassay, EPA identified NOAELs and/or LOAELs based on the original data presented by the
study author. Section 2.4.2 identifies these values in the study summaries and in Table 2-7.
These became candidate PODs for consideration in the derivation of an RfD for TCDD. The
candidate RfD values associated with these candidate PODs are presented in Table 4-5.
Additional PODs were identified using BMD modeling. All PODs were converted to HEDs
using the Emond PBPK models. The remainder of this Section describes the steps in this process
and concludes with the POD candidates from the animal bioassay data that were considered for
derivation of the RfD.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
4.2.4.1. Use of Kinetic Modeling for Animal Bioassay Data
Blood concentrations corresponding to the administered doses in each mouse or rat
bioassay qualifying as a final RfD POD candidate were estimated using the appropriate Emond
rodent PBPK model. In each case, the simulation was performed using the exposure and
observation durations, body weights, and average daily doses from the original studies. For all
multiple exposure protocols, the time-weighted average blood TCDD concentrations over the
exposure period were used as the relevant dose metric. For single (gestational and
nongestational) exposures, the initial peak blood TCDD concentrations were considered to be the
most relevant exposure metric. Gestational exposures were modeled using the species-specific
gestational component of the Emond rodent PBPK model. Bioassays employing exposure
protocols spanning gestational and postpartum life stages were modeled by sequential
application of the gestational and nongestational models.
The Emond PBPK models do not contain a lactation component, so exposure during
lactation was not modeled explicitly. Only one bioassay (Shi et al., 2007) considered as a POD
candidate for RfD derivation included exposure during lactation. In Shi et al. (2007) pregnant
animals were exposed weekly to TCDD throughout gestation and lactation. Exposure was
continued in the offspring following weaning for 10 months. For assessment of maternal effects,
the Emond gestational model was used, terminating at parturition. For assessment of long-term
exposure in the offspring, the Emond nongestational model was used, ignoring prior gestational
and lactational exposure, with the assumption that the total exposure during these periods was
small relative to exposure in the following 10 months. The assumption is conservative in that
effects observed in the offspring would be attributed entirely to adult exposures, which is
somewhat less than the actual total exposure.
The model code, input files and PBPK modeling results for each bioassay are reported in
Appendix C. Note that the modeled output is given in terms of LASC. The corresponding
TCDD whole blood concentrations, which are used directly as the toxiocologically-equivalent
dose metric, are derived by multiplying the LASC by the lipid fraction in rodent serum (0.0033)
and the fraction of blood that is serum (0.55). For the rat gestational model, only, the lipid
fraction in serum is 0.0023, rather than 0.0033.
These predicted TCDD blood concentrations were used for benchmark dose modeling of
bioassay response data and determination of NOAELs and LOAELs. BMD modeling was
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
performed, as described in Section 3.5.2.2.1, by substituting the modeled blood concentrations
for the administered doses and calculating the corresponding BMDL. For each of these LOAEL,
NOAEL, or BMDL blood-concentration equivalents, corresponding HEDs were calculated using
the Emond human PBPK model for the appropriate gestational or nongestational scenario as
described previously (see Section 4.2.1).
4.2.4.2. Benchmark Dose Modeling of the Animal Bioassay Data
Benchmark dose modeling was performed using BMDS 2.1, Build 06/16/09 to estimate
BMDs and BMDLs for each study/endpoint combination. The input data tables for these
noncancer studies are shown in Appendix E, Section El, including both administered doses
(ng/kg-day) and blood concentrations (ng/kg) and either incidence data for the dichotomous
endpoints or mean and standard deviations for the continuous endpoints. (See Section 4.3.1
[below] and Sections 3.3.4 and 3.3.5 for a description of the development of TCDD blood
concentrations using kinetic modeling.) For the continuous endpoints, all available models were
run separately using both the assumption of constant variance and the assumption of modeled
variance. Saturated (0 degrees of freedom) model fits were rejected from consideration.
Parameters in models with power or slope parameters were constrained to prevent supralinear
fits, which EPA considers not to be biologically plausible and which often have undesirable
statistical properties (i.e., the BMDL diverges towards zero). However, if the constrained
parameters were estimated at their lower bounds, the unrestricted model was fit to the data,
primarily for elucidation of the degree of supralinearity present in the data. Depending on the
latter and the magnitude of the BMDL relative to the BMD, unrestricted model fits were
occasionally deemed acceptable. Table 4-2 shows each model and any restrictions imposed. For
the quantal/dichotomous endpoints, all primary BMDS dichotomous models were run. The
alternative dichotomous models were fit to several data sets, but the results were very sensitive to
the assumed independent background response and the fits were not accepted. The confidence
level was set to 95% and all initial parameter values were set to their defaults in BMDS. For the
continuous endpoints, one standard deviation was chosen as the default for the BMR when a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
specific toxicologically-relevant BMR could not be defined. For the dichotomous endpoints, a
BMR of 10% extra risk was used for all endpoints.20
The model output tables in Appendix E show all of the models that were run, both
restricted and nonrestricted, goodness-of-fit statistics, BMD and BMDL estimates, and whether
bounds were hit for constrained parameters. After all models were run, the one giving the best
fit was selected using the selection criteria in the current BMDS guidance (U.S. EPA, 2000a)
where possible. Acceptable model fits were those with chi-square goodness-of-fit/> values
greater than 0.1. For the continuous endpoints, thep-walue for the homogenous variance test
(Test 2) was used to determine whether the constant variance (p > 0.1) model or modeled
(nonconstant) variance (p< 0.1) model should be used. As BMDS offers only one variance
model, model fits for nonconstant variance models were not necessarily rejected if the variance
model did not fit well (Test 3 p-w alue < 0.05). Within the group of models with acceptable fits,
the nominal best fitting model was selected by determining the model with the lowest AIC value
if the BMDLs were within a factor of 3; otherwise, the model with the lowest BMDL was chosen
(U.S. EPA, 2000). However, particularly for continuous models, the fit of the model to the
control mean and standard deviation and in the lower response range was assessed. Models with
higher AICs but much better fit to the lower response data were often chosen over the nominally
best-fitting model. Judgment was also used to assess the plausibility of BMDLs far below the
BMD. In most of these cases, deficiencies in the response data resulted in rejection of the
modeling results.
For many data sets, no models satisfied the acceptance criteria and no clear BMD/BMDL
selection could be made. In theses cases, model fits were examined on an individual basis to
determine the reasons for the poor fits. On occasion, high doses were dropped and the models
were refit. If the fit was still not acceptable, the NOAEL/LOAEL approach was applied to the
study/data set combination.
4.2.4.3. POD Candidates from Animal Bioassays Based on HED
Table 4-3 summaries the PODs that EPA estimated for each key animal study included
for TCDD dose-response modeling. After estimating the blood TCDD concentration associated
20 There were no developmental studies that accounted for litter effects, for which a 5% BMR would be used.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
with a particular toxicity measure (NOAEL, LOAEL, or BMDL) obtained from a rodent
bioassay, EPA estimated a corresponding HED using the Emond human PBPK model (described
in Section 3). Table 4-3 summarizes the NOAEL, LOAEL, or BMDL (ng/kg) based on the
administered animal doses for each key bioassay/data set combination. Table 4-3 also
summarizes the continuous daily HED corresponding to these administered doses. The doses in
Table 4-3 are defined as follows:
• Administered Dose NOAEL: Average daily dose (ng/kg-day) from the study
• Administered Dose LOAEL: Average daily dose (ng/kd-day) from the study
• Administered Dose BMDL: Dose from BMD modeling of the administered doses
converted to a blood concentration (ng/kg) using the Emond rodent PBPK models
• Emond Model NOAEL: Administered Dose NOAEL converted to an HED (ng/kg-day)
using the Emond human PBPK model
• Emond Model LOAEL: Administered Dose LOAEL converted to an HED (ng/kg-day)
using the Emond human PBPK model
• Emond Model BMDL: Dose from BMD modeling of the blood concentrations (ng/kg)
converted to an HED (ng/kg-day) using the Emond human PBPK model
Tables showing the best model fit for each study/endpoint combination and the associated
BMD/BMDL are shown in Appendix E.
An evaluation of key BMD analyses is presented in Table 4-4. The BMD modeling was
largely unsuccessful, primarily because of a lack of response data near the BMR, poor modeled
representation of control values, or nonmonotonic responses yielding poor fits. BMDLs were
often implausibly low (orders of magnitude below the LOAEL) or implausibly high. The
comments column in Table 4-4 lists reasons for poor or implausible results.
4.3. RfD DERIVATION
Table 4-5 lists all the studies and endpoints considered for derivation of the RfD. These
studies were chosen from the entire list of candidate study/data set combinations (see
Section 4.2) based on the toxicologic relevance of the endpoints and covering a range that
includes three of the four human studies21. Figure 4-3 (exposure-response array) shows all of the
21 The RfD derived from the study of Eskenazi et al. (2002) was outside the RfD range presented in Table 4-5.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
endpoints listed in Table 4-5 graphically in terms of PODs in human-equivalent intake units
(ng/kg-day). The human study endpoints are shown at the far left of the figure and the rodent
endpoints are arranged by category to the right. Figure 4-4 demonstrates the same endpoints
arrayed by RfD value, showing the POD, applicable UFs and candidate RfD.
Table 4-5 illustrates the study, species, strain and sex, study protocol, and toxicologic
endpoints observed at the lowest TCDD doses. The table also identifies the human-equivalent
BMDLs (when applicable), NOAELs and LOAELs, as well as the composite UF that applies to
the specific endpoint, and finally, the corresponding candidate RfD.22 The NOAELS, LOAELs,
and BMDLs are presented as HEDs, based on the assumption that blood concentration is the
toxicokinetically-equivalent TCDD dose metric across species and serves as a surrogate for
tissue concentration.23 For rats and mice, these estimates relied on the two Emond PBPK
models—one for the relevant rodent species and one for the human—as described previously
(see Sections 3.3.4.3). The two guinea pig studies that are included in Table 4-5 are given in
HED units based on the first-order body burden model described in Section 3.3.4.2; there is
currently no TCDD PBPK model for the guinea pig. The values listed for guinea pigs are not
directly comparable to those for rats and mice but are probably biased low, as first-order body
burden HED estimates for rats and mice are generally 2- to 5-fold lower than the corresponding
PBPK model estimates. The LOAELs for the human studies also rely on the Emond PBPK
model, as described in Sections 4.2.2 and 4.2.3.
As is evident from the table, very few NOAELs and even fewer BMDLs have been
established for low-dose TCDD studies. BMD modeling was unsuccessful for all of the
endpoints without a NOAEL, primarily because of the lack of dose-response data near the BMR
(see discussion in Section 4.2). Therefore, the RfD assessment rests largely on evaluation of
LOAELs to determine the POD.
The rows in Table 4-5 are arranged in order of increasing candidate RfD magnitude.
Endpoints projected to occur at higher exposure levels are still considered for qualitative support
of the effects shown in Table 4-5.
22Extra significant digits are retained for comparison prior to rounding to one significant digit for the final RfD.
23The procedures for estimating HEDs based on TCDD blood concentration are described in the preceding section.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
4.3.1. Toxicological Endpoints
As can be seen in Table 4-5, a wide array of toxicological endpoints has been observed
following TCDD exposure, ranging from subtle developmental effects to overt chronic liver
toxicity. Developmental effects in rodents include dental defects, delayed puberty in males, and
several neurobehavioral effects. Reproductive effects reported in rodents include altered
hormone levels in females and decreased sperm production in males. Immunotoxicity endpoints
such as decreased response to SRBC challenge in mice and decreased delayed-type
hypersensitivity response in guinea pigs are also observed. Longer durations of TCDD exposure
in rodents elicit results such as organ and body weight changes, renal toxicity, and liver and lung
lesions. Adverse effects in human studies are also observed, which include male reproductive
effects, increased TSH in neonates, and dental defects in children. Analogous results have been
observed in animal bioassays for each of these human endpoints.
All but two of the study/endpoint combinations from animal bioassays listed in Table 4-5
are on TCDD-induced toxicity observed in mice and rats; the other two study/endpoint
combinations are effects in guinea pigs. Although the effects of TCDD have been investigated in
several other species (i.e., hamsters, monkeys, and mink), those studies were not included for
final POD consideration because the effect levels were greater than those in Table 4-5, or
because the effects could not be attributed solely to TCDD exposure (i.e., confounding by
dioxin-like compounds [DLCs]).
Three human studies were also included for final POD consideration in the derivation of
an RfD and are presented in Table 4-5 as candidate RfDs. All three human study/endpoint
combinations are from studies on the Seveso cohort. The developmental effects observed in
these studies were associated with TCDD exposures either in utero or in early childhood between
1 and 10 years of age. Baccarelli et al. (2008) reported increased levels of TSH in newborns
exposed to TCDD in utero, indicating a possible dysregulation of thyroid hormone metabolism.
Mocarelli et al. (2008) reported decreased sperm concentrations, decreased motile sperm counts,
and decreased serum estradiol in men who were 1-9 years old in 1976 at the time of the Seveso
accident (initial TCDD exposure event). Alaluusua et al. (2004) reported dental effects in adults
who were less than 9.5 years of age at the time of the initial exposure (1976).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
4.3.2. Exposure Protocols of Candidate PODs
The studies in Table 4-5 represent a wide variety of exposure protocols, involving
different methods of administration and exposure patterns across virtually all exposure durations
and life stages. Both dietary and gavage administration have been used in rodent studies, with
gavage being the predominant method. Gavage dosing protocols vary quite widely and include
single gestational exposures, multiple daily exposures (for up to 2 weeks, intermittent schedules
that include 5 days/week, once weekly, or once every 2 weeks), and loading/maintenance dose
protocols, in which a relatively high dose is initially given followed by much smaller weekly
doses. The intermittent dosing schedules require dose-averaging over time periods as long as
2 weeks, which introduces uncertainty in the effective exposures. In other words, the high unit
dose may be more of a factor in eliciting the effect than the average TCDD tissue levels over
time. Although the loading/maintenance dose protocols are designed to maintain a constant
internal exposure, these protocols are somewhat inconsistent with the constant daily TCDD
dietary exposures associated with human ingestion patterns.
The epidemiologic studies conducted in the Seveso cohort represent exposures over
different life stages including gestation, childhood, and young adulthood. The Seveso exposure
profile is essentially a high initial pulse TCDD exposure followed by a 20-30 year period of
elimination. Effects are realized, or measured, 10-20 years following the initial exposure; the
critical exposure window for susceptibility varies with effect and is often unknown. Therefore,
the effective exposure profiles for the Seveso cohort studies vary considerably. For the
Mocarelli et al. (2008) and Alaluusua et al. (2004) studies where early childhood exposures
proximate to the initial event are associated with the outcomes, there is some uncertainty as to
the magnitude of the effective doses. Although the effects are associated with TCDD exposure
in the first 10 years of life, it is not clear to what extent the initial peak exposure is primarily
responsible for the effects. It is also not clear if averaging exposure over the critical window is
appropriate given the large difference between initial TCDD body burden and body burden at the
end of the critical exposure window. The LOAELs for both Mocarelli et al. (2008) and
Alaluusua et al. (2004) are calculated as the average of the peak exposure and average exposure
across the critical exposure window (see Section 4.2 for details). For the gestational exposure
study (Baccarelli et al., 2008), the critical exposure window is strictly defined and relatively
short (9 months) and occurs long after the initial exposure (15-20 years). In addition, the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
maternal serum TCDD measurements were taken 10-15 years after the initial exposure and are
proximate to the actual pregnancies, allowing for less uncertainty in the kinetic extrapolation to
time of birth. The narrow critical exposure window at a much later time than the initial exposure
(where the TCDD elimination curve is flattening) is assumed to lead to a relatively steady-state
exposure over the critical time period with much less uncertainty in the magnitude of the
effective dose. With the exception of Eskenazi et al (2002) (see section 4.2), the effective doses
for other effects reported for the Seveso cohort (see Section 2.4.1.1.1.4) have not been quantified
and are not represented in Table 4-5 because no critical exposure windows can be identified or
individual exposure estimates were not reported.
4.3.3. Uncertainty Factors (UFs)
The UF column in Table 4-5 shows the composite (total) UF that would be applied to the
POD for each endpoint. For the animal bioassays, a UF of 3 for the toxicodynamic component
of the interspecies extrapolation factor (UFA) was applied to all PODs. For both animal and
human studies, when a NOAEL was used as the POD, a factor of 10 was applied for human
interindividual variability (UFH). For all of the animal bioassay endpoints lacking a NOAEL, a
UF of 10 for the LOAEL-to-NOAEL UF (UFL) was included. For the human LOAELs, a UFL of
3 was applied because sensitive populations were identified. A sub chronic-to-chronic UF (UFS)
of 1 and a database factor (UFD) of 1 are applied to all endpoints. A rationale for each UF is
provided for the derivation of the RfD below.
4.3.4. Human Studies
For selection of the POD, the human studies are given the highest consideration, as
quality human data are always preferred. Although the lower end of the candidate RfD
distribution is dominated by mouse studies (comprising 6 of the 7 lowest rodent-based RfDs),
EPA considers these candidate RfD estimates to be much more uncertain than human candidate
RfD estimates. The LOAELreds identified in mouse bioassays are low primarily because of the
large toxicokinetic interspecies extrapolation factor used for mice. The ratio of administered
dose to HED (Da:HED) ranges from 65 to 1227 depending on the duration of exposure. The
Da:HED for mice is, on average, about 4 times larger than that used for rats. In addition, each
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-20 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
one of the mouse studies has other qualitative limitations and uncertainties (discussed above and
in Table 4-4) that make them poor candidates as the basis for the RfD.
Most of the other rodent studies between the first six mouse studies and the human
studies of Mocarelli et al (2008) and Baccarelli et al. (2008) are of small size, using 10 or fewer
animals per dose group and are considered too uncertain on which to base the final RfD. Two of
the rat bioassays-Bell et al (2007) and NTP (2006)—however, are of particular note. Both of
these studies were very well designed and conducted, using 30 or more animals per dose group
(see Table 4-6). Bell et al (2007) evaluated several reproductive and developmental endpoints
starting exposure well before mating and continuing through gestation. NTP (2006) is the most
comprehensive evaluation of TCDD chronic toxicity in rodents to date, evaluating dozens of
endpoints at several time points in all major tissues. The toxicokinetic extrapolation to humans,
however, is still very uncertain. Despite the overall strength of the Bell and NTP studies, EPA
still considers the human data to be a better basis for the RfD.
Baccarelli et al. (2008) reported increased levels of TSH in newborns exposed to TCDD
in utero, indicating a possible dysregulation of thyroid hormone metabolism. The study authors
related TCDD concentrations in neonatal blood to TSH levels, reporting group mean TCDD
concentrations associated with TSH levels above or below 5 |i-Units TSH per mL of serum
(5 |iU/mL), The 5 |iU/mL standard was established by the World Health Organization (WHO,
1994) as an indicator of potential iodine deficiency (and potential thyroid problems) in neonates.
Baccarelli et al. (2008) also showed, in graphical form, how the TSH distribution in each of three
categorical exposure groups (reference, zone A, and zone B—representing increasing TCDD
exposure) shifted to higher TSH values with increasing exposure. The individuals comprising
the above 5 |iU/mL group were from all three categorical exposure groups, not just from the
highest exposure group. Therefore, EPA was able to designate a LOAEL independently of the
nominal categorical exposure groups; the LOAEL is designated as the group mean of 39 ppt
TCDD in neonatal plasma as a LOAEL for TSH values above 5 |iU/mL. The daily oral intake at
the LOAEL is estimated to be 0.024 ng/kg-day (see Section 4.2.3.1). A NOAEL is not defined
because it is not clear what maternal intake should be assigned to the group below 5 |iU/mL.
Mocarelli et al. (2008) reported decreased sperm concentrations (20%), decreased motile
sperm counts (11%), and decreased serum estradiol (23%) in men who were 1-9 years old in
1976 at the time of the Seveso accident (initial TCDD exposure event). The sperm
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-21 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
concentrations and motile sperm counts in men who were 10-17 years old in 1976 were not
affected; marginal serum estradiol reductions were reported in this group. Serum (LASC) TCDD
levels were measured within one year of the initial exposure. Serum TCDD levels and
corresponding responses were reported by quartile, with a reference group of less-exposed
individuals assigned a TCDD LASC value of 15 ppt (which was the mean of the TCDD LASC
reported in individuals outside the contaminated area). The lowest exposed group mean was
68 ppt (lst-quartile). Mean sperm concentrations and motile sperm counts were reduced about
20% from the reference group. Further decrease in these values was slight and reached a
maximum of about 33%. Although a decrease in sperm production of 20% would not have
clinical significance for an individual, EPA considers a 20% shift in the population mean to be of
biological significance. Therefore, EPA has designated the lowest exposure group (68 ppt) as a
LOAEL, which translates to a continuous daily oral intake of 0.020 ng/kg-day (see
Section 4.2.3.2). The reference group is not designated as aNOAEL because there is no clear
zero-exposure measurement for any of these endpoints, particularly considering the contribution
of background exposure to DLCs, which futher complicates the interpretation of the reference
group response as a true "control" response (see discussion in Section 4.4). However, males less
than 10 years old can be designated as a sensitive population by comparison to older males who
were not affected.
Alaluusua et al. (2004) reported dental effects in male and female adults were were less
than 9.5 years of age, but not older, at the time of the initial exposure (1976) in Seveso. EPA
used the same approach to estimate daily TCDD intake as was used for the Mocarelli et al.
(2008) data; a window of susceptibility of about 5 years was established. Serum measurements
for this cohort were taken within a year of the accident. Serum TCDD levels and corresponding
responses were reported by tertile, with a reference group of less-exposed individuals assigned a
TCDD LASC value of 15 ppt (ng/kg); the tertile group means were 130, 383, and 1,830 ppt.
Both a NOAEL and LOAEL can be defined for this study. The NOAEL is 0.12 ng/kg-day,
corresponding to the TCDD LASC of 130 ppt at the first tertile. The LOAEL is 0.93 ng/kg-day
at the second tertile. The children in this cohort less than 5 years old can be designated as a
sensitive population by comparison to older individuals who were not affected relative to the
reference group.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
4.3.5. Derivation of the RfD
The two human studies, Baccarelli et al. (2008) and Mocarelli et al. (2008), have similar
LOAELs of 0.024 and 0.020 ng/kg-day, respectively. Together, these two studies constitute the
best foundation for establishing a POD for the RfD, and are designated as co-principal studies.
Therefore, increased TSH in neonates (Baccarelli et al., 2008) and male reproductive effects
(decreased sperm count and motility, increased estradiol) are designated as cocritical effects.
Although the exposure estimate used in determination of the LOAEL for Mocarelli et al. (2008)
is more uncertain than the Baccarelli et al. (2008) exposure estimate, the slightly lower LOAEL
of 0.020 ng/kg-day from Mocarelli et al. is designated as the POD. A composite UF of 30 is
applied to account for lack of a NOAEL (UFL =10) and human interindividual variability
(UFh = 3); the resulting RfD in standard units is 7 x 10~10 mg/kg-day. Table 4-7 presents the
details of the RfD derivation.
4.4. UNCERTAINTY IN THE RfD
Exposure assessment is a key limitation of the epidemiologic studies (of the Seveso
cohort) used to derive the RfD. The Seveso cohort exposure profile consists of an initial high
dose followed by a drop in body burden to background levels over a period of about 20 years, at
which time the effects were observed. This exposure scenario is a mismatch with the constant
daily intake scenario addressed by the RfD methodology. The determination of an effective
average daily dose from the Seveso exposure scenario requires an understanding of the critical
time-window of susceptibility and the influence of the peak exposure on the occurrence of the
observed effects, particularly when the peak exposure is high relative to the average exposure
over the critical exposure window. For one of the principal studies (Mocarelli et al., 2008), a
maximum susceptibility exposure window can be identified based on the age of the population at
risk. However, the influence of the peak exposure on the effects observed 20 years later is
unknown and the biological significance of averaging the exposure over several years, with
internal exposure measures spanning a 4.5-fold range, is unknown. EPA, in this assessment, has
averaged intermittent exposures for rodent bioassays over weekly dosing intervals, but the peak
and average body burdens varied by less than 50%. EPA has not developed guidance for larger-
interval averaging. Furthermore, because there is an assumption of a threshold level of exposure
below which the effects are not expected to occur, averaging over large intervals could include
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
below-threshold exposures. The process used by EPA to estimate the LOAEL exposure for the
Mocarelli study is a compromise between the extremes; as such, there is some uncertainty in the
estimate, perhaps in the range of 3- to 10-fold in either direction. This uncertainty also holds for
the LOAEL determined for the dental effects reported in Alaluusua et al. (2004) and the
increased menstrual cycle length reported in Eskenazi et al. (2002a; see Section 4.2.3.4); in both
of those studies, the uncertainty is greater, as the difference between peak and average internal
exposures is an order of magnitude or more. The LOAEL for increased TSH in neonates
(Baccarelli et al., 2008), however, is less uncertain because the critical exposure window is much
narrower (9 months) and the developmental exposures occurred 10 to 15 years after the initial
exposure, when internal TCDD concentrations for the pregnant women were leveling off; that is,
exposure over the critical window was more constant. However, there is uncertainty in the
magnitude of the exposures because they were extrapolated from serum measurements taken
several years earlier.
Another source of uncertainty using human epidemiologic data is the lack of completely
unexposed populations. The available TCDD epidemiologic data were obtained by comparing
populations that experienced elevated TCDD exposures to populations that experienced lower
exposures, rather than to a population with no TCDD exposure. An additional complicating
factor is coexposure to DLCs, which can behave in the same way as TCDD. Although the
accidental exposure to the Seveso women's cohort was virtually 100% TCDD, background
exposure was largely to DLCs. Eskenazi et al. (2004) reported that TCDD comprised only 20%
of the total toxicity equivalence (TEQ) in the serum of the reference group that was not exposed
as a result of the factory explosion, which implies that the effective background exposure could
have been as much as 5-fold higher. The higher background exposure could be significant at the
lower TCDD exposure levels, with the effect diminishing as TCDD exposure increased. For
dose-response modeling, the effect of a higher background dose (i.e., total TEQ), if included,
would be to shift the response curve to the right (responses associated with higher exposures)
but, primarily, would reduce the spread of the exposures, which would tend to alter the shape of
the dose response towards sublinear. Both the right shift and the more sublinear shape would
result in higher EDX estimates, such as BMDs and BMDLs, from fitting dose-response models.
However, for determination of a LOAEL, which is the case for all the human studies in
Table 4-5, the impact may be minimal, as the LOAEL depends only on establishing that an effect
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
of sufficient magnitude was observed at some TCDD exposure level. In this case, the effect of
the increased effective background exposure would be to inflate the "control" (zero-TEQ)
response, providing the threshold for the response had been exceeded. The potential impact of
an inflated control response would be to mask a significant effect of the added TCDD exposure,
when the latter effect is determined by comparison to the reference group response. To
compensate for this, EPA has been somewhat conservative in interpreting the magnitude of
responses defining LOAELs for the Seveso cohort studies. The actual magnitude of the impact
of the DLC background exposure is impossible to assess without knowing the true (TEQ-free)
background response.
A primary strength of the TCDD database is that analogous effects have been observed in
animal bioassays for most of the human endpoints, increasing the overall confidence in the
relevance to humans of the effects reported in rodents and the association of TCDD exposure
with the effects reported in humans. Table 4-5 shows that low dose TCDD exposures are
associated with a wide array of toxicological endpoints in rodents including developmental
effects, reproductive effects, immunotoxicity and chronic toxicity. Effects reported in human
studies are similar, including male reproductive effects, increased TSH in neonates and dental
defects in children; other human health effects such as female reproductive effects and chloracne
have been observed at higher exposures (see Section 2.4.1). Other effects reported in rodent
studies such as liver toxicity and overt immunological endpoints have not been reported in
human studies. However, with respect to immunological effects, Baccarelli et al (2002, 2004)
evaluated immunoglobin and complement levels in the sera of TCDD-exposed individuals from
the Seveso cohort and found slightly reduced immunoglobulin in the highest exposure groups but
no effect on other immunoglobulins or on C3 or C4 complement levels. The latter finding
indicates that at least one immunological measure in humans is not a sensitive endpoint, as it is
for mice, with large reductions in serum complement at low exposure levels (White et al., 1986).
Although there is a substantial amount of qualitative concordance of effects between
rodents and humans, quantitative concordance is not evident in Table 4-5. The differential
sensitivity of mice and humans for the serum complement endpoint is one example. Other
examples of differential sensitivity are developmental dental effects and thyroid hormonal
dysregulation. Developmental dental defects are relatively sensitive effects in rodents, appearing
at exposure levels in mice (Keller et al., 2007, 2008a,b) more than an order of magnitude lower
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
than effect levels in humans (Alaluusua et al., 2004). In contrast, thyroid hormone effects are
seen in rats (Crofton et al., 2005) at 30-fold higher exposures than for humans (Baccarelli et al.,
2008). Male reproductive effects (sperm production) occur in rats (Latchoumycandane and
Mathur, 2002) and humans (Mocarelli et al., 2008) at about the same dose. To what extent these
differential sensitivities depend on specifics of the comparison, such as species (mouse vs. rat),
life-stage (e.g., fetal vs. adult), endpoint measure (e.g., thyroxine [T4] vs. TSH) or magnitude of
the lowest dose tested, cannot be determined, so strong conclusions about quantitative
concordance cannot be made.
Considering the issue of lowest tested dose, the general lack of NOAELs and acceptable
BMDLs is a primary weakness of the rodent bioassay database. None of the 6 most sensitive
rodent studies in Table 4-5, spanning a 30-fold range of LOAELs, had defined NOAELs or
BMDLs. For only 4 of the next 10 rodent studies were NOAELs or BMDLs established. In
addition, many of these LOAELs are characterized by relatively high responses with respect to
the control population, so it is not certain that a 10-fold lower dose (based on the application of
UFl of 10) would be approximately equivalent to a NOAEL. A major reason for the failure of
BMD modeling was that the responses were not "anchored" at the low end; first response levels
were far from the BMR (see Table 4-4). Another major problem with the data was non-
monotone and flat response profiles. The small dose-group sizes and large dose intervals
probably contributed to many of these response characteristics that prevented successful BMD
modeling. Larger study sizes with narrower dose intervals at lower doses are still needed to
clarify rodent response to TCDD.
Lower TCDD doses have been tested in rodents but almost entirely for investigation of
specialized biochemical endpoints24 that would not be considered to be adverse health effects
(see Appendix G). There is, however, a fundamental limit to the lowest dose of TCDD that can
be tested meaningfully, as TCDD is present in feed stock and accumulates in unexposed animals
prior to the start of a study. This issue is illustrated by the presence of TCDD in tissues of
unexposed control animals, often at significant levels relative to the lowest tested dose in low
dose studies (Vanden Heuvel et al., 1994; Ohsako et al., 2001; Bell et al., 2007a; see
24Enzyme induction, oxidative stress indicators, mRNA levels, etc.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-26 DRAFT—DO NOT CITE OR QUOTE
-------
1 Text Box 4-1). Some DLCs have been measured in animal feeds and are anticipated to
2 accumulate in unexposed test animals further complicating the interpretation of low dose studies.
3
Text Box 4-1.
TCDD tissue levels in control animals are rarely reported either explicitly or implicitly.
VandenHeuvel et al. (1994), however, reported TCDD concentrations in livers of control
animals (10-week-old female Sprague-Dawley rats) of 0.43 ppt (ng/kg) compared to 0.49 ppt in
the livers of animals given a single oral TCDD dose of 0.1 ng/kg. Assuming proportionality of
liver concentration to total body burden, the body burden of untreated animals was 87.8% of
that of treated animals. The equivalent administered dose for untreated animals (do) can be
calculated as equal to 0.878 x (0.1 + d0), assuming proportionality of body burden to
administered dose and that all animals started with the same TCDD body burdens. The
calculation yields a value of 0.72 ng/kg for d0, which represents the accumulated TCDD from
all sources in these animals prior to being put on test. This value would raise the nominal
0.1 ng/kg TCDD dose 8-fold to 0.82 ng/kg. The next higher dose of 1 ng/kg would be nearly
doubled to 1.72 ng/kg. The impact on higher doses would be negligible. Bell et al. (2007a)
reported slightly higher levels (0.66 ppt) in the livers of slightly older untreated pregnant female
Sprague-Dawley rats (mated at 16-18 weeks of age and tested 17 days later).
Ohsako et al. (2001) reported TCDD concentrations in the fat of offspring of untreated
pregnant Holtzman rats that were 46% of the TCDD fat concentrations in animals exposed in
utero to 12.5 ng/kg (single exposure on GD 15). This level of TCDD would imply a very large
background exposure, but quantitation based on simple kinetic assumptions probably would not
reflect the more complicated indirect exposure scenario
Bell et al. (2007a) also reported concentrations of 0.1 and 0.6 ppt TCDD measured in
two samples of feed stock. Assuming that the average of 0.35 ppt is representative of the entire
supply of feed stock and a food consumption factor of 10% of body weight per day, the average
daily oral exposure from feed to these animals would be 0.035 ng/kg. Discrimination of
outcomes from longer-term repeated exposures might be problematic at exposure levels around
0.1 ng/kg-day. Background exposure was not much of an issue for Bell et al. (2007a), as the
lowest TCDD exposure level was 2.4 ng/kg-day (28-day dietary exposure).
NTP (2006), however, found virtually no TCDD in the tissues of untreated animals or
in the feed stock. In all of these studies, except the 28-day exposure in Bell et al. (2007a),
control animals were gavaged with corn oil vehicle. TCDD concentrations in corn oil were not
4 reported in any of the studies.
5
6
7 Table 4-6 compares the qualitative strengths and limitations/uncertainties associated with
8 the top animal bioassays listed in Table 4-5.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-27 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 4-1. POD candidates for epidemiologic studies of TCDD
2
Study
POD (ng/kg-day)
Critical effects
Alaluusua et al., 2004
7.20E-023 (NOAEL)
Dental effects in adults exposed to TCDD in childhood
Baccarelli et al., 2008
2.40E-2b (LOAEL)
Elevated TSH in neonates
Eskenazi et al., 2002
1.66E+0C (LOAEL)
Increased length of menstrual cycle in women exposed
to TCDD in childhood
Mocarelli et al., 2008
2.00E-2d (LOAEL)
Decreased sperm count and motility and increased
estradiol in men exposed to TCDD in childhood
3
4 aMean of peak exposure (0.15 ng/kg-day) and average exposure over 10-year critical window (0.0093 ng/kg-day).
5 bMaternal exposure corresponding to neonatal TSH concentration exceeding 5 |iU/mL.
6 °Mean of peak exposure (3.2 ng/kg-day) and average exposure over 10-year critical window (0.12 ng/kg-day).
7 dMean of peak exposure (0.035 ng/kg-day) and average exposure over 10-year critical window (0.0078 ng/kg-day).
8
9
10 Table 4-2. Models run for each study/endpoint combination in the animal
11 bioassay benchmark dose modeling
12
Model
Restrictions imposed
Continuous models
Exponential M2-M5,
not grouped
Adverse direction specified according to the response data; power > 1
Hill
Adverse direction is automatic; n> 1
Linear
Adverse direction is automatic; degree of polynomial = 1
Polynomial
Adverse direction is automatic; degree of polynomial = 2; restrict the sign of
the power to non-negative or non-positive, depending on the direction of the
responses
Power
Adverse direction is automatic; power >1
Dichotomous models
Gamma
Power >1
Logistic
None
Log-Logistic
Slope >1
Log-Probit
None
Multistage
Beta >0, 2nd degree polynomial
Probit
None
Weibull
Power >1
13
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-28 DRAFT—DO NOT CITE OR QUOTE
-------
0
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
^s
^s
TO
TO'
*
^5
«
0
J\. Co
1 ^
to ^
vo o
s
1
a,
2 §-
C^ TO
>;
a
H o
.. ^k
TO
O
s
2! »
O ^
H r?
<*^
hh Oq
H to
W|
o £?
o
H
W
Table 4-3. Summary of key animal study NOAELs, LOAELs, and BMDLs for different dose metrics (ng/kg-
day)
Study
Endpoint
Administered dose"
lst-order body burden HEDb
Blood concentration HEDC
NOAEL
LOAEL
BMDLd
NOAEL
LOAEL
BMDLd
NOAEL
LOAEL
BMDLd
Amin etal., 2000
Saccharin preference ratio,
female
-
2.50E+01
5.15E+01
-
2.51E+04
5.17E+04
-
1.71E-01
3.19E-01
Bell et al., 2007a
Balano-preputial separation
in male pups
-
2.40E+00
1.25E-01
-
1.41E+02
7.36E+00
-
1.10E-01
7.40E-03
Cantoni et al., 1981
Urinary coproporhyrins
-
1.43E+00
2.82E-01
-
1.65E+02
3.25E+01
-
6.51E-02
1.60E-03
Chuetal., 2001
Tissue weight changes
2.50E+02
1.00E+03
-
8.28E+04
3.31E+05
-
-
-
-
Chu et al., 2007
Liver lesions
2.50E+00
2.50E+01
-
8.28E+02
8.28E+03
-
3.56E-02
5.76E-01
-
Crofton et al., 2005
Serum T4
3.00E+01
1.00E+02
3.01E+01
4.69E+04
1.56E+05
4.70E+04
1.73E-01
7.62E-01
1.41E-01
Croutch et al., 2005
Decreased body weight
5.43E+01
2.17E+02
-
1.33E+04
5.30E+04
-
-
-
-
DeCaprio et al.,
1986
Decreased body weight
6.10E-01
4.90E+00
-
9.05E+01
7.27E+02
-
-
-
-
Fattore et al., 2000
Decreased hepatic retinol
-
2.00E+01
-
-
3.25E+03
-
-
8.01E-01
-
Fox et al., 1993
Increased liver weight
5.73E-01
3.27E+02
-
2.31E+02
1.32E+05
-
-
-
-
Franczak et al.,
2006
Abnormal estrous cycle
-
7.14E+00
-
-
8.57E+02
-
-
0.326716
-
Hojo et al., 2002
DRL response per min
-
2.00E+01
1.10E+02
-
7.60E+04
4.18E+05
-
5.51E-02
2.59E-05
Ikeda et al., 2005a
Sex ratio
-
1.65E+01
-
-
2.60E+03
-
-
2.7500259
-
Ishihara et al., 2007
Sex ratio
1.00E-01
1.00E+02
-
3.15E+01
3.15E+04
-
-
-
-
Kattainen et al.,
2001
3rd molar length
-
3.00E+01
2.14E+00
-
1.14E+05
8.15E+03
-
9.00E-02
1.71E-03
Keller et al., 2007a,
2008a, 2008b
Missing mandibular molars
-
1.00E+01
9.17E+00
-
3.88E+04
3.56E+04
-
9.81E-03
1.70E-02
Kocibaetal., 1976
Liver and hematologic
effects and body weight
changes
7.14E+00
7.14E+01
1.13E+03
1.13E+04
1.60E-01
1.71E+00
O
O
-------
0
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
^s
^s
TO
TO'
*
^5
«
0
J\. Co
1 ^
U> ^
o o
s
1
a,
2 §-
7? ^
J> >3
Tl S
H ©
3 §
2 §
O ^
H r?
<*^
hh Oq
H to
W|
O £?
o
H
W
Table 4-3. Summary of key study NOAELs, LOAELs, and BMDLs for different dose metrics (ng/kg-day)
(continued)
Study
Endpoint
Administered dose"
lst-order body burden HEDb
Blood concentration HEDC
NOAEL
LOAEL
BMDLd
NOAEL
LOAEL
BMDLd
NOAEL
LOAEL
BMDLd
Kocibaetal., 1978
Liver and lung lesions,
increased urinary
porphyrins
1.00E+00
1.00E+01
7.30E-01
9.31E+01
9.31E+02
6.80E+01
6.46E-02
6.46E-01
2.00E-02
Latchoumycandane
and Mathur., 2002
Sperm production
-
1.00E+00
1.56E-02
-
2.59E+02
4.03E+00
-
1.70E-02
3.90E-05
Li et al., 1997
Increased serum FSH
3.00E+00
1.00E+01
-
1.14E+04
3.80E+04
-
2.97E-03
1.72E-02
-
Li et al., 2006
Hormone levels (serum
estradiol)
-
2.00E+00
1.11E+02
-
4.06E+03
2.27E+05
-
1.57E-03
3.46E-01
Markowski et al.,
2001
FR2 revolutions
-
2.00E+01
5.81E+01
-
6.40E+04
1.86E+05
-
5.17E-02
1.34E-01
Maronpot et al.,
1993
Increased relative liver
weight
1.07E+01
3.50E+01
-
1.28E+03
4.18E+03
-
-
-
-
Miettinen et al.,
2006
Cariogenic lesions in pups
-
3.00E+01
4.99E+00
-
1.14E+05
1.90E+04
-
8.94E-02
9.32E-03
Murray et al., 1979
Fertility index in fl
generation
1.00E+00
1.00E+01
-
1.06E+02
1.06E+03
-
2.96E-02
3.88E-01
-
NTP, 1982
Liver lesions
-
1.39E+00
1.38E+01
-
2.98E+02
2.97E+03
-
2.19E-02
1.78E-02
NTP, 2006
Liver and lung lesions
-
2.14E+00
1.40E+00
-
1.96E+02
1.28E+02
-
1.39E-01
8.76E-02
Nohara et al., 2000
Decreased spleen
cellularity
8.00E+02
-
-
3.04E+06
-
-
5.34E+00
-
-
Ohsako et al., 2001
Anogenital distance in
pups
1.25E+01
5.00E+01
1.22E+01
4.75E+04
1.90E+05
4.62E+04
2.87E-02
1.80E-01
2.67E-02
Seo et al., 1995
Decreased thymus weight
2.50E+01
1.00E+02
-
2.51E+04
1.00E+05
-
1.69E-01
9.31E-01
-
Sewall et al., 1995
Serum T4
1.07E+01
3.50E+01
5.20E+00
1.28E+03
4.18E+03
6.20E+02
5.15E-01
1.76E+00
7.25E-02
Shi et al., 2007
Serum estradiol in female
pups
1.43E-01
7.14E-01
2.24E-01
1.67E+01
8.32E+01
2.61E+01
4.71E-03
2.75E-02
4.95E-03
-------
0 ^
&
to
s
S
to
s
>;*
a
a,
Sf
^s
TO
TO'
*
^5
«
0
_L^ >;
1 TO
LtJ ^
^ O
s
1
a,
3 §-
C^ TO
>;
a
H o
.. ^k
TO
O
s
2! »
O ^
H r?
<*^
hh Oq
H to
W|
o £?
o
H
ffl
Table 4-3. Summary of key study NOAELs, LOAELs, and BMDLs for different dose metrics (ng/kg-day)
(continued)
Study
Endpoint
Administered dose"
lst-order body burden HEDb
Blood concentration HEDC
NOAEL
LOAEL
BMDLd
NOAEL
LOAEL
BMDLd
NOAEL
LOAEL
BMDLd
Simanainen et al.,
2002
Decreased serum T4
1.00E+02
3.00E+02
-
3.80E+05
1.14E+06
-
-
-
-
Simanainen et al.,
2003
Decreased thymus weight
and change in EROD
activity
1.00E+02
3.00E+02
~
3.80E+05
1.14E+06
~
~
~
~
Simanainen et al.,
2004
Decreased daily sperm
production
1.00E+02
3.00E+02
-
3.80E+05
1.14E+06
-
-
-
-
Smialowicz et al.,
2004
Decreased antibody
response to SRBCs
3.00E+02
1.00E+03
-
1.16E+06
3.88E+06
-
-
-
-
Smialowicz et al.,
2008
PFC per 10A6 cells
-
1.07E+00
7.19E+01
-
2.29E+02
1.54E+04
-
6.38E-03
2.00E-03
Toth et al., 1979
Skin lesions
-
1.00E+00
6.85E-02
-
2.70E+02
1.85E+01
-
1.00E-02
8.61E-01
VanBirgelen et al.,
1995a,b
Decreased liver retinyl
palmitate
-
1.40E+01
9.89E+02
-
2.27E+03
1.61E+05
-
5.25E-01
5.00E+00
Vosetal., 1973
Decreased delayed-type
hypersensitivity response
to tuberculin
1.14E+00
5.71E+00
2.02E+02
1.01E+03
White et al., 1986
Decreased serum
complement
-
1.00E+01
2.89E+01
-
4.49E+03
1.30E+04
-
2.83E-02
4.65E-02
Yang et al., 2000
Increased endometrial
implant survival
1.79E+01
-
-
4.73E+02
-
-
-
-
-
O
O
"Average administered daily dose over the experimental exposure period.
bHED based on lst-orderbody burden model described in Section 3.2.4.4.
°HED based on Emond rodent and human PBPK models described in Section 3.3.6.
dBMR = 0.1 for quantal endpoints and 1 standard deviation control mean for continuous endpoints, except for body and organ weights, where BMR = 10%
relative deviation from control mean.
- = value not established or not modeled.
-------
Table 4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as LASCa)
Study
NOAEL/
LOAEL
Endpoint
Control
response
First
responseb
Max
response0
Model fit detail
BMD/
BMDL
Comments
Smialowicz et
al., 2008
(mouse)
None/
2.41E+2
PFC per spleen
(n = 15)
24% |
(0.5 SD)
89% |
Continuous power,
unrestricted
(p = 0.27)
6.54E+3
2.07E+3
BMDL > LOAEL; fit at control and
low dose inconsistent with data;
constrained parameters in other models
hit lower bounds
PFC per 10A6 cells
(« = 8-15)
24% |
(0.5 SD)
9.3-fold |
Continuous power
unrestricted, constant
variance
(p = 0.48)
1.05E+3
1.19E+2
Constant variance test failed; observed
control variance underestimated by
35%; poor fits for all non-constant
variance models
Li et al., 2006
(mouse)
None/
8.75E+1
serum estradiol
(n = 10)
2.0-fold t
(0.8 SD)
2.4-fold t
Continuous linear
(p = 0.16)
8.85E+3
2.96E+3
BMDL > LOAEL; high control CV
(1.25); near maximal response at low
dose; nonmonotonic response; other
model fits are step-function-like
serum progesterone
(,n = 10)
33% |
(2.0 SD)
61% |
Continuous Hill
(p = 0.44)
8.95E-1
6.31E-3
No response data near BMR; large CVs
(>1) for treatment groups; poor fit for
variance model; Hill coefficient at
lower bound (step-function);
implausible BMDL
Tothetal.,
1979
(mouse)
None/
3.15E+2
skin lesions
(,n = 38-44)
0/38
5/44
25/43
Dichotomous log-logistic
(p = 0.67)
1.1E-2
4.7E-4
Constrained parameter lower bound hit;
implausible BMDL
Dichotomous
log-logistic, unrestricted
(p = 0.98)
2.60E+2
3.18E+0
Supralinear fit (slope = 0.48);
implausible BMDL
dermal amyloidosis
(n = 38-44)
0/38
5/44
17/43
Dichotomous log-logistic
(p = 0.02)
8.3E+3
4.8E+3
Poor fit; constrained parameter lower
bound hit; BMDL > LOAEL
Dichotomous
log-logistic, unrestricted
(p = 0.90)
2.67E+2
2.93E+0
Supralinear fit (slope = 0.33);
implausible BMDL
-------
0 ^
&
to
s
S
to
s
>;*
a
a,
Sf
^s
TO
TO'
*
^5
«
0
_L^ >;
1 TO
LtJ ^
OJ 0
S
1
a,
3 §-
C^ TO
>;
a
H o
.. ^k
TO
O
s
2! »
O ^
H r?
<*^
hh Oq
H to
W|
o £?
o
H
ffl
Table 4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as LASCa) (continued)
Study
NOAEL/
LOAEL
Endpoint
Control
response
First
response
Max
response
Model fit detail
BMD/
BMDL
Comments
Latchoumy-
candane and
Mathur, 2002
(rat)
None/
4.37E+2
daily sperm
production
(n = 6)
29% |
(1.0 SD)
41% |
Continuous Hill,
restricted
(p = 0.98)
3.39E+2
1.51E-2
Near maximal response at LOAEL;
constrained parameter bound hit;
implausible BMDL; standard
deviations given in paper interpreted as
standard errors
Continuous Hill,
unrestricted
(p = 0.96)
3.32E+2
8.77E-3
Slightly supralinear fit (n =0.91);
implausible BMDL
NTP, 1982
(mouse)
None/
4.20E+2
Toxic hepatitis;
males
(n = 50)
1/73
5/49
44/50
Dichotomous multistage
(p = 0.01)
8.3E+2
3.7E+2
No acceptable model fits; lowest
BMDL shown
White et al.,
1986
(mouse)
6.03E+2
Total hemolytic
complement activity
(CH50)
(» = 8)
41% |
(2.6 SD)
81% |
Continuous
Hill, restricted
(p = 0.002)
4.76E+3
8.56E+2
Poor fit; no response near BMR;
constrained parameter bound hit;
BMDL > LOAEL
Continuous
Hill, unrestricted
(p = 0.07)
8.17E+1
7.55E+1
Supralinear fit (n = 0.25); implausible
BMDL
Keller et al.,
2007, 2008a, b
(mouse)
None/
2.96E+2
Missing molars
(n = 23-36)
0/29
2/23
30/30
Dichotomous 1° multi-
stage
(p = 0.26)
6.01E+2
4.20E+2
Poor fit at first response level; not most
sensitive endpoint; other endpoints not
amenable to BMD modeling
Shi et al., 2007
(rat)
1.88E+2
5.92E+2
Serum estradiol in
female pups
(n = 10)
38% |
(0.4 SD)
62% |
Continuous exponential
(M4)
(p = 0.69)
4.45E+2
1.95E+2
Adequate fit; selected
Cantoni et al.,
1981
(rat)
None/
1.02E+3
Urinary uro-
porhyrins
(n = 4)
2.4-fold t
(5.7 SD)
87-fold t
Continuous exponential
(M2)
(p = 0.0003)
2.07E+3
1.52E+3
No response near BMR; poor fits for all
non-constant variance models; constant
variance poor representation of control
SD; BMDL > LOAEL
O
O
-------
Table 4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as LASCa) (continued)
Study
NOAEL/
LOAEL
Endpoint
Control
response
First
response
Max
response
Model fit detail
BMD/
BMDL
Comments
Cantoni et al.,
1981
(rat, continued)
None/
1.02E+3
(cont.)
Urinary copro-
porhyrins
\n = 4)
2.4-fold t
(3.1 SD)
4.0-fold t
Continuous exponential
(M4)
(p = 0.49)
2.94E+2
9.93E+1
No response near BMR
Continuous power,
unrestricted
(p = 0.61)
1.52E+1
2.31E-6
Supralinear fit (n = 0.30); implausible
BMDL; poor model choice for plateau
effect
NTP, 2006
(rat)
None/
1.41E+3
Hepatocyte
hypertrophy
(n = 53-54)
0/53
19/54
52/53
Dichotomous multistage
(p = 0.03)
5.01E+2
4.33E+2
Poor fits for all models
Alveolar metaplasia
(n = 52-54)
2/53
19/54
46/52
Dichotomous log-logistic
(P = 0.72)
3.58E+2
2.07E+2
No response near BMR
Oval cell hyperplasia
(n = 53-54)
0/53
4/54
53/53
Dichotomous probit
(p = 0.23)
3.13E+3
2.64E+3
Relatively poor fit for control and low
dose groups; negative response
intercept (same for logistic);
implausible model; BMDL > LOAEL
Dichotomous Weibull
(p = 0.08)
3.15E+3
2.25E+3
Marginal fit; BMDL > LOAEL
Gingival hyperplasia
(n = 53-54)
1/53
7/54
16/53
Dichotomous log-logistic
(p = 0.06)
3.22E+3
2.25E+3
Poor fit; constrained parameter bound
hit; BMDL > LOAEL
Dichotomous log-
logistic, unrestricted
(p = 0.66)
3.88E+2
6.95E-3
Supralinear fit (slope = 0.33);
implausible BMDL
Eosinophilic focus,
multiple
(n = 53-54)
3/53
8/54
42/53
Dichotomous probit
(p = 0.46)
3.08E+3
2.68E+3
Relatively poor fit to control response;
BMDL > LOAEL
Liver fatty change,
diffuse
(n = 53-54)
0/53
2/54
48/53
Dichotomous Weibull
(P = 0.72)
2.16E+3
1.57E+3
BMDL > LOAEL; otherwise adequate
fit
-------
Table 4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as LASCa) (continued)
Study
NOAEL/
LOAEL
Endpoint
Control
response
First
response
Max
response
Model fit detail
BMD/
BMDL
Comments
NTP, 2006
(rat, continued)
None/
1.41E+3
(cont.)
Liver necrosis
(« = 53-54)
1/53
4/54
17/53
Dichotomous log-probit,
unrestricted
(p = 0.80)
4.13E+3
1.93E+3
Adequate fit; slightly supralinear;
BMDL > LOAEL
Liver pigmentation
(n = 53-54)
4/53
9/54
53/53
Dichotomous log-probit
(p = 0.96)
1.36E+3
1.04E+3
Adequate fit
toxic hepatopathy (n
= 53-54)
0/53
2/54
53/53
Dichotomous multistage,
restricted
(p = 0.81)
2.11E+3
1.70E+3
BMDL > LOAEL; otherwise adequate
fit
Bell et al.,
2007a
(rat)
None/
2.00E+3
Balano-preputial
separation in male
pups
(n = 30 [dams])
1/30
5/30
15/30
Dichotomous log-
logistic, restricted
(p = 0.79)
1.96E+3
1.22E+3
Adequate fit; constrained parameter
bound hit; not litter based
Dichotomous log-
logistic, unrestricted
(P = 0.51)
1.81E+3
2.64E+2
Adequate fit; slightly supralinear
(slope = 0.95); selected
Kociba et al.,
1978
(rat)
8.53E+2
3.94E+3
Uroporphyrin per
creatinine, females
(» = 5)
15% t
(0.48 SD)
89% t
Continuous linear
(p = 0.79)
7.20E+3
5.12E+3
BMDL > LOAEL; otherwise adequate
fit
Urinary
coproporphyria,
females
(» = 5)
67% t
(5.1 SD)
78% t
Continuous exponential
(M4, non-constant var)
(p = 0.01)
8.6E+2
4.0E+2
Poor fit; no response near BMR;
BMDL > LOAEL
Liver lesions
(n = 50)
No data presented
Lung lesions
(n = 50)
No data presented
Markowski et
al., 2001
(rat)
None/
1.23E+3
FR5 run
opportunities
(n = 4-7)
10% |
(0.21 SD)
51% |
Continuous Hill
(p = 0.94)
1.37E+3
7.21E+2
Constrained parameter upper bound hit;
-------
Table 4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as LASCa) (continued)
Study
NOAEL/
LOAEL
Endpoint
Control
response
First
response
Max
response
Model fit detail
BMD/
BMDL
Comments
Markowski et
al., 2001
(rat, continued)
None/
1.23E+3
(cont.)
Continuous power,
unrestricted
(p = 0.13)
2.11E+3
8.15E-12
Saturated model; supralinear fit (power
= 0.39); BMD/BMDL ratio » 100
FR2 revolutions
(n = 4-7)
—
9% |
(0.15 SD)
43% |
Continuous Hill
(p = 0.65)
1.46E+3
4.76E+2
Constrained parameter bound hit (upper
bound)
Continuous power,
unrestricted
(p = 0.16)
4.54E+3
8.15E-12
Supralinear fit (power =0.32);
implausible BMDL
FR10 run
opportunities
(n = 4-7)
15% 4
(0.24 SD)
57% |
Continuous exponential
(M2)
(p = 0.30)
6.77E+3
2.28E+3
BMDL > LOAEL
Hojo et al.,
2002
(rat)
None/
1.29E+3
DRL reinforce per
min
(« = 12)
55% t
(1.0 SD)
80% t
Continuous exponential
(M4)
(p = 0.054)
1.04E+3
4.94E+0
Poor fit; near maximal response at
lowest dose, BMD/BMDL ratio » 100
DRL response per
min
(« = 12)
105% |
(2.4 SD)
105% |
Continuous exponential
(M4)
(p = 0.48)
3.02E+2
7.55E+0
No response data near BMR; maximal
response at lowest dose, BMD/BMDL
ratio » 20
Murray et al.,
1979
(rat)
6.19E+2/
3.24E+3
fertility in f2 gen.
(no litters)
(n = 20)
4/32
0/20
9/20
Dichotomous multistage
(p = 0.08)
1.50E+3
7.50E+2
Poor fit; non-mo no tonic response; no
response data near BMR
Kattainen et
al., 2001
(rat)
None/
1.76E+3
3rd molar length in
pups
(,n = 4-8)
15% |
(4.2 SD)
27% |
Continuous Hill,
restricted
(/?<0.01)
2.48E+2
1.33E+2
No response data near BMR;
Constrained parameter lower bound hit
Continuous Hill,
unrestricted
(/?<0.01)
2.01E+5
BMDL could not be calculated
-------
Table 4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as LASCa) (continued)
Study
NOAEL/
LOAEL
Endpoint
Control
response
First
response
Max
response
Model fit detail
BMD/
BMDL
Comments
Kattainen et
al., 2001
(rat, continued)
None/
1.76E+3
(cont.)
3rd molar eruption in
pups
(n = 4-8)
1/16
3/17
13/19
Dichotomous log-
logistic, restricted
(p = 0.98)
1.90E+3
1.05E+3
Constrained parameter lower bound hit
Dichotomous log-
logistic, unrestricted
(p = 0.95)
1.53E+3
1.46E+2
Supralinear fit (slope = 0.91)
Miettinen et
al., 2006
(rat)
None/
1.76E+3
Cariogenic lesions
in pups
(n = 4-8)
25/42
23/29
29/32
Dichotomous log-
logistic, restricted
(p = 0.60)
1.13E+3
4.09E+2
Constrained parameter lower bound hit;
near maximal response at LOAEL; high
control response
Dichotomous log-
logistic, unrestricted
(p = 0.73)
3.91E+1
Supralinear fit (slope = 0.47); BMDL
could not be calculated
Ohsako et al.,
2001
(rat)
8.45E+2/
2.76E+3
Ano-genital distance
in male pups
(« = 5)
12% |
(1.0 SD)
17% |
Continuous Hill,
restricted
(p = 0.26)
3.63E+3
8.05E+2
Constrained parameter lower bound hit;
near maximal response at LOAEL
Continuous Hill,
unrestricted
(P = 0.11)
4.74E+3
4.52E+2
Supralinear fit (n = 0.62)
Amin etal.,
2000
(rat)
None/
2.67E+3
Saccharin
consumed, female,
(0.25%) (n = 10)
22% |
(0.3 SD)
66% |
Continuous linear
(p = 0.55)
7.22E+3
4.81E+3
BMDL > LOAEL; restricted power
model, constrained parameter hit lower
bound
Continuous power,
unrestricted
(p = NA)
6.61E+3
2.70E+3
Saturated model; supralinear fit (power
= 0.74)
Saccharin
consumed, female
(0.50%) (n = 10)
49% |
(0.7 SD)
80% |
Continuous linear
(p = 0.06)
8.02E+3
5.18E+3
Restricted power model, constrained
parameter hit lower bound
-------
Table 4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as LASCa) (continued)
Study
NOAEL/
LOAEL
Endpoint
Control
response
First
response
Max
response
Model fit detail
BMD/
BMDL
Comments
Amin etal.,
2000
(continued)
None/
2.67E+3
(cont.)
Continuous power,
unrestricted
(p = NA)
5.19E+3
9.41E+2
Saturated model; supralinear fit (power
= 0.40)
Saccharin
preference ratio,
female (0.25%)
(n = 10)
29% |
(1.8 SD)
33% |
Continuous linear
(p = 0.002)
9.17E+3
4.39E+3
BMDL > LOAEL; no response near
BMR; near maximal response at
LOAEL
Saccharin
preference ratio,
female (0.50%)
(n = 10)
39% |
(1.1 SD)
54% |
Continuous linear
(p = 0.14)
6.43E+3
4.03E+3
BMDL > LOAEL; near maximal
response at LOAEL; restricted power
model, constrained parameter hit lower
bound
Continuous power,
unrestricted
(p = NA)
2.05E+3
1.26E-5
Saturated model; supralinear fit (power
= 0.28)
Crofton et al.,
2005
(rat)
1.91E+3/
5.10E+3
Serum T4,
(n = 4-14)
29% |
(1.9 SD)
51% |
Continuous exponential
(M4)
(p = 0.94)
2.86E+3
1.67E+3
No response near BMR
Sewall et al.,
1995
(rat)
3.92E+3
9.16E+3
Serum T4
(n = 9)
—
9.1%|
(0.6 SD)
40% |
Continuous Hill
(p = 0.90)
5.68E+3
1.98E+3
Constrained parameter hit lower bound;
otherwise acceptable fit
Continuous Hill,
unrestricted (p = 0.90)
5.35E+3
1.09E+3
Supralinear fit (power = 0.57);
otherwise acceptable fit
Schantz et al.,
1996
(rat)
None/
None
Maze errors per
block, female
(« = 10)
22% |
(0.2 SD)
34% |
Continuous linear
(p = 0.16)
5.53E+3
3.63E+3
BMDL > LOAEL; near maximal
response at LOAEL; restricted power
model, constrained parameter hit lower
bound
Continuous power,
unrestricted
(p = NA)
2.02E+3
8.11E-6
Saturated model; supralinear fit (power
= 0.37); implausible BMDL
Li etal., 1997
(rat)
1.46E+2/
4.40E+2
FSH in female rats
(n = 10)
3.6-fold t
(2.0 SD)
19-fold t
Continuous power,
restricted
(/?<0.01)
1.11E+5
7.50E+4
Power hit lower bound
-------
Table 4-4. TCDD BMDL analysis (NOAEL, LOAEL, BMD, and BMDL values given as LASCa) (continued)
0 <5;
&
to
s
S
to
s
>;*
a
a,
Sf
TO
TO'
*
^5
«
0
>;
1 TO
LtJ ^
vo o
s
1
a,
3 §-
TO
>;
a
H o
..
TO
O
s
2! »
O ^
H r?
hh Oq
H to
W|
o
o
H
ffl
Study
NOAEL/
LOAEL
Endpoint
Control
response
First
response
Max
response
Model fit detail
BMD/
BMDL
Comments
Van Birgelen
et al., 1995ab
(rat)
None/
3.97E+3
Hepatitic retinol
(« = 8)
—
44% |
(0.74 SD)
96% |
Continuous exponential
(M4) (p < 0.01)
1.37E+4
1.85E+3
Poor fit
Continuous power,
unrestricted (p = 0.01)
1.03E+5
7.92E+4
Poor fit; supralinear fit (power =0.14),
Hepatitic retinyl
palmitate (n = 8)
—
80% |
(1.4 SD)
99% |
Continuous exponential
(M4) (p < 0.01)
7.79E+4
2.01E+4
Poor fit; no response near BMR
Continuous power,
unrestricted (p = 0.24)
2.90E+1
3.25E-2
Supralinear fit (power = 0.06);
implausible BMDL
o
o
'Converted to whole blood concentrations as described in Section 3 prior to determining HEDs in Table 4-5.
bMagnitude of response at first dose where response differs from control value (in the adverse direction); continuous response magnitudes given as relative to
control; quantal response given as number affected/total number.
°Magnitude of response maximally differing from control value (in the adverse direction).
SD = standard deviation.
-------
o
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
^s
^s
TO
TO'
*
^5
«
s
TO
>;
0
s
1
a,
§•
TO
>;
s
o
^•k
TO
O
s
2! »
O ^
H r?
<*^
hh Oq
H to
W|
O £?
o
H
W
Table 4-5. Candidate points of departure for the TCDD RfD using blood-concentration-based human
equivalent doses
Study
Species, strain
(sex, if not
both)
Protocol
Endpoint
NOAELhed(N) or
BMDLhed (B)
(ng/kg-day)
LOAELhed
(ng/kg-day)
UFa
RfD
(mg/kg-day)
Li et al., 2006
Mouse, NIH (F)
Gavage GD 1-3;
n = 10
Hormone levels in pregnant dams (decreased
progesterone, increased estradiol)
-
1.6E-03
300
5.3E-12
Smialowicz et
al., 2008
Mouse, B6C3F1
(F)
90-day gavage;
n = 8-15
Decreased SRBC response
-
6.4E-03
300
2.1E-11
Keller et al.,
2007, 2008a, bb
Mouse, CBA/J
and C3H/HeJ
Gavage GD 13;
n = 23-36 (pups)
Missing molars, mandibular shape changes in
pups
-
9.8E-03
300
3.3E-11
Tothetal.,
1979
Mouse, Swiss/
H/Riop (M)
1-year gavage;
n = 38—44
Dermal amyloidosis, skin lesions
-
1.0E-02
300
3.3E-11
Latchoumy-
candane and
Mathur, 2002
Rat, Wistar (M)
45-day oral
pipetting;
n = 6
Decreased sperm production
"
1.7E-02
300
5.7E-11
NTP, 1982
Mouse, B6C3F1
(M)
2-year gavage;
n = 50
Liver lesions
-
2.2E-02
300
7.3E-11
White et al.,
1986
Mouse, B6C3F1
(F)
14-day gavage;
n = 6-8
Decreased serum complement
-
2.8E-02
300
9.4E-11
Li et al., 1997
Rat, S-D
(F, 22 day-old)
Single gavage;
n= 10
Increased serum FSH
3.0E-03 (N)
1.7E-02
o
o
co
1.0E-10
DeCaprio et al.,
1986
Guinea pig,
Hartley
90-day dietary;
n = 10
Decreased body weight, organ weight
changes (liver, kidney, thymus, brain)
4.1E-03d (N)
3.3E-02d
o
o
co
1.4E-10
Shi et al., 2007
Rat, S-D (F)
11-month gavage;
n = 10
Decreased serum estradiol
4.7E-03 (N)
5.0E-03 (B)
2.8E-02
o
o
co
1.6E-10
Markowski et
al., 2001
Rat, Holtzman
Gavage GD 18;
n =4-7
Neurobehavioral effects in pups (running,
lever press, wheel spinning)
-
5.2E-02
300
1.7E-10
Hojo et al.,
2002
Rat, S-D
Gavage GD 8;
n = 12
Food-reinforced operant behavior in pups
-
5.5E-02
300
1.8E-10
Vosetal., 1973
Guinea pig,
Hartley (F)
8-week gavage;
n = 10
Decreased delayed-type hypersensitivity
response to tuberculin
6.4E-03d (N)
3.2E-02d
o
o
2.1E-10
o
O
O
-------
$ 1
'Vl
0
Co'
&
to
s
S
to
s
>;*
a
a,
Sf
<§>
^s
^s
TO
TO'
*
^5
«
0
J\. Co
1 ^
4^ co
^ o
s
1
a,
2 §-
7? ^
J> Co
nr\ S
H ©
3 §
2 §
o ^
H r?
<*^
hh Oq
H to
W|
o £?
o
H
W
Table 4-5. Candidate points of departure for the TCDD RfD using blood-concentration-based human
equivalent doses (continued)
Study
Species, strain
(sex, if not
both)
Protocol
Endpoint
NOAELhed(N) or
BMDLhed (B)
(ng/kg-day)
LOAELhed
(ng/kg-day)
UFa
RfD
(mg/kg-day)
Cantoni et al.,
1981
Rat, CD-COBS
(F)
45-week gavage;
n = 4
Increased urinary porhyrins
-
6.5E-02
300
2.2E-10
Bell et al., 2007
Rat, CRL:WI
(Han) (M)
17 week dietary;
n =30
Delay in onset of puberty
7.4E-03 (B)
1.1E-01
o
o
2.5E-10
Miettinen et al.,
2006
Rat, Line C
Gavage GD 15;
n = 3-10
Cariogenic lesions in pups
-
8.9E-02
300
3.0E-10
Kattainen et al.,
2001
Rat, Line C
Gavage GD 15;
n = 4-8
Inhibited molar development in pups
-
9.0E-02
300
3.0E-10
NTP, 2006
Rat, S-D (F)
2-year gavage;
n =53
Liver and lung lesions
-
1.4E-01
300
4.7E-10
Amin etal.,
2000
Rat, S-D
Gavage GD 10-16;
n = 10
Reduced saccharin consumption and
preference
-
1.7E-01
300
5.7E-10
Schantz et al.,
1996
Rat, S-D (F)
Gavage GD 10-16;
n = 10
Altered maze performance
-
1.7E-01
300
5.7E-10
Mocarelli et
al., 2008
Human (M)
Childhood
exposure; n = 157
Decreased sperm cone, sperm motility and
increased estradiol, as adults
-
2.0E-02e
30f
6.7E-10
Baccarelli et
al., 2008
Human infants
Gestational
exposure; n = 51
Increased TSH in newborn infants
-
2.4E-028
30f
8.2E-10
Ohsako et al.,
2001
Rat, Holtzman
Gavage GD 15;
n = 5
Decreased ano-genital distance in male pups
2.9E-02 (N)
1.8E-01
o
o
9.6E-10
Murray et al.,
1979
Rat, S-D
3-generation dietary
Reduced fertility and neonatal survival (f 0
and f 1)
3.0E-02 (N)
3.9E-01
o
o
1.0E-09
Chu et al., 2007
Rat, S-D (F)
28-day gavage,
n = 5
Liver lesions
3.6E-02 (N)
5.8E-01
o
o
co
1.2E-09
Van Birgelen et
al., 1995
Rat, S-D (F)
13-week dietary;
n = 8
Decreased liver retinyl palmitate
-
5.3E-01
300
1.8E-09
-------
S"4
>3*
&
to
S
S
to
s
>;*
a
a,
Sf
^s
TO
TO'
*
^5
«
s
TO
>;
0
s
1
a,
§•
TO
>;
s
o
^•k
TO
O
s
2! »
O ^
H r?
<*^
hh Oq
H to
W|
o £?
o
H
W
Table 4-5. Candidate points of departure for the TCDD RfD using blood-concentration-based human
equivalent doses (continued)
Study
Species, strain
(sex, if not
both)
Protocol
Endpoint
NOAELhed(N) or
BMDLhed (B)
(ng/kg-day)
LOAELm.|)
(ng/kg-day)
UFa
RfD
(mg/kg-day)
Kociba et al.,
1978
Rat, S-D (F)
2-year dietary;
n = 50
Liver and lung lesions, increased urinary
porhyrins
6.5E-02 (N)
6.5E-01
o
o
2.2E-09
Sewall et al.,
1995
Rat, S-D (F)
30-week gavage;
n = 9
Decreased serum T4
5.2E-01 (N)
7.3E-02 (B)
1.8E+00
o
o
2.4E-09
Fattore et al.,
2000
Rat, S-D
13-week dietary;
n = 6
Decreased hepatic retinol
-
8.01E-1
300
2.7E-09
Seo et al., 1995
Rat, S-D
Gavage GD 10-16;
n = 10
Decreased serum T4 and thymus weight
1.7E-01 (N)
9.3E-01
o
o
5.6E-09
Crofton et al.,
2005
Rat, Long-Evans
(F)
4-day gavage;
n = 4-14
Decreased serum T4
1.7E-01 (N)
7.6E-01
o
o
5.7E-09
Alaluusua et al.,
2004
Human
Childhood exposure;
n = 48
Dental defects
1.2E-01h(N)
9.3E-011
J
3.9E-08
to
O
O
aExcept where indicated, UFA = 3 (for dynamics), UFH = 10, UFL = 10.
bResults from 3 separate studies with identical designs combined.
cUFl = 1 (NOAEL or BMDL).
dHED determined from lst-orderbody burden model; no PBPK model available for guinea pigs.
e Mean of peak exposure (0.0319 ng/kg-day) and average exposure over 10-year critical window (0.00802 ng/kg-day).
fUFH = 3, UFl = 10.
gMaternal exposure corresponding to neonatal TSH concentration exceeding 5 |iU/mL.
hMean of peak exposure (0.200 ng/kg-day) and average exposure over 10-year critical window (0.0335 ng/kg-day).
'Mean of peak exposure (1.71 ng/kg-day) and average exposure over 10-year critical window (0.153 ng/kg-day).
jUFh = 3.
S-D = Sprague-Dawley.
-------
Table 4-6. Qualitative analysis of the strengths and limitations/uncertainties associated with animal bioassays
possessing candidate points-of-departure for the TCDD RfD
Study
Strengths
Limitations
Remarks
Li et al, 2006
• Female reproductive effects (i.e., early embryo loss
and changes in serum progesterone and estradiol)
were tested at multiple exposure times—early
gestation, preimplantation, and peri- to
postimplantation.
• Relatively small sample size of mouse dams/dose
employed (n= 10)
Study may have human
relevance based on perceived
TCDD-induced female
reproductive effects; uterine
sequestration of TCDD
observed.
Smialowicz et
al., 2008
• Sheep red blood cell (SRBC) plaque forming cell
assay is highly sensitive and reproducible across
laboratories when examining TCDD
• Small sample size of mice/dose employed (n = 8)
• Only female mice were tested
• Thymus and spleen weights were only other
immune response-related endpoints tested
Study adds to a substantial
database on the
immunotoxicity of TCDD in
laboratory animals
Tothetal., 1979
• Large sample size of mice/dose employed
• One-year exposure duration
• Reporting of findings is terse and lacks sufficient
detail (e.g., materials and methods, thorough
description of pathological findings, etc.)
• Only male mice were tested for amyloidosis and
skin lesions
Study has human relevance
based on similarity of ulcerous
skin lesions and amyloidosis in
mice to chloracne observed in
humans
Latchoumy-
candane and
Mathur, 2002
• Compared to epididymal sperm counts, the
testicular spermatid head count provides better
quantitation of acute changes in sperm production
and can indicate pathology
• Small sample size of rats/dose employed (n = 6)
• Oral pipette administration of TCDD may be a less
efficient dosing method than gavage
Study has human relevance
based on observed TCDD-
induced male reproductive
effects
-------
Table 4-6. Qualitative analysis of the strengths and limitations/uncertainties associated with animal bioassays
possessing candidate points-of-departure for the TCDD RfD (continued)
Study
Strengths
Limitations
Remarks
Keller et al.
2007, 2008a,b
• Six different inbred mouse strains were utilized
• Large sample size of mouse offspring/dose/strain
evaluated
• Low TCDD dose levels used compared to typical
mouse studies allowed for identification of subtle
sensitivity differences in presence of absence of
third molars, variant molar morphology, and
mandible structure in offspring
• Unknown sample size of mouse dams/dose/strain
employed
• All inbred strains possessed sensitive b allele at
the Ahr locus (i.e., a potentially resistant sub-
population was not evaluated for comparison
purposes)
• Morphological dental and mandibular changes
induced by TCDD at earlier or later gestational
dosing dates are unknown because of single
gavage administration on GD 13
• Difficulties breeding A/J mice led to abandonment
of that strain in the analysis (Keller et al.,
2008a, b)
Studies have human relevance
based on observed TCDD-
induced developmental dental
effects.
NTP, 1982
• Large sample size of mice and rats/dose employed
• Comprehensive 2-year bioassay that assessed body
weights, clinical signs, and pathological changes in
multiple tissues and organs
• Elevated background levels of hepatocellular
tumors in untreated male mice
• Gavage exposure was only 2 days/week
One of two comprehensive
chronic toxicity evaluations of
TCDD in rodents
White et al., 1986
• Total hemolytic complement (CH50) is
representative functional assay of the complete
complement sequence
• Small sample size of rats/dose employed (n = 6-8)
• Individual complement factors may be
significantly depleted without affecting CH50
activity (only C3 is measured)
• TCDD used for dosing was of unknown purity
First report on TCDD-induced
effects on serum complement
Shi et al., 2007
• Study design evaluated TCDD effects on aging
female reproductive system (i.e., exposure began in
utero and spanned across reproductive lifespan)
• Several female reproductive endpoints were
evaluated, including cyclicity, endocrinology,
serum hormone levels, and follicular reserves
• Relatively small sample size of rats/dose
employed (n= 10)
Study may have human
relevance based on perceived
TCDD-induced female
reproductive effects
-------
Table 4-6. Qualitative analysis of the strengths and limitations/uncertainties associated with animal bioassays
possessing candidate points-of-departure for the TCDD RfD (continued)
Study
Strengths
Limitations
Remarks
Markowski et al.,
2001
• Low TCDD dose levels used allowed for subtle
behavioral deficits to be identified in rat offspring.
• Several training sessions on wheel apparatuses
were extensive.
• Neurobehavioral effects are exposure-related and
cannot be attributed to motor or sensory deficits.
• Unknown sample size of rat dams/dose employed.
• Small sample size of rat offspring/dose evaluated
(n = 4-7).
• TCDD used for dosing was of unknown purity and
origin.
• Neurobehavioral effects induced by TCDD at
earlier or later gestational dosing dates are
unknown because of single gavage administration
on GD 18.
• Although BMD analysis was conducted, the model
parameters were not constrained according to EPA
guidance, so the results cannot be used.
Study informs a dearth of
information on
neurobehavioral toxicity of
TCDD; observed
neurobehavioral changes
signify meaningful
developmental outcome.
DeCaprio et al.,
1986
• Subchronic oral dosing duration up to 90 days.
• Male and female guinea pigs tested
• Relatively small sample size of guinea pigs/dose
employed (n= 10)
• No histopathological analyses performed
• TCDD used for dosing was of unknown purity
Standard subchronic oral study
Vosetal, 1973
• Three different animal species tested (guinea pigs,
mice, rats)
• Effects of TCDD tested on both cell-mediated and
humoral immunity
• Small sample size of animals/dose employed in
each experiment (n = 5-10)
• Only female guinea pigs and rats were tested, and
only male mice were tested
• Only one experimental assay was utilized to assess
cell-mediated and humoral immunity in each
animal species; humoral immunity was only
investigated in guinea pigs
• TCDD used for dosing was of unknown purity
Early study on the
immunotoxicity of TCDD in
laboratory animals
Cantoni et al.,
1981
• Experiments were designed to test qualitative and
quantitative composition and the course of urinary
excretion in TCDD-induced porphyria
• Small sample size of rats/dose employed (n = 4)
• Concurrent histological changes with tissue
porphyrin levels were not examined
• TCDD used for dosing was of unknown purity
Early study on porphyrogenic
effects of TCDD
-------
Table 4-6. Qualitative analysis of the strengths and limitations/uncertainties associated with animal bioassays
possessing candidate points-of-departure for the TCDD RfD (continued)
Study
Strengths
Limitations
Remarks
Bell et al., 2007
• Large sample size of both rat dams and
offspring/dose employed
• Several developmental effects tested
• Batch-to-batch variation of up to 30% in TCDD
concentration in the diet
• Chronic dosing of dams does not accurately define
gestational period when fetus is especially
sensitive to TCDD-induced toxicity
Study adds to a substantial
database on the developmental
toxicity of TCDD in laboratory
animals
Hojo et al., 2002
• Low TCDD dose levels used allowed for subtle
behavioral deficits to be identified in rat offspring.
• Preliminary training sessions in operant chamber
apparatuses were extensive.
• Neurobehavioral effects are exposure-related and
cannot be attributed to presence of learning or
discrimination deficits.
• Relatively small sample size of rat dams/dose
employed (n= 12).
• Small sample size of rat offspring/dose evaluated
(n = 5-6).
• Neurobehavioral effects induced by TCDD at
earlier or later gestational dosing dates are
unknown because of single gavage administration
on GD 8.
• Although BMD analysis was conducted, the
model parameters were not constrained according
to EPA guidance, so the results cannot be used.
Study informs a dearth of
information on
neurobehavioral toxicity of
TCDD; observed
neurobehavioral changes
signify meaningful
developmental outcome.
NTP, 2006
• Chronic exposure duration with several interim
sacrifices
• Large number of dose groups with close spacing
• Large number of animals per dose group
• Comprehensive suite of endpoints evaluated
• Comprehensive biochemical, clinical and
histopathological tests and measures
• Detailed reporting of results, with individual
animal data presented as well as group summaries
• Evaluation of background exposure to TCDD and
DLCs in feed
• Single species, strain and sex
• Lowest dose tested too high for establishing
NO A F. I.
Study is the most
comprehensive chronic TCDD
toxicity evaluation in rats to
date
-------
Table 4-7. Basis and derivation of the TCDD reference dose
Princi
jal study detail
Study
POD (ng/kg-day)
Critical effects
Mocarelli et al., 2008
0.020 (LOAEL)
Decreased sperm count (20%) and motility (11%)
and decreased estradiol (23%) in men exposed to
TCDD during childhood
Baccarelli et al., 2008
0.024 (LOAEL)
Elevated TSH (> 5 |iU/mL) in neonates
RfD derivation
POD
0.020 ng/kg-day (2.0E-8 mg/kg-day)
UF
30 (UFL = 10, UFh = 3)
RfD
7 x 10"10 (7E-10) mg/kg-day (2.0E-8 - 30)
Uncertainty factors
LOAEL-to-NOAEL
(UFl)
10
No NOAEL established; cannot quantify lower exposure
group in Baccarelli et al. (2008); magnitude of effects at
LOAEL sufficient to require a 10-fold factor.
Human interindividual
variability
(UFh)
3
A factor of 3 (10°'5) is used because the effects were
elicited in sensitive populations. A further reduction to
1 was not made because the sample sizes were relatively
small, which, combined with uncertainty in exposure
estimation, may not fully capture the range of
interindividual variability.
Interspecies extrapolation
(UFa)
1
Human study.
Sub chroni c-to-chroni c
(UFS)
1
Chronic effect levels are not well defined for humans;
however, animal bioassays indicate that developmental
effects are the most sensitive, occurring at doses lower
than other effects noted in chronic studies. Considering
that exposure in the principal studies encompasses the
critical window of susceptibility associated with
development, an UF to account for exposure duration is
not warranted.
Database sufficiency
(UFd)
1
The database for TCDD contains an extensive range of
human and animal studies that examine a
comprehensive set of endpoints. There is no evidence to
suggest that additional data would result in a lower
reference dose.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-47 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
No
Yes
Does the
study provide data
on noncancer effects and TCDD
dose for determining a POD on
\a toxicologically relevant/^
endpoint?
Exclude study from
POD candidate estimation
Include as POD candidate
List of key noncancer epidemiologic studies for
quantitative dose-response analysis of TCDD
Identify a study NOAEL or LOAEL
as a candidate POD
Use kinetic model to estimate
continuous oral daily intake (ng/kg-day)
in the affected study population
Figure 4-1. EPA's process to select and identify candidate PODs from key
epidemiologic studies for use in the noncancer risk assessment of TCDD. For
each noncancer study that qualified for TCDD dose-response assessment using
the study inclusion criteria, EPA first evaluated the dose-response information
developed by the study authors for whether the study provided noncancer effects
and TCDD dose data for a toxicologically relevant endpoint. If such data were
available, then EPA identified a NOAEL or LOAEL as a candidate POD. Then,
EPA used a human kinetic model to estimate the continuous oral daily intake
(ng/kg-day) for the candidate POD that could be used in the derivation of an RfD
based on the study data. If all of this information was available, then the result
was included as a candidate POD.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
No
^ Is the ^
endpoint observed
\at the LOAEL?/
Is the BMDL less
than the LOAEL?
No
No
yes
Yes
Is the endpoint less
than the minimum
LOAEL X 100?
No
Yes
Is the
endpoint under consideration
toxicologically
relevant?
Exclude endpoint
as a POD candidate
Include NOAEL/LOAEL/BMDL
as a POD candidate
List of key noncancer animal studies for
quantitative dose-response analysis of TCDD
Estimate a Human Equivalent Dose (HED)
corresponding to each blood concentration NOAEL, LOAEL, or BMDL
using the Emond human PBPK model
Determine NOAEL, LOAEL, and BMDL (if possible) human equivalent dose
(HED) based on 1 st-order body burden for each study/endpoint combination
Determine a NOAEL, LOAEL, and BMDL (if possible) for each
study/endpoint combination, based on blood concentrations from the
Emond rodent PBPK model
Figure 4-2. EPA's process to select and identify candidate PODs from key
animal bioassays for use in noncancer dose-response analysis of TCDD. For
each noncancer endpoint found in the studies that qualified for TCDD dose-response assessment
using the study inclusion criteria, EPA attempted to estimate six candidate PODs, i.e., estimates of
a NOAEL, LOAEL, and BMDL using administered average daily doses (ADD) and using blood
concentrations from the animal kinetic model. Benchmark dose modeling was not always possible
due to poor model fits. Then for all of the candidate PODs that were estimated, HEDs were
estimated using the human kinetic model. Next, the toxicological relevance of each candidate
POD was evaluated relative to its usefulness in human health risk assessment. Finally, the lowest
group of the toxicologically relevant candidate PODs are selected for final use in derivation of an
RfD.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 4-49 DRAFT—DO NOT CITE OR QUOTE
-------
W y f!
< < Q
O O ^
I IZ CQ
o < o
03
O)
c
'(/)
o
Q
c
0)
E
Q.
o
0)
>
0)
Q
si
£ O
* £
c
TO
£ -
5 £
+j
TO
£!
0
¦O
<
-a
CM
T—
O
o
+
+
LU
LU
o
o
T—
T—
o
o
+
LU
O
(p-6>j/6u '8sop ju0|BAmb0-UBiunH)
ajey uoijssBui aaoj.-8'Z'e'Z
T—
CN
CO
o
O
o
o
LU
LU
LU
LU
o
o
o
o
T—
T—
T—
T—
(y) ezoos 'ie i9 nag
(lAl) 9003 "lB I9 n
(y) 9661- ' |B}0Z}UBHOS
(y) 0003 'ibi9 ujwv
(y) 1.003 "|B19 o>|BSHO
(y) 9003 ' |B I9 U9Uift9!|/\|
(y) 1-003 ' |B19 !>|SM0>|JB|/\|
(lAl) q'B8003 Z003 "|B}0 J0||0>1
(y) 1.003 ' |B}0 U0U|BUB>1
(y) 3003 "|B}0 ofoH
(y) 6Z6I. ' |B}0 ABJjn|/\|
(y) ZOOS "|B}0 !MS
(y) 3003 'JniflB|/\| PUB no}Bn
(y) 9003 'dlN
(lAl) 6Z61 "|B}0 H}01
(y) 8Z6I. "|B}0 Bqpox
(lAl) 3861. 'dlN
(y) 1861 ' |B}0 !UO}UBQ
(y) S66I. ' |B}0 ||BM0S
(y) BS66I. ' |B}0 U0|06j|g UBA
(lAl) 8003 ' |B }0 zo!MO|B|lus
(y) S66I. "|B}0 00s
(y)OOO3'|B}0 0JOHBz|
(y) Z003 "|B}0nno
(lAl) 9861. "|B}0 0}!H/\A
(y) S003 ' |B}0 uoyojQ
(y) Z66I. "ib}0 n
(H) 8003 ' |B}0 !||0JBOOO|/M
(H) 8003 ' |B}0 !||0JBOOBg
(H) V00Z ' |B}0 Bnsnn|B|v
Q
P
U
H
o
u
3
Xfl
©
a
x
c
#o
V*
OC
#c
£
s-
i*
c
0
a
0)
s-
1
u
3
{*5
O
a
*
H
ro
Tt
u
3
OX)
77ws document is a draft for review purposes only and does not constitute Agency policy
1/15/10 4-50 DRAFT: DO NOT CITE OR QUOTE
-------
>
re
TS
6)
O)
E
¥
3
in
o
o.
X
LU
1.0E-06
1.0E-07
1.0E-08
1.0E-09
Human
75 1.0E-10
1.0E-11
1.0E-12 H i i 1 i i i i i i i i i i i i i i i i i i i i i i i i i i i
^ o3? oP c# o?? $ c# c^N oP ^ J? ^ <&>* c# c#
er . er & . & fA' e> ^ &¦.<&" & . & . & er er . er . & & & & .& &¦ er &<&¦&¦
'V^yy ^yy^^w
A~ vpV*v
-cS4 „e.
rCr ,<# ^ ^ £••
*> ^ ^ <*
J-
Figure 4-4. Candidate RfD array.
-------
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
5. CANCER ASSESSMENT
5.1. QUALITATIVE WEIGHT-OF-EVIDENCE CARCINOGEN CLASSIFICATION
FOR 2,3,7,8-TETRACHLORODIBENZO-jp-DIOXIN (TCDD)
5.1.1. Summary of National Academy of Sciences (NAS) Comments on the Qualitative
Weight-of-Evidence Carcinogen Classification for 2,3,7,8-Tetrachlorodibenzo-
jp-Dioxin (TCDD)
In its charge, the National Academy of Scinces (NAS) was requested to comment
specifically on U.S. Environmental Protection Agency (EPA)'s conclusion that
2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) is best characterized as "carcinogenic to humans."
While indicating that distinction between the categories of "carcinogenic to humans" and "likely
to be carcinogenic to humans" is ".. .based more on semantics than on science.
(NAS, 2006a, p. 141) and recommending that EPA "... spend its energies and resources on more
carefully delineating the assumptions used in quantitative risk estimates for TCDD..."
(NAS, 2006a, p. 141) rather than on the qualitative cancer descriptor for TCDD, the NAS
provided the following comments:
.. .the classification of dioxin as "carcinogenic to humans" versus "likely to be
carcinogenic to humans" depends greatly on the definition and interpretation of
the specific criteria used for classification, with the explicit recognition that the
true weight of evidence lies on a continuum with no bright line that easily
distinguishes between these two categories. The committee agreed that, although
the weight of epidemiological evidence that dioxin is a human carcinogen is not
strong, the human data available from occupational cohorts are consistent with a
modest positive association between relatively high body burdens of dioxin and
increased mortality from all cancers. Positive animal studies and mechanistic data
provide additional support for classification of dioxin as a human carcinogen.
However, the committee was split on whether the weight of evidence met all the
necessary criteria described in the cancer guidelines for classification of dioxin as
"carcinogenic to humans." EPA should summarize its rationale for concluding
that dioxin satisfies the criteria set out in the most recent cancer guidelines for
designation as either "carcinogenic to humans" or "likely to be carcinogenic to
humans (NAS, 2006a, p. 140).
If EPA continues to designate dioxin as "carcinogenic to humans," it should
explain whether this conclusion reflects a finding that there is a strong association
between dioxin exposure and human cancer or between dioxin exposure and a key
precursor event of dioxin's mode of action (presumably AhR binding). If EPA's
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
finding reflects the latter association, EPA should explain why that end point
(e.g., AhR binding) represents a "key precursor event (NAS, 2006a, p. 141).
5.1.2. EPA's Response to the NAS Comments on the Qualitative Weight-of-Evidence
Carcinogen Classification for TCDD
A cancer descriptor is used to express the conclusion of the weight of evidence regarding
the carcinogenic hazard potential of a compound. EPA agrees with the NAS committee that
cancer descriptors represent points along a continuum of evidence. Relatedly, EPA
acknowledges that there are gradations and borderline situations that cannot be communicated
through a descriptor and are best clarified by a full weight of evidence narrative.
The 2003 Reassessment contains a detailed discussion of TCDD carcinogenicity in both
humans (Part II, Chapter 7a; 8) and animals (Part II, Chapter 6; 8) as well as an overall summary
of TCDD carcinogenicity (Part III, Chapter 2.2.1). Since the release of the 2003 Reassessment,
the database pertaining to TCDD carcinogenicity has been strengthened and expanded by
numerous publications (see U.S. EPA, 2008b), including a new chronic bioassay in female rats
(NTP, 2006) and several new follow-up epidemiological investigations (see Section 2.4.1 and
references therein). Many of these studies have been published subsequent to the NAS review.
These new data are summarized and evaluated in Section 2.4 of this document.
As noted by the NAS, the 2003 Reassessment was released prior to EPA's publication of
the U.S. EPA Guidelines for Carcinogen Risk Assessment ("2005 Cancer Guidelines"; U.S. EPA,
2005). Using EPA's guidance at the time of its release (U.S. EPA, 1996a), the 2003
Reassessment determined that the available evidence was sufficient to classify TCDD as a
"human carcinogen." The 1996 guidance suggested "human carcinogen" to be an appropriate
descriptor of carcinogenic potential when there is an absence of conclusive epidemiologic
evidence to clearly establish a cause-and-effect relationship between human exposure and
cancer, but there are compelling carcinogenicity data in animals and mechanistic information in
animals and humans demonstrating similar modes of carcinogenic action.
The 2005 Cancer Guidelines (U.S. EPA, 2005) are intended to promote greater use of the
increasing scientific understanding of the mechanisms that underlie the carcinogenic process.
The 2005 Cancer Guidelines expand upon earlier guidance applied in the 2003 Reassessment and
encourage the use of chemical- and site-specific data versus default options, the consideration of
mode of action information and understanding of biological changes, fuller characterization of
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
carcinogenic potential, and consideration of differences in susceptibility. The 2005 Cancer
Guidelines also emphasize the importance of weighing all of available evidence in reaching
conclusions about the human carcinogenic potential of an agent. As noted above, additional
information on TCDD carcinogenicity has been published since the release of the 2003
Reassessment. This information has expanded the TCDD database and provided additional
support for conclusions made in the 2003 Reassessment regarding the carcinogenic potential of
TCDD.
Under the 2005 Cancer Guidelines (U.S. EPA, 2005), TCDD is characterized as
carcinogenic to humans, based on the available data as of 2009. The 2005 Cancer Guidelines
indicate that this descriptor is appropriate when there is convincing epidemiologic evidence of a
causal association between human exposure and cancer or when all of the following conditions
are met (a) there is strong evidence of an association between human exposure and either cancer
or the key precursor events of the agent's mode of action, but not enough for a causal
association, and (b) there is extensive evidence of carcinogenicity in animals, and (c) the mode(s)
of carcinogenic action and associated key precursor events have been identified in animals, and
(d) there is strong evidence that the key precursor events that precede the cancer response in
animals are anticipated to occur in humans and progress to tumors, based on available biological
information.
As noted above, the NAS commented that EPA should ".. .explain whether this
conclusion reflects a finding that there is a strong association between dioxin exposure and
human cancer or between dioxin exposure and a key precursor event of dioxin's mode of action
(presumably AhR binding)" (NAS, 2006a). When evaluating the carcinogenic potential of a
compound, EPA employs a weight of evidence approach in which all available information is
evaluated and considered in reaching a conclusion. The following sections provide a summary
of EPA's weight of evidence evaluation for TCDD.
5.1.2.1. Summary Evaluation of Epidemiologic Evidence of TCDD and Cancer
The available occupational epidemiologic studies provide convincing evidence of an
association between TCDD exposure and all cancer mortality. Among the strongest of these are
the studies of over 5,000 U.S. chemical manufacturing workers (the National Institute for
Occupational Safety and Health [NIOSH] cohort) (Fingerhut et al., 1991; Steenland et al., 1999,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2001; Aylward et al., 1996; Cheng et al., 2006; Collins et al., 2009); a study of nearly
2,500 German workers involved in the production of phenoxy herbicides and chlorophenols (the
Hamburg cohort) (Becher et al., 1996, 1998; Manz et al., 1991; Nagel et al., 1994; Flesch-Janys
et al., 1995, 1998); a study of more than 2,000 Dutch workers in two plants involved in the
synthesis and formulation of phenoxy herbicides and chlorophenols (the Dutch cohort) (Bueno
de Mesquita et al., 1993; Hooiveld et al., 1998); a smaller study of roughly 250 workers involved
in a chemical accident cleanup (the BASF cohort) (Thiess et al., 1982; Zober et al., 1990; Ott and
Zober, 1996); and an international study of more than 18,000 workers exposed to phenoxy
herbicides and chlorophenols (Saracci et al., 1991; Kogevinas et al., 1997) including newer
studies of smaller subsets of these workers (t'Mannetje et al., 2005; McBride et al., 2009a, b).
The findings from these studies have been thoroughly described either in the 2003 Reassessment
or in Section 2.4.1 of this document.
As noted in Section 2.4, there are considerable challenges inherent in addressing potential
sources of confounding from smoking and co-exposure to other carcinogens, (which could
produce inflated or spurious associations), the healthy worker effect, (which could result in
attenuated effects through comparison with a referent background with an inappropriately high
background risk), and quantifying exposure to the populations included in many of these
retrospective studies. The more recent studies of these cohorts have made significant advances
in reducing the potential for bias from the healthy worker effect through use of internal cohort
analyses and/or controlling for potential confounders through statistical adjustment, restriction,
and use of internal comparisons. Although some exposure assessment uncertainties remain,
some of these studies have also collected individual-level TCDD exposure estimates that allow
quantification of effective dose necessary for dose-response modeling. Overall, the occupational
data provide consistent support for an association between exposure to TCDD and increased
cancer mortality.
Additional epidemiologic evidence supporting an association between TCDD exposure
and cancer comes from studies investigating the morbidity and mortality of residents exposed to
TCDD following an accidental release from a chemical plant near Seveso, Italy (the Seveso
cohort) (Bertazzi et al., 1989, 1993, 1997, 2001; Pesatori et al., 1998, 2003; Consonni et al.,
2008; Warner et al., 2002). Pesatori et al. (2003) and Consonni et al. (2008) were not available
at the time the 2003 Reassessment was released. Among individuals with relatively high
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
exposure at Seveso (Zones A and B combined), all-cancer mortality in the 20-year post-accident
period and all-cancer incidence in the 15-year post-accident period failed to exhibit significant
departures from the expected (Bertazzi et al., 2001; Pesatori et al., 2003; Consonni et al., 2008).
However, an increased risk of all-cancer mortality was noted among men 15-20 years after first
exposure; not only is the association similar in magnitude to other studies (relative risk
[RR] = 1.3 ; 95% confidence interval [CI] = 1.0-1.7) but also emphasizes the importance of
consideration of latency (Bertazzi et al., 2001). Furthermore, associations between TCDD and
some specific cancer sites were detected in this cohort, including increased incidence (based on
15 years of follow-up) and mortality (based on 20 years follow-up) from lymphatic and
hematopoietic neoplasms in both males and females from Zones A and B (Consonni et al., 2008).
This excess was primarily due to non-Hodgkin's lymphoma. Additionally, there was an increase
in lung and rectal cancer mortality in men (Bertazzi et al., 2001) and limited evidence of
increased liver cancer incidence in women based on the 15-year follow-up study (Bertazzi et al.,
1993). In a separate analysis of 981 women in Zone A, breast cancer incidence (n= 15) was
associated (a 2-fold increase for a 10-fold increase in serum TCDD) with TCDD measurements
first collected in 1976 and 1977 (Warner et al., 2002). The authors also reported a 2-3-fold
increase in all cancer incidence (n = 21) for the two upper quartiles of TCDD exposure.
Overall, the newer studies of the Seveso cohort have reported significant increases in
cancer incidence and elevations in cancer mortality that were not evident in earlier studies of this
cohort. While these studies demonstrate an association between TCDD exposure and different
types of cancer, one of the main limitations is the small number of cancer cases to assess
site-specific associations with TCDD exposure. Ongoing studies in that cohort should help
further elucidate potential risk for specific cancer types (and other endpoints) associated with
TCDD exposures among this population.
5.1.2.1.1. Evidence for causality.
The evidence for causality for cancer from the human studies is briefly summarized in the
paragraphs that follow and is based on recommendations from the 2005 Cancer Guidelines. It
should be noted that there are methodological limitations of the epidemiologic studies that may
temper some of the conclusions regarding causality. These limitations include limited statistical
power, exposure assessment uncertainty, and lack of control of confounders (e.g., dioxin-like
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
compounds and smoking) in some studies. There is additional uncertainty regarding lack of
organ specificity in TCDD associated cancers, as the most consistent results occurred for
all-cancer mortality. Despite these uncertainties, many of the more recent studies have greatly
improved exposure assessments compared to earlier studies of the same cohorts and have
addressed the potential for confounding and other types of biases.
Temporality—exposure must precede the effect for causal inference. Given the long
induction period for many types of cancers, exposure must precede the effect with a sufficient
latency (i.e., typically 15-20 years for environmental carcinogens). In all the occupational
studies reviewed (with the exception of McBride et al., 2009a), TCDD exposure has preceded
the effect with sufficient latency to be considered causally associated. In the studies of the
Seveso cohort, the follow-up exposure period has now reached 20 years, a latency sufficient to
address some carcinogenic endpoints. Since most of the studies are based on occupational
exposures or accidental releases into the environment, temporality is more readily established
due to the obvious determination of the specific exposure windows prior to disease onset.
Strength of Association—refers to the magnitude of measures of association such as the
ratio of incidence or mortality (e.g., standardized mortality ratio [SMRs], standardized incidence
ratios, RRs, or odds ratios) irrespective of statistical significance. Effect estimates that are large
in magnitude are less likely to be due to chance, bias, or confounding. Reports of modest risk,
however, do not preclude a causal association and may reflect an agent of lower potency, lower
levels of exposure or attenuation due to nondifferential exposure misclassification. The four
occupational cohorts with the highest exposures (NIOSH, Hamburg, Dutch, and BASF)
consistently showed statistically significant, although moderate, elevations in cancer mortality.
When the data were combined, the SMR for all four subcohorts was 1.4 [95% CI = 1.2-1.6]
(IARC, 1997). Based on findings from the International Agency for Research on Cancer (IARC)
Working Group, increases in all cancer (combined) mortality of the magnitude reported for
TCDD have rarely been found in occupational cohort studies (IARC, 1997). Although these
estimates are higher than the all-cancer mortality results among Seveso men (RR =1.1;
95% CI = 1.0-1.3), they are comparable to the risk estimated in this population (RR = 1.3;
95% CI = 1.0-1.7) 15-20 years after first exposure. These consistent results comparable in
magnitude from the occupational cohorts and Seveso population are not likely due to chance.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
The occupational cohort studies also show an increased risk for lung cancer in the
previously mentioned four subcohorts. The relative risk for lung cancer in the combined highly
exposed subcohorts was estimated to be 1.4 (95% CI = 1.1-1.7) (IARC, 1997). This is
consistent with the lung cancer mortality findings for the highest exposed group of men in
Seveso (RR = 1.3; 95% CI = 1.0-1.7). Additionally, there was an increase in rectal cancer
mortality in the Seveso cohort (RR = 2.4; 95% CI = 1.2-4.6) (Bertazzi et al., 2001) with a
corresponding increase in incidence. Consistent relative risks of more than two were also
detected for rectal cancer in the Hamburg and New Zealand cohorts, but increased risks were not
found in the other cohorts. Although there was limited evidence of increased incidence or
morality from hepatobiliary cancers across the cohorts, liver cancer incidence was elevated in the
15-year postaccident period among women in the Seveso cohort (RR = 2.4; 95% CI = 1.1-5.1,
Warner et al., 2002). An association in this population was also detected for between breast
cancer incidence (RR = 2.1; 95% CI = 1.0-4.6) and serum TCDD levels (per a 10-fold increase
in serum TCDD). Although findings were based on small numbers, three- and four-fold
increased risks of soft tissue sarcoma were detected among the NIOSH (Collins et al., 2009) and
New Zealand cohorts (McBride et al., 2009a). No other cases of this very rare cancer were
detected in the exposed populations from the other cohorts.
Elevated risk of lymphohemopoietic cancer mortality was noted among the Seveso cohort
(RR = 1.7; 95% CI = 1.2, 2.5) (Consonni et al., 2008). Increased SMRs for lymphohemopoietic
cancer comparable in magnitude (range: 1.6-2.2) were also detected among the Hamburg and
New Zealand occupational cohorts, but limited evidence (range: 1.0 to 1.2) of increased
mortality was found in the BASF, NIOSH and Ranch Hands employees (Ott and Zober, 1996;
Steenland et al., 1999; Akhtar et al., 2004). Most of the lymphohemopoietic cancer mortality
risk was reportedly due to non-Hodgkin's lymphoma in most of the cohorts. Relative risks for
non-Hodgkin's lymphoma among TCDD exposed populations from the NIOSH, Hamburg, New
Zealand, Dutch, and Seveso cohorts ranged from 1.2 to 3.8. Although statistical power was
limited in most of these studies, relative risks exceeded 3.0 for non-Hodgkin's lymphoma in
three of these cohorts (Flesch-Janys et al., 1998; Consonni et al., 2008; Hooiveld et al., 1998).
Consistency—the observation of the same site-specific effect across several independent
study populations strengthens an inference of causality. Despite differences across occupational
cohorts, most studies have consistently reported increases in all-cancer mortality with TCDD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
exposure. Several of these studies have also reported increases in lung cancer related to TCDD
exposure. As noted above, there is also suggestive evidence of an increased risk in all-cancer
and lung cancer mortality among the Seveso cohort consistent in magnitude to the occupational
cohorts. Elevated risk of lymphohemopoietic cancer mortality consistent in magnitude
(range: 1.6-2.2) was also detected among the Seveso, Hamburg and New Zealand cohorts. An
increased risk for non-Hodgkin's lymphoma was found in two of the occupational cohorts as
well as in the Seveso cohort, although the relative risks largely did not achieve statistical
significance. Among those studies detecting an association, consistent two-fold relative risks
were found for rectal cancer (McBride et al., 2009a; Flesch-Janys et al., 1998; Bertazzi et al.,
2001) and relative risks in excess of three were detected for soft tissue sarcoma (McBride et al.,
2009a; Collins et al., 2009).
Biological Gradient—refers to the presence of a dose-response and/or duration-response
between a health outcome and exposure of interest. Several of the occupational cohort studies
(Ott and Zober, 1996; Flesch-Janys et al., 1998; Manz et al., 1991; Steenland et al., 1999;
Michalek and Pavuk, 2008) found evidence of a dose-response relationship for all cancers and
various TCDD exposure measures. The SMR analyses based on internal comparisons within the
occupational cohorts show a biological gradient by comparing highly TCDD exposed workers to
low or unexposed workers. A biological gradient was also demonstrated in the Seveso cohort by
comparing highly exposed individuals (Zones A and B) to individuals in lower exposure zones
(Zones C and R). Warner et al. (2002) also reported evidence of a dose-response trend for breast
cancer and increasing TCDD exposures.
Biological Plausibility—refers to the observed effect having some biological link to the
exposure. Most evidence suggests that toxic effects of TCDD are mediated by interaction with
the aryl hydrocarbon receptor (AhR). AhR is a highly conserved protein among mammals,
including humans (Fujii-Kuriyama et al., 1995; Harper et al., 2002; Nebert et al., 1991). Several
hypothesized modes of action have been presented for TCDD-induced tumors in rodents, all
involving AhR activation. The available evidence does not preclude the relevance of these
hypothesized modes of action to humans.
Specificity—as originally intended, this refers to increased inference of causation if a
single site effect, as opposed to multiple effects, is observed and associated with exposure.
Based on current biological understanding, this is now considered one of the weaker guidelines
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
for causality. As stated in the 2005 Cancer Guidelines, given the current understanding that
many agents cause cancer at multiple sites, and cancers have multiple causes, the absence of
specificity does not detract from evidence for a causal effect. Given that the most consistent
findings associating TCDD and cancer are for all-cause cancer mortality, epidemiological
evidence suggests that TCDD lacks specificity for particular tumor sites. A key event in
TCDD's mechanism of action is binding to and activating AhR; however, downstream events
leading to tumor formation are uncertain and may likely be tissue specific. Given that the AhR is
highly conserved within species and is expressed in various human tissues, the lack of tumor site
specificity does not preclude a determination of causality.
In summary, EPA finds the available epidemiological information provides strong
evidence of an association between TCDD exposure and human cancer that cannot be reasonably
attributed to chance or confounding and other types of bias, and with a demonstration of
temporality, strength of association, consistency, biological plausibility, and a biological
gradient. Additional evidence from animal studies and from mechanistic studies (described
below) provides additional support for the classification of TCDD as carcinogenic to humans.
5.1.2.2. Summary of Evidence for TCDD Carcinogenicity in Experimental Animals
An extensive database on the carcinogenicity of TCDD in experimental animals is
described in detail in Part II, Chapter 6 of the 2003 Reassessment. There is substantial evidence
that TCDD is carcinogenic in experimental animals based on long-term bioassays conducted in
both sexes of rats and mice (Kociba et al., 1978; NTP, 1982, 2006) and in male hamsters (Rao et
al., 1988). Additionally, National Toxicology Program (NTP, 2006) has completed a new
chronic bioassay in female Sprague Dawley rats. These studies are summarized in Section 2.4.2
of this document. All studies have produced positive results, with TCDD increasing the
incidence of tumors at sites distant from the site of treatment and at doses well below the
maximum tolerated dose. In both sexes of rodents, when administered by different routes and at
low doses, TCDD caused tumors at multiple sites; tumors were observed in liver, lung,
lymphatic system, soft tissue, nasal turbinates, hard palate, thyroid, adrenal, pancreas, and
tongue. The most consistent and best characterized carcinogenic responses to TCDD are in the
rodent liver, lung, and thyroid (discussed below in Section 5.1.2.3).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
5.1.2.3. TCDD Mode of Action
The 2005 Cancer Guidelines defines the term "mode of action" as "a sequence of key
events and processes, starting with interaction of an agent with a cell, proceeding through
operational and anatomical changes, and resulting in cancer formation." A "key event" is an
empirically observable precursor step that is itself a necessary element of the mode of action or is
a biologically based marker for such an element. Mode of action is contrasted with "mechanism
of action," which implies a more detailed understanding and description of events, often at the
molecular level. In the case of TCDD, the terms 'mechanism of action' and 'mode of action' are
often used interchangeably in the scientific literature in reference to TCDD's interaction with the
AhR. A thorough discussion of TCDD's interaction with the AhR can be found in the 2003
Reassessment (Part II, Chapter 2; Part III, Chapter 3), and is summarized below (see
Section 5.1.2.3.1).
Most evidence suggests that the majority of toxic effects of TCDD are mediated by
interaction with the AhR. EPA considers interaction with the AhR to be a necessary, but not
sufficient, event in TCDD carcinogenesis. The sequence of key events following binding of
TCDD to the AhR and that ultimately leads to the development of cancer is unknown.
Therefore, in the strictest sense, TCDD's interaction with the AhR does not constitute a "mode
of action" as defined by the 2005 Cancer Guidelines because information about the progression
of necessary events is lacking. However, AhR binding and activation by TCDD is considered to
be a key event in TCDD carcinogenesis.
5.1.2.3.1. The aryl hydrocarbon receptor (AhR).
While substantial evidence suggests that most toxic effects of TCDD are mediated by
interaction with the AhR, less is known about the complex responses that result in tumor
formation. Nonetheless, a picture is emerging wherein TCDD is considered a
"receptor-mediated carcinogen" in laboratory animals (see Figure 5-1), acting in a manner
similar to peroxisome proliferators, phorbol esters, or estrogen (Woods et al., 2007).
The mechanism of action of TCDD has been extensively studied. TCDD activates the
AhR, a member of the basic helix-loop-helix, Per-Arnt-Sim (bHLH-PAS) family of transcription
factors. AhR is present in most cell types and in the unactivated state is cytosolic and exists in a
complex with chaperone proteins, such as heat shock protein 90 (Hsp90). Binding of TCDD to
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
AhR leads to nuclear translocation and heterodimerization with its partner protein Arnt, another
bHLH-PAS family member. The AhR: Arnt heterodimer binds to specific cognate DNA
sequence elements known as dioxin/xenobiotic response elements (DRE/XRE) present in the
regulatory region of specific genes. Binding of the AhR: Arnt heterodimer to these elements, and
subsequent recruitment of transcription coactivator complexes, leads to increased transcription of
specific genes, known as "target genes." There is a battery of genes affected in this manner and
targets include certain xenobiotic-metabolizing enzymes, such as cytochrome P450 (CYP)l Al,
CYP1A2, CYP2B1, and UDP-glucuronosyltransferase (UGT)1A6 (reviewed in Schwartz and
Appel [2005]). In addition, genes affected by the TCDD/AhR-complex code for both inhibitory
and stimulatory growth factors and their gene products affect cellular growth, differentiation and
homeostasis and have been shown to contribute to carcinogenicity as well as other forms of
toxicity (reviewed in Popp et al. [2006]).
Detailed molecular biology research has been performed to identity the extent of the
genes regulated by AhR (Woods et al., 2007); however a complex and still ill-defined profile
remains. Additionally, it is important to note that the extent of the response of individual genes
may not correlate with site-specific tumorigenicity. For example, while TCDD is ineffective as a
tumor promoter in ovariectomized rats and does not stimulate liver cell proliferation in these
animals, it is still capable of inducing CYP1A2 in roughly the same magnitude as in the intact
female rats (Lucier et al., 1991). Similarly, CYP1 Al induction by TCDD is very similar in male
and female rats even though males are almost completely resistant to TCDD carcinogenicity
(Wyde et al., 2002).
Some of AhR's effects on gene expression may be the result of interaction with other
transcription factors (such as the retinoblastoma protein [Ge and Elferink, 1998], NF-kB [Tian et
al., 1999] or with the tyrosine kinase c-Src (Blankenship and Matsumura, 1997]) rather than via
direct interaction with DNA. By far the most extensive studies involving cross talk between
AhR and another transcription factor are those involving the estrogen receptor alpha (ERa). The
anti-estrogenic properties of TCDD have been well-documented, beginning with the
observations that TCDD repressed estradiol function in rat uterus and liver. The AhR-ERa
cross-talk can be manifested at several levels including direct interaction, association of the
receptors with the other's response element and altered metabolism of estradiol by AhR ligand
(Takemoto et al., 2004). The interactions between AhR/Arnt- and estrogen receptor-dependent
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-11 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
signaling pathways, which mediate anti-estrogenic effects of dioxins and dioxin-like
polychlorinated biphenyls (PCBs; Bock, 1994), is probably causal for the well-documented
gender-specificity of the carcinogenic effects of these agents (e.g., hepatocarcinogenicity of
TCCD in female as opposed to male rats) (Lucier et al., 1991). In addition, cross-talk between
AhR/Arnt and other nuclear receptors, their coactivators, and corepressors, has been described.
In fact, cross-talk has been reported for AhR and numerous signaling pathways involved in a
broad range of physiological processes. The molecular mechanisms by which the AhR interferes
with these signaling networks are multifaceted and occur at multiple levels of regulation (many
beyond transcriptional control) (Haarmann-Stemmann et al., 2009).
Pertinent to human risk assessment, there are wide inter- and intraspecies differences in
the toxicological responses to TCDD (Ema et al., 1994; Poland and Glover, 1990; Poland et al.,
1994) some of which can be explained by polymorphisms in AhR. For instance there is a
10-fold difference in susceptibility to TCDD-induced toxicity between the TCDD-sensitive
C57BL/6 and the resistant DBA/2 strains of mice (Poland and Glover, 1980) that can be
explained by polymorphic variations in the ligand-binding domain and in the C-terminal region
of the AhR molecule of each strain (Dolwick et al., 1993). Depending on the system examined,
the estimated affinity of binding of TCDD (and related compounds) to the human AhR is about
10-fold lower than that observed to the AhR from "responsive" rodent species and is comparable
to that observed to the AhR from "nonresponsive" mouse strains (Ramadoss and Perdew, 2004).
This reduced affinity is due, in part, to a single amino acid substitution within the ligand binding
domain of the human and "nonresponsive" mouse AhRs (Ramadoss and Perdew, 2004).
However, there is considerable tissue and species variability in response to TCDD that cannot be
ascribed solely to polymorphisms of the AhR gene (Pohjanvirta and Tuomisto, 1994; Geyer et
al., 1997). Although the affinity of binding of TCDD and related compounds to the human AhR
is reduced compared with rodent AhRs, the qualitative and quantitative rank-order potency of
these chemicals is similar.
5.1.2.3.1.1. Other AhR considerations.
In addition to the potent agonist TCDD, there are many other exogenous ligands for the
AhR, including certain polycyclic aromatic hydrocarbons, polychlorinated dibenzofurans, and
PCBs (Bock, 1994). Several natural and endogenous compounds are also regulators of AhR
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
(Chiaro et al., 2008). The classes of endogenous compounds that have been shown to induce
CYP1 and/or activate AhR include: (a) tryptophan metabolites, other indole-containing
molecules, and phenylethylamines (Gielen and Nebert, 1971); (b) tetrapyrroles such as bilirubin
and biliverdin; (c) sterols such as 7-ketocholesterol and the horse steroid equilenin; (d) fatty acid
metabolites, including at least six different prostaglandins (Seidel et al., 2001) and lipoxin A4;
and (e) the ubiquitous second messenger cAMP (reviewed in McMillan and Bradfield [2007] and
Barouki et al. [2007]). The existence of multiple ligands with varying affinity and responses
suggests that "selective receptor modulators" (or SRMs) of the AhR exist. SRMs are ligands for
a receptor that, upon binding, elicit a conformational change in the receptor that results in
differential recruitment of coregulatory molecules to the target gene promoter region, thereby
imparting a different biological activity relative to the prototypical ligand. This phenomenon has
been most studied for nuclear receptors such as the ERa with the classic example being
tamoxifen, which has estrogen-like activity in the uterus but anti-estrogen-like effects in the
breast. Thus, the relative abilities of compounds to stimulate gene expression or other effects
vary in promoter- and cell type-specific manners. It is now apparent that SRMs exist for the
AhR as well (SAhRMs, Fretland et al. [2004]). For example,
6-methyl-l,3,8-trichlorodibenzofuran (6-MCDF), a SAhRM whose structure is similar to that of
the prototypical AhR agonist TCDD, can induce CYP1 Al gene expression in liver but does not
lead to the toxic responses associated with TCDD (Fritz et al., 2009).
It has been demonstrated that AhR-deficient (AhR-/-) mice show no signs of toxicity at
doses of TCDD approximating the lethal dose eliciting 50% response (LD50) dose (200 (J,g/kg) in
mice expressing the AhR (Fernandez-Salguero et al., 1996). However, a single high exposure of
2,000 (J,g/kg to AhR-deficient mice produced several minor lesions including scattered necrosis
and vasculitis in the liver and lungs. These data suggest that a pathway leading to toxicity exists,
albeit at very high doses, that is independent of the AhR. However, these data also indicate that,
at least in mice, the major in vivo effects of TCDD are mediated through the AhR. The finding
of carcinogenicity in hamsters (Rao et al., 1988) is of special interest since hamsters have been
found to be relatively resistant to the lethal effects of TCDD (Henck et al., 1981; Olson et al.,
1980). To date, there have been no chronic bioassay studies of TCDD carcinogenicity in
AhR-deficient transgenic animals.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
There are additional insights into the complexity of TCDD's mechanism of action
involving AhR. Some biochemical responses to TCDD treatment in isolated cells have been
reported in cells lacking Arnt, in cells expressing a mutated Arnt protein and in cells with highly
reduced levels of AhR (Puga et al., 1992; Kolluri et al., 1999), implying either a non nuclear role
of the AhR in mediating these events or an AhR-independent process.
Additionally, recent studies have linked AhR activation in the absence of exogenous
ligand to a multitude of biological effects, ranging from control of mammary tumorigenesis to
regulation of autoimmunity (reviewed in Hahn et al., 2009). Finally, constitutively activated
AhR in rodents has been shown to induce stomach tumors (Andersson et al., 2002). This
indicates that AhR activation alone (i.e., in the absence of ligand) is sufficient to induce tumors.
5.1.2.3.2. TCDD as a tumor promoter.
The role of TCDD as a tumor promoter is discussed in the 2003 Reassessment (Part II,
Chapter 6). The following is a brief summary of the information regarding TCDD as a tumor
promoter.
TCDD is typically designated as a nongenotoxic carcinogen because it does not damage
DNA directly through the formation of DNA adducts, is negative in most short-term assays for
genotoxicity, and is a potent tumor promoter and a weak initiator or noninitiator in multistage
models for chemical carcinogenesis (Pitot et al., 1980; Graham et al., 1988; Lucier et al., 1991;
Clark et al., 1991; Flodstrom and Ahlborg, 1991; Poland et al., 1982). However, mechanisms
have been proposed that support the possibility that TCDD might be indirectly genotoxic, either
through the induction of oxidative stress or by altering the DNA-damaging potential of
exogenous and endogenous compounds, such as estrogens. In addition, there have been
numerous reports demonstrating TCDD-induced modifications of growth factor signaling
pathways and cytokines in experimental animals and cell culture systems. Some of the altered
signaling pathways include those for epidermal growth factor, transforming growth factor alpha,
glucocorticoids, estrogen, tumor necrosis factor-alpha, interleukin 1-beta, plasminogen
inactivating factor-2, and gastrin. Many of these pathways are involved in cell homeostasis,
proliferation, and differentiation and provide plausible mechanisms responsible for the
carcinogenic actions of TCDD.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
5.1.2.3.3. Hypothesized modes of action of TCDD in rodents.
TCDD has been shown to consistently induce multiple tumors in both sexes in several
rodent species. These tumors are observed in various tissues, including (but not limited to):
liver, lung, thyroid, lymphatic system, soft tissue, nasal turbinates, hard palate, adrenal, pancreas,
and tongue. While the mode of action of TCDD in producing cancer has not been elucidated for
any tumor type, the best characterized carcinogenic actions of TCDD are in rodent liver, lung,
and thyroid. The hypothesized mode of action for each of these three tumor types is briefly
discussed below and is described in Figure 5-2. The hypothesized sequence of events following
TCDD interaction with the AhR is markedly different for each of these three tumor types. No
detailed hypothesized mode of action information exists for any of the other reported tumor
types. Further, no single definitive mode of action of TCDD-mediated carcinogenicity has been
identified.
5.1.2.3.3.1. Liver tumors.
The mode of action of TCDD in producing liver cancer in rodents has not been
elucidated. One hypothesized mode of carcinogenic action of TCDD in the liver is mediated
through hepatotoxicity. Generically speaking, TCDD activation of the AhR leads to a variety of
changes in gene expression, which then lead to hepatotoxicity, followed by compensatory
regenerative cellular proliferation and subsequent tumor development. The details of the
mechanism of TCDD-induced hepatotoxicity have not been fully determined but both CYP
induction and oxidative stress have been postulated to be involved (Maronpot et al., 1993;
Viluksela et al., 2000). The enhanced cell proliferation arising from either altered gene
expression or hepatotoxicity, or both, could be the principal factor leading to promotion of
hepatocellular tumors (Whysner and Williams, 1996). The sensitivity of female rat liver to
TCDD, which apparently does not extend to the mouse, depends on ovarian hormones (Lucier et
al., 1991; Wyde et al., 2001). This sensitivity has been ascribed to induction of estradiol
metabolizing enzymes (Graham et al., 1988) and is hypothesized to lead either to generation of
reactive metabolites of endogenous estrogen or to active oxygen species of estrogens. Oxidative
DNA damage has been implicated in liver tumor promotion (Umemura et al., 1999).
A dose-response relationship exists for TCDD-mediated hepatotoxicity, and this parallels
the dose-response relationship for tumor formation (or formation of foci of cellular alteration as a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
surrogate of tumor formation). However, the dose-response relationship for other
TCCD-induced responses such as enhanced gene expression is different from the dose-response
for tumor formation in terms of both efficacy and potency (see Popp et al. [2006] for review). It
is important to note that differences in potency between events (i.e., gene expression versus cell
proliferation) does not necessary imply alternative mechanisms of action.
5.1.2.3.3.2. Lune tumors.
The mode of action of TCDD in producing lung cancer in rodents (predominantly
keratinizing squamous cell carcinoma, [Larsen, 2006]) has not been elucidated. One
hypothesized mechanism of the carcinogenic action of TCDD in the lung involves disruption of
retinoid homeostasis in the liver. Retinoic acids and their corresponding nuclear receptors, the
retinoic acid receptors (RARs) and the retinoid X receptors (RXRs), work together to regulate
cell growth, differentiation, and apoptosis. It is hypothesized that TCDD, through activation of
the AhR, can affect parts of the complex retinoid system and/or other signaling systems
regulated by, and/or cross-talking with, the retinoid system (reviewed in Nilsson and Hakansson,
2002). These effects are then hypothesized to lead to lung tumor development; however, the
mechanisms underlying this hypothesis are not well-defined. Pulmonary squamous proliferative
lesions have been reported following oral exposure to TCDD in rats (Tritscher et al., 2000). In
general, squamous metaplasia with some inflammation is associated with significant forms of
injury via inhalation of toxic compounds but is also seen with vitamin A deficiency (Tritscher et
al., 2000) and gives some credence to this hypothesis.
Another hypothesized mechanism for the carcinogenic action of TCDD in the lung is
through induction of metabolic enzymes. Through activation of AhR and subsequent induction
of metabolizing enzymes (such as CYP1A1), TCDD may enhance bioactivation of other
carcinogens in lung (Tritscher et al., 2000). There have been few studies to support this
hypothesis; however, in a long-term continuous-application study of carcinogenesis using
airborne particulate extract (APE), squamous cell carcinoma occurred in 8 of 17 AhR+/+ mice
(47%) while no tumors were found in AhR-/- mice (Matsumoto et al., 2007). In addition
CYP1 Al was induced in AhR+/+ mice but not in AhR-/- mice in this study. These results
suggest that AhR plays a significant role in APE-induced carcinogenesis in AhR+/+ mice and
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
CYP1A1 activation of carcinogenic polycyclic aromatic hydrocarbons (the primary carcinogenic
component of APE) is also of importance.
5.1.2.3.3.3. Thyroid tumors.
The mode of action of TCDD in producing thyroid cancer in rodents has not been
elucidated. It is hypothesized that TCDD increases the incidence of thyroid tumors through an
extrathyroidal mechanism. The prevailing hypothesis for the induction of thyroid tumors by
TCDD involves the disruption of thyroid hormone homeostasis via induction of Phase II
enzymes UGTs in the liver (reviewed in Brouwer et al., 1998) by an AhR-dependent
transcriptional mechanism (Bock et al., 1998; Nebert et al., 1990). This induction of hepatic
UGT results in increased conjugation and elimination of thyroxine (T4), leading to reduced
serum T4 concentrations. T4 synthesis is controlled by the thyroid stimulating hormone (TSH)
which is under negative and positive regulation from the hypothalamus, pituitary, and thyroid via
thyrotrophin-releasing hormone, TSH, T4, and triiodothyronine. Consequently, the reduced
serum T4 concentrations lead to a decrease in the negative feedback inhibition on the pituitary
gland. This would then lead to a rise in secreted TSH and stimulation of the thyroid. The
persistent induction of UGT by TCDD and the subsequent prolonged stimulation of the thyroid
could result in thyroid follicular cell hyperplasia and hypertrophy of the thyroid, thereby
increasing the risk of progression to neoplasia. Increases in blood TSH levels are consistent with
prolonged stimulation of the thyroid and may represent an early stage in the induction of thyroid
tumors identified in animal bioassays. Statistically significant increases in neonatal blood TSH
levels have been recently been reported in children born to TCDD-exposed mothers in the
Seveso cohort (Baccarelli et al., 2008; discussed in Section 2.4.1.1.1.4.4). Support for this
hypothesis comes from several studies showing that TCDD decreases serum total thyroxine and
free thyroxine concentrations in rats following both single dose and repeated dose exposures
(Bastomsky, 1977; Brouwer et al., 1998; Pohjanvirta et al., 1989; Potter et al., 1983, 1986;
Sewall et al., 1995; Van Birgelen et al., 1995). Further support comes from studies of transgenic
animals in which TCDD exposure resulted in a marked reduction of total thyroxin and free T4
levels in the serum of AhR+/- mice but not AhR-/- mice (Nishimura et al., 2005). Additionally,
gene expression of UGT1A6, CYP1 Al, and CYP1A2 in the liver was markedly induced by
TCDD in AhR+/- but not AhR-/- mice (Nishimura et al., 2005).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
5.1.2.3.4. Summary of TCDD mode of action in rodents.
Overall, there are inadequate data to support the conclusion that any of the particular
mode of action hypotheses described above is operant in TCDD-induced carcinogenesis.
However, the wealth of scientific evidence available indicates that most, if not all, of the
biological and toxic effects of TCDD are mediated by the AhR. Although the receptor may be
necessary for the occurrence of these events, it is not sufficient because other proteins and
conditions are known to affect the activity of the receptor and its ability to alter gene expression
or to induce other effects. Certain studies could be interpreted to indicate AhR-independent
mechanisms, although these studies have not clearly ruled out involvement of the AhR. The
only consistent, but limited, evidence for TCDD-induced effects that do not involve the AhR
comes from studies using AhR-deficient transgenic animals. Here however, only minor effects
occurred following treatment with extremely high doses of TCDD. Thus, a toxic response to
TCDD has AhR interaction as a key event, but there are various species-, cell-, development-,
gender-, and disease-dependent differences in the cellular milieu that can affect the nature and
extent of the response observed.
The findings that many AhR-modulated effects are regulated with distinct specificity
supports the understanding that the molecular and cellular pathways leading to any particular
toxic event are extremely complex. Precise dissection of these events represents a considerable
challenge, especially in that a toxic response may depend on timely modulation of several genes
rather than of just one particular gene, and possibly modulation of these genes in several rather
than just one cell type or tissue.
While a defined mechanism at the molecular level or a defined mode of action for
TCDD-induced carcinogenicity is lacking, EPA concludes the following
• interaction with the AhR is a necessary early event in TCDD carcinogenicity in
experimental animals.
• through interaction with the AhR, TCDD modifies one or more of a number of cellular
processes, such as induction of enzymes, changes in growth factor and/or hormone
regulation, and/or alterations in cellular proliferation and differentiation.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
• AhR activation is anticipated to occur in humans and to progress to tumors. AhR is
present in human cells and tissues, studies using human cells are consistent with the
hypothesis that the AhR mediates TCDD toxicity and no data exist to suggest that the
biological effects of AhR activation by TCDD are precluded in humans.
• non-AhR mediated carcinogenic effects of TCDD are possible.
5.1.3. Summary of the Qualitative Weight of Evidence Classification for TCDD
Under the 2005 Cancer Guidelines (U.S. EPA, 2005), TCDD is characterized as
carcinogenic to humans, based on the available data as of 2009. This conclusion is based on
• Multiple occupational epidemiologic studies showing strong evidence of an association
between TCDD exposure and increased mortality from all cancers.
• Epidemiological studies showing an association between TCDD exposure and certain
cancers in individuals accidentally exposed to TCDD in Seveso, Italy.
• Extensive evidence of carcinogenicity at multiple tumor sites in both sexes of multiple
species of experimental animals.
• General scientific consensus that the mode of TCDD's carcinogenic action in animals
involves AhR-dependent key precursor events and proceeds through modification of one
or more of a number of cellular processes, such as induction of enzymes, changes in
growth factor and/or hormone regulation, and/or alterations in cellular proliferation and
differentiation.
• The human AhR and rodent AhR are similar in structure and function and human and
rodent tissue and organ cultures respond to TCDD in a similar manner and at similar
concentrations.
• General scientific consensus that AhR activation is anticipated to occur in humans and to
progress to tumors.
5.2. QUANTITATIVE CANCER ASSESSMENT
5.2.1. Summary of NAS Comments on Cancer Dose-Response Modeling
5.2.1.1. Choice of Response Level and Characterization of the Statistical Confidence Around
Low Dose Model Predictions
The NAS commented on the low dose model predictions in the 2003 Reassessment,
including EPA's development of ED0i (effective dose eliciting x percent response) estimates for
numerous study/endpoint combinations. The committee also suggested that EPA had not
appropriately characterized the statistical confidence around such model predictions in the low-
response region of the model.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
The committee concludes that EPA did not adequately justify the use of the 1%
response level (the ED0i) as the POD for analyzing epidemiological or animal
bioassay data for both cancer and noncancer effects. The committee recommends
that EPA more explicitly address the importance of the selection of the POD and
its impact on risk estimates by calculating risk estimates using alternative
assumptions (e.g., the ED0s) (NAS, 2006a, p. 18)
It is critical that the model used for determining a POD fits the data well,
especially at the lower end of the observed responses. Whenever feasible,
mechanistic and statistical information should be used to estimate the shape of the
dose-response curve at lower doses. At a minimum, EPA should use rigorous
statistical methods to assess model fit, and to control and reduce the uncertainty of
the POD caused by a poorly fitted model. The overall quality of the study design
is also a critical element in deciding which data sets to use for quantitative
modeling (NAS, 2006a, p. 18).
EPA should ... assess goodness-of-fit of dose-response models for data sets and
provide both upper and lower bounds on central estimates for all statistical
estimates. When quantitation is not possible, EPA should clearly state it and
explain what would be required to achieve quantitation (NAS, 2006a, p. 10).
The NAS also suggested that EPA report information describing the adequacy of dose-
response model fits, particularly in the low-response region. For those cases where biostatistical
modeling was not possible, the NAS recommended that EPA identify the reasons.
The Reassessment should also explicitly address the importance of statistical
assessment of model fit at the lower end and the difficulties in such assessments,
particularly when using summary data from the literature instead of the raw data,
although estimates of the impacts of different choices of models would provide
valuable information about the role of this uncertainty in driving the risk estimates
(NAS, 2006a, p. 73).
5.2.1.2. Model Forms for Predicting Cancer Risks Below the Point of Departure (POD)
The NAS focused much of its review on EPA's derivation of a cancer slope factor.
Specifically, the NAS commented extensively on the selection of the appropriate point of
departure (POD) and the extrapolation of dose response modeling below the POD.
The NAS questioned EPA's choice of a linear, nonthreshold model for extrapolating risk
associated with exposure levels below the POD, concluding that the current scientific evidence
was sufficient to justify the use of nonlinear methods when extrapolating below the POD for
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-20 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
TCDD carcinogenicity. The committee further recommended that EPA include a nonlinear
model for low dose cancer risk estimates as a comparison to the results from the linear model.
The committee concludes that EPA's decision to rely solely on a default linear
model lacked adequate scientific support. The report recommends that EPA
provide risk estimates using both nonlinear and linear methods to extrapolate
below PODs (NAS, 2006a, p. 5).
After reviewing EPA's 2003 Reassessment and additional scientific data
published since completion of the Reassessment, the committee unanimously
agreed that the current weight of scientific evidence on the carcinogenicity of
dioxin is adequate to justify the use of nonlinear methods consistent with a
receptor-mediated response to extrapolate below the POD. The committee points
out that data from NTP released after EPA generated the 2003 Reassessment
provide the most extensive information collected to date about TCDD
carcinogenicity in test animals, and the committee found the NTP results to be
compelling. The committee concludes that EPA should reevaluate how it models
the dose-response relationships for TCDD... (NAS, 2006a, p. 16).
Because EPA's assumption of linearity at doses below the 1% excess risk level
for carcinogenic effects of TCDD, other dioxins, and DLCs is central to the
ultimate determination of regulatory values, it is important to critically address the
available scientific evidence on the most plausible shape of the dose-response
relationship at doses below the POD (LED0i). On the basis of a review of the
literature, including the detailed review prepared by EPA and presented in Part II
of EPA's Dioxin Risk Assessment and new literature available since the last EPA
review, the committee concludes that, although it is not possible to scientifically
prove the absence of linearity at low doses, the scientific evidence, based largely
on mode of action, is adequate to favor the use of a nonlinear model that would
include a threshold response over the use of the default linear assumption (NAS,
2006a, p. 122).
On the whole, the committee concluded that the empirical evidence supports a
nonlinear dose-response below the ED0i, while acknowledging that the possibility
of a linear response cannot be completely ruled out. The Reassessment
emphasizes the lack of such nonlinear models, hence its adoption of the approach
of linear extrapolation below the POD level. Although this approach remains
consistent with the cancer guidelines (EPA 2005, see also Appendix B), EPA
should acknowledge the qualitative evidence of nonlinear dose response in a more
balanced way, continue to fill in the quantitative data gaps, and look for
opportunities to incorporate mechanistic information as it becomes available. The
committee recommends adopting both linear and nonlinear methods of risk
characterization to account for the uncertainty of dose-response relationship shape
below EDqi (NAS, 2006a, p. 72).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-21 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
5.2.2. Overview of EPA Response to NAS Comments on Cancer Dose-Response Modeling
EPA agrees with the NAS that the approaches to cancer dose-response modeling for
TCDD should be clearly communicated and justified. Furthermore, due to the abundance of new
information on TCDD carcinogenicity published since the 2003 Reassessment, EPA has
reevaluated the cancer dose-response modeling for TCDD presented in the 2003 Reassessment.
As detailed below in Section 5.2.3, EPA has conducted an updated cancer dose-response
assessment for TCDD that incorporates key NAS recommendations discussed in this document,
reflects the current state-of-the science in cancer dose-response modeling and integrates new
TCDD carcinogenic information. Detailed responses to the NAS comments summarized above
are found in Section 5.2.3.3.
The 2003 Reassessment presents an extensive dose-response assessment of TCDD and
provides a comprehensive summary of dose-response relationships. The analyses and
discussions synthesized a considerable breadth of data and model types, highlighting the
strengths and weaknesses of the then-available scientific information. Modeling included both
administered dose and steady state body burden dose metrics, taking into account variation in
half-lives of TCDD across species. These body burden calculations used a simple one-
compartment kinetic model based on the assumption of a first-order decrease in the levels of
administered dose as a function of time. An excess risk of 1% was chosen to model the cancer
data, but comparative results were also shown for 5% and 10% excess risk (see Table 8-2 of the
2003 Reassessment). Dose response was also explored thoroughly for a number of in vitro and
biochemical endpoints in addition to the in vivo data analyses, and ranges of these values were
presented (see Figures 8-1, 8-2 and 8-3 of the 2003 Reassessment). Thus, the 2003
Reassessment provides an initial evaluation of the carcinogenic database for TCDD and serves as
the foundation for the analyses presented below.
5.2.3. Updated Cancer Dose-Response Modeling for Derivation of Oral Slope Factor
The following sections describe the dose-response analysis of the cancer data from
epidemiologic cohort studies (see Section 2.4.1) and rodent bioassays (see Section 2.4.2),
concluding with the derivation of oral slope factors for TCDD based on epidemiologic data (see
Section 5.2.3.1) and rodent bioassay data (see Section 5.2.3.2).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
5.2.3.1. Dose-Response Modeling Based on Epidemiologic Cohort Data
The 2003 Reassessment included dose-response analyses and the development of oral
slope factors from the following three occupational cohorts: the NIOSH cohort, the Hamburg
cohort, and the BASF cohort. In this document, EPA determined that specific studies from each
of these cohorts (Steenland et al., 2001; Ott and Zober, 1996; Becher et al., 1998) met the
epidemiologic study inclusion criteria (see Section 2.3.1 and Section 2.4.1). In Section 5.2.3.1.1,
the oral slope factors derived from these studies in the 2003 Reassessment are reviewed. Since
the publication of the 2003 Reassessment, additional cancer epidemiology studies based on these
cohorts have been published in the peer-reviewed literature. Of these, Cheng et al. (2006) met
the epidemiologic study inclusion criteria (see Section 2.3.1 and Section 2.4.1). In
Section 5.2.3.1.2, EPA evaluates the suitability of deriving an oral slope factor from this study
and derives oral slope factor estimates.
Another study that met the current epidemiologic study inclusion criteria (Warner et al.,
2002) was also briefly discussed in the 2003 Reassessment, but an oral slope factor (OSF) was
not derived from that study. In Section 5.2.3.1.2.2, EPA discusses its unsuccessful attempt to use
the categorical results published by Warner et al. (2002) to develop an oral cancer risk estimate.
5.2.3.1.1. Evaluation of Epidemiologic Studies in the 2003 Reassessment for OSF Derivation.
In the 2003 Reassessment, EPA reported dose-response modeling results for three
epidemiologic studies of human occupational cohorts: the NIOSH cohort with data published by
Steenland et al. (2001); the Hamburg cohort with data published by Becher et al. (1998); and the
BASF cohort with data published by Ott and Zober (1996). Each of these studies is summarized
in Section 2.4.1 of this document and in the 2003 Reassessment (Part II, Chapter 8; Part III,
Chapter 5). Furthermore, EPA has evaluated the suitability of these studies for use in TCDD
dose-response modeling and concluded that each of these studies meet the inclusion criteria for
epidemiology studies presented in Section 2.3.1.
Each of these studies reports all cancer mortality as an outcome. Steenland et al. (2001)
and Becher et al. (1998) analyzed cohorts of primarily male workers who experienced
occupational exposures to TCDD over long periods of time, while Ott and Zober (1996) studied
a cohort of primarily male workers who were exposed to high TCDD concentrations at a single
point in time due to an industrial accident.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
The authors of all three of these studies measured and then back-extrapolated TCDD
levels in a subset of workers to estimate exposures during employment and then the authors used
this information to estimate exposures in the remainder of the cohort. These measured TCDD
samples generally were collected decades after the last known occupational exposure. In each
study, the authors relied on TCDD measures in the cohort to back-calculate serum lipid or body
fat levels of TCDD using a simple one-compartment kinetic model based on the assumption of a
first-order decrease in the levels of exposure dose as a function of time. The assumed half-life of
TCDD used in the models varied from study to study. None of the studies sampled TCDD levels
from the entire cohort; for example, Ott and Zober collected samples from 138/243 workers
(57% of the cohort), which was the highest percentage of workers sampled among the three
studies. Steenland et al. (2001) and Becher et al. (1998) used the measured and back-
extrapolated TCDD concentrations to estimate the exposures that were associated with various
occupations within the cohort, and subsequently used this information to develop exposure
matrices (i.e., the TCDD load per unit time for an occupation) that then could be used to estimate
the cumulative dioxin dose for each cohort member. Ott and Zober (1996) used regression
procedures with data on time spent at various occupational tasks to estimate TCDD levels for all
members of the cohort. Following the estimation of worker exposures in each of these three
studies, the studies' authors divided these cohorts into exposure subgroups based on the
estimated TCDD levels.
In the 2003 Reassessment, EPA identified a POD based on a 1% response in cancer
mortality (ED0i) for the Steenland et al. (2001), and the Ott and Zober (1996) studies. EPA
extrapolated from this POD to lower doses using a straight line drawn from the POD to the
origin—zero incremental dose, zero incremental response—to give a probability of extra risk.
Because there was insufficient evidence to support an assumption of nonlinearity, EPA chose to
develop these models using a linear model.
5.2.3.1.1.1. Steenland et al. (2001).
Steenland et al. (2001) developed dose-response models based on TCDD exposures and
all cancer mortalities from eight plants in the NIOSH cohort. Serum lipid levels of TCDD in
1988 were measured in 193 workers at one of these plants. Steenland and coauthors relied on a
first-order kinetic model (assuming a constant 8.7 year half-life) to back-extrapolate to serum
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
TCDD levels at the time of the last occupational exposure. The study authors assigned exposure
estimates to each of the 3,538 workers in the cohort based on a job-exposure matrix. This matrix
was based on (1) an estimated level of contact with TCDD, (2) the degree of TCDD
contamination of the products the workers produced, and (3) the fraction of a workday during
which the worker likely contacted the TCDD-contaminated products. They then estimated each
worker's serum TCDD levels as an area under the concentration curve (AUC) for lipid-adjusted
serum levels over time. The mortality analysis was conducted on 256 cancer decedents.
Several different dose-response models were fit to these data to provide estimates of fatal
cancer risk. The best-fitting model was a Cox regression exposure-response model using the
log(AUC) of TCDD lipid concentration (ppt-year) lagged by 15 years as the exposure metric.
Steenland and colleagues also developed a piecewise linear regression model with no lag, in
which two separate linear slopes were estimated. This analysis assumed a background exposure
of 0.5 pg/kg-day. The lipid concentrations were converted to body burdens by dividing by 4.
The central tendency estimate and lower bound ED0is from the piecewise linear model and their
associated cancer slope factors for the most sensitive endpoint (male cancer mortality) are
presented in Table 5-1.
5.2.3.1.1.2. Becheretal. (1998).
Based on the Hamburg cohort, Becher et al. (1998) reported a dose-response analysis for
all fatal cancers combined. The mortality analysis was conducted in 1992 on 124 cancer
decedents. The exposure variable in the study was the integrated blood levels for TCDD
concentration over time (AUC, ng/kg-years), as estimated by Flesch-Janys et al. (1998); these
were converted to body burdens by dividing by 4. Estimates of the half-life of TCDD, based on
the sample of 48 individuals with repeated measures, were incorporated into a model that back-
extrapolated TCDD exposures to the end of the employment after accounting for the workers'
ages and body fat percentages. These estimated exposure measures were then applied to the
entire cohort, which consisted of all 1,189 regular male employees who were employed for at
least 3 months between 1952 and 1984 at the Boehringer chemical plant in Hamburg, Germany.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Becher et al. (1998) used a Cox regression approach for the dose-response modeling and
developed three models: a multiplicative model, an additive model, and a power model.25 The
response variable in each model was the SMR for total cancer mortality. The models were
calculated with lag times of up to 20 years. The multiplicative model provided the best fit;
however, the study authors judged the fits for all three models to be acceptable. The model
results were used to calculate unit risk estimates derived as the risk of cancer death through age
70 given a daily dose of 1 pg/kg body weight of TCDD minus the risk given no exposure to
TCDD. These calculations were based on background German cancer mortality rates. The
model results were used to calculate cancer risk estimates. The lower bound estimates on the
dose were not available for models published by Becher et al. due to the absence of statistical
parameter measures. The central tendency estimate EDoiS from the three statistical models and
their associated cancer slope factors are presented in Table 5-1.
5.2.3.1.1.3. Ott and Zober (1996).
In the 2003 Reassessment, EPA also developed a dose-response analysis based on a study
reported by Ott and Zober (1996) for cancer incidence and mortality experienced by 243 men,
who were exposed to TCDD in 1953 during an accident at the BASF plant in Germany. The
cohort was followed through 1992. TCDD blood lipid levels were available for 138 of these
men 30 years after the accident. These levels were back-extrapolated and used to estimate the
AUC for TCDD. Body burdens (ng/kg) were estimated by dividing AUC by 4, and steady-state
25The "multiplicative model" set relative risk (RR) equal to cxp(/;c/). where the dose d is the AUC. The "additive
model" set RR = 1 +fid, and the "power model" set RR = c\p(/; log (kcl+1)). The values /; and k are estimated
parameters.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-26 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
body burdens were estimated assuming a constant half-life of approximately 7.1 years.26 Ott and
Zober (1996) used Cox proportional hazard approaches to estimate cancer mortality risk per unit
TCDD dose.27 The central tendency estimate and lower bound ED0iS from the modeling and
their associated cancer slope factors are presented in Table 5-1.
5.2.3.1.2. Evaluation of Epidemiologic Studies Published Since the 2003 Reassessment for
OSF Derivation.
Two additional epidemiological studies that met the study inclusion criteria (see
Section 2.3) for use in dose response modeling as set forth in this document are evaluated in this
section for the estimation of cancer risk estimates. Each of these studies is summarized in
Section 2.4.1 of this document.
5.2.3.1.2.1. Chens et al. (2006).
As discussed in Section 2.4.1.1.1, Cheng et al. (2006) analyzed the relationship between
TCDD dose and all cancer mortality for the same subset of NIOSH workers as analyzed
previously by Steenland et al. (2001). In contrast to Steenland et al., Cheng et al. (2006) used the
"concentration- and age-dependent elimination model" (concentration- and age-dependent
elimination [CADM], discussed in Section 3.3; see also Aylward et al. [2005a]), rather than a
constant 8.7-year half-life, and calculated serum-derived TCDD estimates for use in dose-
response analysis. An important feature of CADM is that it incorporates concentration- and age-
dependent elimination of TCDD from the body, meaning that the effective half-life of TCDD
26Based on the initial body burden (B0) EPA estimated the body burden at time t using the following formula:
k t
B(t) = B0e , where ke is an elimination constant equal to ln(2)/(half-life in years). This implies that the AUC at
Br, -k T
time T after initial exposure is A UC = — (1 -e e ). T in this case was 39 years (time from the accident in 1953 to
K
the follow-up in 1992). Dividing by a lifetime of 71 years (mean age in 1954, 33 years, plus 38 years from 1954 to
the followup in 1992) yields the lifetime mean body burden as:
D i
B = °—(\-ee v In the 2003 Reassessment, EPA converted the steady-state body burden to units of equivalent
mean _, , ^ '
i\he
1 k T
initial dose by dividing by the constant (l - e " ). With the given values for half-life and T, that constant is
l\ke
0.1411 and l/(the constant) is 7.09.
27The model from Ott and Zober has risk proportional to el> d"sc with [1 = ln(1.22). The corresponding slope for the
mean (steady-state) body burden is 7.0851 * log(1.22) * 0.001 (the 0.001 converts nanograms to micrograms).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
elimination varies based on exposure history, body burden, and age of the exposed individuals.
As discussed in Section 3.3, the use of the CADM model to simulate TCDD kinetics in the
NIOSH cohort results in time-integrated body burden estimates four to five times greater than
those obtained with the simple first-order model, with smaller differences between the two
methods at lower exposures.
Following the estimation of dose using the CADM-derived AUC values, Cheng and
colleagues derived dose-response estimates for the NIOSH cohort using Cox regression and
piecewise linear modeling. Their results are summarized in Table 5-2. For comparison, the Cox
regression coefficient from the analysis conducted by Steenland et al. (2001), which relied on a
first-order elimination rate model assuming a constant 8.7-year half-life, is also shown on the
first line of the table. As in Steenland et al. (2001), Cheng et al. (2006) found a much stronger
relationship between cancer mortality and exposure metrics lagged 15 years compared to the
relationships for untagged exposures. Cheng et al. (2006) also noted that the dose-response
relationship plateaued above the 95th percentile of exposure. For exposures lagged 15 years, the
regression coefficient of the linear slope derived by Cheng et al. (2006) was 3.3 x io~6 per
ppt-year lipid-adjusted serum TCDD (in Table III in their analysis—the standard error of this
regression coefficient was 1.4 x 10 6), The upper 5% of the exposure range (individuals
>252,950 ppt-year lipid adjusted serum TCDD) was excluded in estimating this slope. Because
this exclusion reduces the upper portion of the response where the slope is shallow, this likely
better represents the slope in the region of the curve where the fatal cancer risk is increasing with
dose, which is the equivalent of dropping the highest dose in an animal bioassay.
To develop an OSF for TCDD, EPA used data and equations from the Cheng et al. (2006)
in its calculations as follows
• Upper 95th percentile slope (7?q0. To estimate P95 of this regression coefficient, EPA
summed the regression coefficient (P) and the product of 1.96 and the standard error of
the regression coefficient, yielding an estimate of 6.0 x 10 6 per ppt-year lipid adjusted
serum TCDD, as follows
P95 =(3+ 1.96* SE
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
• Background cancer mortality risk estimate (Ro). EPA used an R0 of 0.112 as reported by
Cheng et al. (2006)28
• Risk in the exposed group associated with a 1% extra risk of fatal cancer (7C„). EPA
estimated Rexp to be 0.12088, using the following relationship for extra risk:
D _ D
0 oi = —^ 2.
1-^0
• Incremental cancer mortality risk in the exposed population based on a 1% extra risk
(Rn). Rd, was then calculated to be 8.9 x 10 3 using the following equation:
Rd Rexp Ro
• Cumulative TCDD concentration in the fat compartment for a 1% extra risk (AUCoi).
EPA then estimated AUCoi using Equation 3 from Cheng et al. (2006):
AUCoi = \n(Ru + R(J Ro)!figs
Cheng et al. (2006) assume that the TCDD concentration in fat is the same as ppt-yr lipid
adjusted serum concentration. The AUCoi was calculated to be 1.26 x 104 ppt-yr.
• Oral slope factor associated with 1% extra risk |"OSF(AUCm )1. AUCoi of
1.26 x 104 ppt-yr is used by EPA as a POD for linear extrapolation to zero. OSF(AUC0i)
is calculated to be 7.92 x 10~7 by dividing 0.01 by the AUCoi. OSF(AUCoi) is linear
with the TCDD concentration in fat. Ingested TCDD doses, however, are not linear with
the predicted TCDD fat concentrations in the Emond pharmacokinetic model. Thus, the
OSF(AUCoi) that is linear with TCDD in the fat is not linear with ingested TCDD dose.
Thus, to estimate the fatal cancer risk associated with an oral intake of TCDD, estimates
of both the average TCDD concentration in the fat resulting from the oral intake and the risk
using the OSF(AUC0i) are needed. Next in this report, EPA presents estimates of OSFs for
specific TCDD intake rates based on target risk levels (RLs) of 1 x 10 3, 1 x 10 4, 1 x 10 5,
1 x 10~6, and 1 x 10 7, using the following calculations:
• Area under the TCDD fat concentration curve associated with a target risk level
(AUCrt ) (ppt-yr). For each target risk level, EPA calculated an AUCRLby dividing the
target risk level by the OSF(AUCoi) [i.e., RL/7.92 x 10"7 (ppt-yr)"1].
• Lifetime averaged concentration of TCDD in the fat compartment associated with the
target risk level (FATrt ) (ng/kg). The AUCrl estimates were then further divided by
28In Table IV, Cheng et al. (2006) report two estimates of background fatal cancer risk, R0, for males aged 75 years:
0.112 and 0.124. A R0 estimate of 12.4% was used by Steenland et al. (2001), and 11.2%, as estimated for all males
in the 1999-2001 Surveillance Epidemiology and End Result data set. EPA chose to use 11.2% as this reflects more
current cancer mortality rates in the U.S. population.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
70 years to identify (FATRL) (ng/kg). This step essentially reverses the integration
undertaken to calculate AUC0i.
• Continuous daily TCDD intake Drt (ng/kg-dav) associated with a target risk level over a
lifetime. Using the Emond human physiologically based pharmacokinetic (PBPK)
model, EPA estimated the DRL necessary to achieve the FATRL. Table 5-3 shows the
ingested TCDD doses (Drl) corresponding to these target risk levels.
• Oral slope factor at the target risk level (OSFRT) (per mg/kg-dav). Table 5-3 also shows
that at the target risk levels, the associated OSFRLs range from 3.7 x io5 to 1.3 x io6 per
mg/kg-day. These are calculated as OSFRL = RL/Drl x 106.
Table 5-4 shows analogous results based on the best estimate of regression coefficient
(3.3 x 10 6) for all fatal cancers as reported by Cheng et al. (2006) by comparing lipid-adjusted
serum concentrations, fat concentrations, risk specific dose estimates and equivalent oral slope
factors for risk levels of 1 x 10~3, 1 x 10~4, 1 x 10~5, 1 x 10~6, and 1 x 10~7.
5.2.3.1.2.2. Warner et al (2002).
Warner et al. (2002) is a study of 981 females exposed to elevated TCDD levels
following the Seveso accident of 1976. The TCDD exposure pattern involving a single period of
elevated TCDD exposures followed by an extended period of lower ambient level TCDD
exposures and elimination is similar to that of the BASF cohort (Ott and Zober, 1996). TCDD
levels, measured or estimated in blood lipids shortly after the time of the accident, were available
for all women. These women were divided into four exposure groups of <20, 20-44, 44.1-100,
and >100 ppt. In this cohort, 21 total cancers have been observed; 15 of these were breast cancer
cases and 3 were thyroid cancer cases. Cox proportional hazards modeling showed that the
hazard ratio for breast cancer associated with a 10-fold increase in serum TCDD levels (logio
(TCDD)) was significantly increased to 2.1 (95% CI = 1.0-4.6). Rate ratios (95% CI) for cancer
incidence in these 4 groups were 1.0, 1.0 (0.2-5.5), 2.2 (0.5-10.8) and 2.5 (0.5-11.8). Using a
Cox proportional hazards model and assuming continuous exposure, the rate ratio was 1.7
(0.9-3.4) for each 10-fold increase in serum TCDD; that is, a logio transformation of the
exposure estimates in their analysis was presented. They reported a test for trend ofp = 0.09.
EPA attempted to estimate an ED0i from the modeled results of Warner et al. (2002) from
the statistically significant hazard ratio for breast cancer. However, EPA had to estimate the
slope of the tangent to the log-linear relationship. Because the slope of a log-linear relationship
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
is not constant but varies with dose, and because the lowest exposure measure was well-above
the 1% response region of interest, EPA could not confidently estimate the tangent to the
log-linear relationship. The transformation of the logio dose units to linear units of TCDD
yielded an implausibly low ED0i and correspondingly high cancer risk that was inconsistent with
a visual inspection of the untransformed plot. EPA was not confident in these values for health
risk assessment because of uncertainties in the transformation in the low response region of the
original model. Thus, EPA did not derive an ED0i or oral slope factor for this study.
5.2.3.2. Dose-Response Modeling Based on Animal Bioassay Data
Figure 5-3 provides a summary of the process EPA has utilized to select and identify
candidate TCDD OSFs from key animal bioassays that were identified in Section 2.4.3 of this
document. For each in vivo animal cancer study that qualified for TCDD dose-response
assessment using the study inclusion criteria specified in Section 2.3.2, EPA first selected the
species/sex/tumor data set combinations that had been characterized as having statistically
significant increases in tumor incidence by either a pair-wise test between the treated group and
the controls or by a trend test showing increases in tumors with increases in dose. Next, EPA
used the Emond animal kinetic model discussed in Section 3 to estimate blood concentrations
corresponding to each study's average daily administered doses for use in dose response
modeling. Benchmark dose lower confidence bounds (BMDLois) were then estimated for the
blood concentrations by (1) using the multistage cancer model for each species/sex/tumor
combination within each study and (2) using a Bayesian Markov Chain Monte Carlo framework
that assumes independence of tumors, modeling all tumors together for each species/sex
combination within each study. The final selected models were subjected to goodness-of-fit tests
and visual inspections of fit to the raw data. Thus, for each sex/species combination within each
study, this process generated a BMDLoi for each single tumor type and another BMDLoi for the
combined tumors. Finally, using the Emond human kinetic model discussed in Section 3, human
equivalent doses (BMDLreds) were then estimated for each of the BMDLois and, using a linear
extrapolation, OSFs were calculated by OSF = 0.01/BMDLHed- The lowest OSF for a
species/sex combination for either a single tumor type or all combined tumors was selected as a
candidate OSF for TCDD cancer assessment. These steps in Figure 5-3 are further described in
detail in the following sections.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
5.2.3.2.1. Selection of key data sets.
Based on the study selection criteria outlined in Section 2.3.2 (see Figure 2-3), EPA
selected four animal bioassays for use in the cancer dose-response assessment for TCDD (see
Table 2-6 and Section 2.4.2 for detailed study descriptions). Three of these studies (Kociba et
al., 1978; NTP, 1982; Toth et al., 1979) were evaluated in the 2003 Reassessment, while one
study (NTP, 2006) was published after the 2003 Reassessment was released. The NTP (2006)
study was specifically called out by the NAS (2006a) report as cancer bioassay data that EPA
should evaluate prior to completing its TCDD dose-response assessment. As discussed below,
EPA has chosen to conduct dose-response modeling for a number of tumor types from each of
the sex/species combinations in these studies in order to maximize the amount of information
available to support OSF derivation. Because tumors occurred in multiple sites in the exposed
animals, each tumor type was considered separately (individual tumor models) and were also
combined into composite tumor incidence dose estimates (multiple tumor models).
The tumor incidence tables for these four bioassays are shown in Tables 5-5 through 5-8.
The data in these tables are summarized from each study's reference publication and are the
species/sex/tumor incidence data used for TCDD dose-response modeling in this report. In the
case of the Kociba et al. (1978) female rat combined hepatocellular adenomas and carcinomas
only, EPA used data from a reanalysis of the pathology slides that was published by Goodman
and Sauer (1992). The data sets in Tables 5-5 through 5-12 were selected because they had been
characterized by the study authors as having statistically significant increases in tumor incidence
by either a pair-wise test between the treated group and the controls or by a trend test showing
increases in tumors with increases in dose. One exception is that Goodman and Sauer (1992) did
not statistically analyze the revised tumor incidence data from their reanalysis of the Kociba et
al. (1978) female rat combined hepatocellular adenomas and carcinomas, so a Fischer's Exact
Test to evaluate the statistical significance of those data was used in this document. In the case
of the NTP (2006) study only, information was available regarding the length of time that the
animals stayed on test (105 weeks); animals who died within the first year were censored from
analysis in this document because animals who died within the first year were not considered to
have been alive long enough to develop tumors. Therefore, those animals were not included in
the denominators in Table 5-11. These adjusted incidence data were used in the analysis of
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-32 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
tumor dose-response for NTP (2006) in this document. The tumor incidence data in Tables 5-5
through 5-12 include
• nasal, tongue and adrenal tumors in males, and liver, nasal and lung tumors in females
from the Kociba et al. (1978) 2-year study of Sprague-Dawley rats,
• liver, thyroid and adrenal tumors in males, and subcutaneous tissue, liver, adrenal and
thyroid tumors in females and from the NTP (1982) 2-year study of Osborne-Mendel rats,
• lung and liver tumors in males, and subcutaneous tissue, hematopoietic system, liver and
thyroid tumors in females from the NTP (1982) 2-year study of B6C3Fi mice
• liver, oral mucosa, pancreas and lung tumors in females from the NTP (2006) 2-year
study of Sprague-Dawley rats, and
• liver tumors in males from the Toth et al. (1979) 1-year study of Swiss/H/Riop mice.
For each cancer endpoint, the reported (administered) doses from each study were converted,
where necessary, to average daily doses in ng/kg-day (e.g., doses administered 5 days/week were
adjusted by multiplying by 5 and dividing by 7 to get average daily doses). These doses were
then subjected to kinetic modeling to generate blood concentrations for use in TCDD dose-
response modeling.
5.2.3.2.2. Dose adjustment and extrapolation methods for selected data sets.
5.2.3.2.2.1. Dose metric estimation for dose-response modeling.
Tables 5-5 through 5-12 show the blood concentrations that were used in TCDD dose-
response modeling of the animal bioassay data. Based on kinetic analysis (see Section 3), a
choice for whole blood concentration of TCDD was made for the purpose of dose extrapolation
between animals and humans. In order to estimate blood concentrations for each study selected,
the Emond PBPK model was run using ACSLX® software, version 2.5.0.6 (see Section 3).
Depending on the selected study, either rat or mouse versions of the model were used. In each
case, the simulation was performed using the exposure and observation durations, the body
weights, and the adjusted doses from the original studies. Details of PBPK model input
parameters are given for each study's m-file in Appendix C.2. In the case of Toth et al. (1979)
study, which dosed the animals for a year and then followed up for the lifetime of the animal,
only the one-year simulation was performed. The m-files were used to run the appropriate PBPK
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-33 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
model to estimate time-averaged, maximum, and terminal (end of exposure) lipid adjusted blood
serum concentration (see Appendix C.3). Other model simulated dose metrics such as
concentrations for liver, fat, Ah-receptor bound in liver, body burden, and the time at which the
maximum concentration was reached for each dose metric are also reported for illustrative
purposes in Appendix C.3. The complete results for each study modeled are shown in
Appendix C.3.
Applying the Emond PBPK model to each study/endpoint combinations in Tables 5-5
through 5-12, whole blood concentrations were derived by converting the model estimated lipid-
adjusted to whole blood concentration using model constants for the lipid fraction in blood serum
for rodents (0.0033) and the blood serum fraction (0.55). Once the Emond modeling runs were
performed, the model outputs of whole blood concentrations were used as dose metrics to
estimate BMDLs (in ng/kg). To obtain BMDLs, benchmark dose modeling was performed as
described below by substituting the model simulated dose metrics (whole blood concentrations)
for the original study doses and calculating the corresponding BMDL (results appear below and
in Appendix F).
5.2.3.2.2.2. Calculation of human equivalent doses (HEDs).
Human equivalent doses (ng/kg-day), corresponding to each BMDL (ng/kg) were
calculated using the Emond human PBPK model (see Section 3) and are denoted as BMDLheds.
The Emond human PBPK model was run for 70 years assuming a constant daily dose starting
from birth. The model concentrations were averaged over both the entire 70 year lifetime
(lifetime average) and over the five years surrounding the peak concentration (five-year average)
(see Section 3.3.1, describing first order body burden estimation). The human equivalent doses
were estimated by adjusting the daily dose model input until the time-averaged whole blood
concentration matched the associated alternative dose BMDL (derived earlier from animal PBPK
model). For animal studies which lasted longer than 540 days, the lifetime average was used; for
studies lasting less than 540 days, the five year average was used. The process was iterative and
continued until the modeled human concentration was within 1% of the BMDL. In general,
however, the concentrations matched to within 0.1%.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-34 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
5.2.3.2.3. Dose-response modeling approaches for rodent bioassays.
5.2.3.2.3.1. Modeling of individual tumors.
EPA's BMDS Software, version 2.1 was used to estimate the BMDLois for each of the
species/sex/tumor combinations, using the blood concentrations and incidence data shown in
Tables 5-5 through 5-12. Each data set was modeled using the multistage cancer model, and a
BMDLoi in blood concentration was estimated. The multistage model has been used by EPA in
the majority of its quantitative cancer assessments because it is statistically robust and able to
provide good fits to a wide range of dose-response patterns. It is also consistent with the
multistage nature of the carcinogenic process. The mathematical form of the multistage model is
P(d) = 1 - exp[-(V/0 + q\d + q2d2 + ... + qid)\ (Eq. 5-3)
where
P(d) = lifetime excess risk (probability) of cancer at dose d
qi = parameters estimated in fitting the model, i = 1, ...,k.
To estimate the BMDoiS and BMDLois, BMDS was run with all parameters set to their
defaults; up to three degrees of freedom were specified for the dichotomous, multistage cancer
model; and a 95% confidence level. A 1% benchmark response (BMR) was used for each tumor
type to ensure the PODs were estimated in the linear portion of the dose-response curve. The
BMDLoi (ng/kg) was then converted to a BMDLhed (ng/kg-day) using the Emond human model,
and an OSF in units of (mg/kg-day)-1 was calculated by, OSF = 0.01/BMDLHed x 106. It may be
noted that the OSF is linear with the blood concentration, and is not linear with ingested TCDD
doses; thus, using the Emond et al. pharmacokinetic model, risk-specific doses of TCDD intake
(ng/kg-day) corresponding to the target risk levels would need to be provided for use in human
health risk assessment. In the following sections, results are presented for the models that
provided the best overall fit to the data as judged by comparison of likelihood ratios for models
that had an acceptable fit (chi-squared goodness of fit statisticp > 0.05).
5.2.3.2.3.2. Multiple tumor (Bayesian) models.
Statistically significant increased tumor incidences were observed at multiple sites in
male and/or female rats (Kociba et al., 1978; NTP, 1982, 2006) and male and female mice (NTP,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-35 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
1982) following oral exposures to TCDD. With this multiplicity of tumors, the concern is that a
potency or risk estimate based solely on one tumor site (e.g., the most sensitive site) may
underestimate the overall cancer risk associated with exposure to this chemical. Relevant
approaches in the 2005 Cancer Guidelines (U.S. EPA, 2005) for characterizing total risk include
the following: (1) analyze the incidence of tumor-bearing animals, or (2) combine the potencies
associated with significantly elevated tumors at each site. The NRC (1994) concluded that an
approach based on counts of animals with one or more tumors (tumor-bearing animals) would
tend to underestimate overall risk when tumor types occur independently, and thus an approach
based on combining the risk estimates from each separate tumor type should be used. On
independence of tumors, NRC (1994) stated ".. .a general assumption of statistical independence
of tumor-type occurrences within animals is not likely to introduce substantial error in assessing
carcinogenic potency."
Because potencies are typically upper bound estimates, summing such upper bound
estimates across tumor sites is likely to overstate the overall risk. Therefore, following the
recommendations of the NRC (1994) and the 2005 Cancer Guidelines (U.S. EPA, 2005), a
statistically valid upper bound on combined risk was derived, assuming independence, in order
to gain some understanding of the overall risk resulting from tumors occurring at multiple sites.
In the case of TCDD, tumors are thought to be independent across the sites found in these three
studies because: (1) they are in different organs and tissues, specifically liver, lung, thyroid,
subcutaneous tissue, oral cavity, tongue, pancreas, adrenal cortex and the hematopoietic system;
(2) different kinds of tumors were found, even within the same organ (e.g., both
cholangiocarcinomas and hepatocellular adenomas were found in female rat livers in NTP,
2006); and (3) the tumors found in these studies were not progressive (i.e., they did not
metastasize to other sites in the body). It is important to note that this estimate of overall
potency describes the risk of developing tumors at any combination of the sites and is not the
risk of developing tumors at all sites simultaneously.
For modeling individual tumor data, the multistage model is specified as shown in the
previous section (see Eq. 5-3). Under the assumption of independence, the model for the
combined (or composite) tumor risk is still multistage, with a functional form that has the sum of
stage-specific multistage coefficients as the corresponding multistage coefficient.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-36 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Pc(d) = 1 - exp[-(Lq0i + dZqn + d2I.q2l + ... + cTlqmi)\, for /' = 1,..k, (Eq. 5-4)
where k = total number of sites.
The resulting equation for fixed extra risk (BMR) is polynomial in dose (when logarithms
of both sides are taken) and can be solved in a straightforward manner for the combined BMD.
However, the current version of BMDS cannot estimate confidence bounds for this combined
BMD.
Therefore, a Bayesian approach to finding confidence bounds on the combined BMD was
implemented using WinBUGS (Spiegelhalter et al., 2003). WinBUGS software is freely
available and implements Markov Chain Monte Carlo (MCMC) computations. Use of
WinBUGS has been demonstrated for derivation of a distribution of BMDs for a single
multistage model (Kopylev et al., 2007) and is easily generalized (Kopylev at al., 2009) to derive
the distribution of BMDs for the combined tumor load, following the NRC (1994) methodology
described above. The advantage of a Bayesian approach is that it produces a distribution of
BMDs that allows better characterization of statistical uncertainty. For the current analysis, a
diffuse (high variance or low tolerance) Gaussian prior restricted to be nonnegative was used.
The posterior distribution was based on three simulation chains with 50,000 burn-in (i.e., the
initial 50,000 iterations were dropped) and a thinning rate of 20, resulting in 150,000 interactions
total. The median and 5th percentile of the posterior distribution provided the BMD0i (central
estimate) and BMDLoi (lower bound) for combined tumor load, respectively.
The methodology above was applied to the statistically significant dose-response data
from Kociba et al. (1978), NTP (1982), and NTP (2006) (see Section 2.3.2 for data set selection
criteria). As with the risk estimates generated for individual tumor sites, the combined analysis
used the internal dose metric, whole blood concentration (see Section 3). For the combined
tumors for each sex/species combination, a BMDLoi in blood concentrations was estimated. The
BMDLoi (ng/kg) was then converted to a BMDLhed (ng/kg-day) using the Emond human model,
and an OSF in units of (mg/kg-day)-1 was calculated by, OSF = 0.01/BMDLHed x 106. It may be
noted that the OSF factor is linear with the blood concentration, and is not linear with ingested
TCDD doses; thus, using the Emond et al. pharmacokinetic model, risk-specific doses of TCDD
intake (ng/kg-day) corresponding to the target risk levels would need to be provided for use in
human health risk assessment.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-37 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
5.2.3.2.4. Results of dose-response modeling for rodent bioassays.
Table 5-13 presents the benchmark dose modeling results for both the individual tumors
and the combined tumors based on blood concentrations. The /^-values in the table are for a
chi-square goodness of fit statistic with significance ofp > 0.05. Goodness of fit was acceptable
&tp > 0.05 for all models. The difference in log likelihood (dLL) statistic documents the
difference in log likelihoods between stages of the models in cases where the stage is above 1; it
shows the difference between the stage in the table and the lower stage. For example, for the
NTP (2006) liver cholangiocarcinomas, twice the difference of 2.92 would be >3.84, the test
statistic from the assumed chi-square distribution,29 withp = 0.95, justifying the choice of
3 stages over 2 stages. The best fitting multistage models include a 1-stage (linear) model for all
of the individual tumor data sets from Kociba et al. (1978), NTP (1982), and Toth et al. (1979),
as well as for the pancreatic and oral mucosa tumors in NTP (2006), a 2-stage model for the lung
tumors in NTP (2006), and a 3-stage model for the liver cholangiocarcinoma and liver adenoma
data sets from NTP (2006). For the Toth et al. (1979) liver tumors, the model fit to all of the data
was poor, and the highest dose group was dropped in order to achieve an acceptable fit
(p = 0.29). The BMDoiS and BMDLoiS (ng/kg) were estimated from these multistage models for
the individual tumors. BMDoiS and BMDLoiS (ng/kg) were also provided in Table 5-13 for the
combined tumors for each sex/species combination within a study. These were estimated from
the distributions of BMDoiS produced by the Bayesian MCMC simulation (see
Section 5.2.3.1.2.3.2). The BMDoiS and BMDLoiS (ng/kg) for the combined tumors in
Table 5-13 are the mean and lower 95% percentile values from these distributions, respectively.
5.2.3.2.4.1. Individual tumor models.
Table 5-14 shows the BMDLredS (ng/kg-day) that were estimated from the BMDLoiS in
Table 5-13 using the Emond human model (see Section 5.2.3.1.2.2.2) and the OSFs calculated
by, OSF = 0.01/BMDLhed x 106 to convert the OSF to (mg/kg-day)-1 units. BMDS results,
details of the model fits and dose-response graphics for all endpoints are shown in Appendix F.
29The chi-square distribution with 1 degree of freedom is the correct distribution only under standard conditions
(e.g., no boundary parameters in null hypothesis). Thus, the correct distribution for the situation where the
parameter of interest is on the boundary, as happens with testing for the order of the multistage model, and, possibly
nuisance parameters (estimated parameters of the model), is very difficult to derive (Self and Liang, 1987).
Therefore the p-valuc of chi-square with one degree of freedom is used as the best available choice.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-38 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Although only the blood concentration results are presented in this section, for comparison
purposes, Appendix F also provides modeling results for the studies' administered average daily
doses. Table 5-14 lists the OSFs in decreasing value. It can be seen that liver tumors in male
mice yield the highest slope factors; OSF values are 5.9 xlO6 and 5.2 xlO6 per mg/kg-day in
NTP (1982) and Toth et al. (1979), respectively. The OSFs for the new NTP (2006) study in
female rats are two orders of magnitude lower, ranging from 1.8 xlO4 to 1.8 xlO5 per mg/kg-day,
representing the lowest OSFs for TCDD from the individual tumor models.
5.2.3.2.4.2. Multiple tumor (Bayesian) models.
Table 5-15 shows the BMDLredS (mg/kg-day) that were estimated from the BMDLoiS in
Table 5-13 using the Emond human model (see Section 5.2.3.1.2.2.2) and the OSFs calculated
by, OSF = 0.01/BMDLhed x 106 to convert the OSF to (mg/kg-day)-1 units. Table 5-15 lists the
OSFs in decreasing value. It can be seen that the combined liver and lung tumors in male mice
yield the highest OSF value of 9.4 x 106 per mg/kg-day from NTP (1982), and the combined
adrenal, tongue and nasal tumors in male rats yield the lowest OSF value of 3.2 xio5 from
Kociba et al. (1978). The OSF for the combined liver, oral mucosa, lung, and pancreatic tumors
in female rats from the newer NTP (2006) study is 4.4 x 105.
5.2.3.2.5. Summary evaluation of slope factor estimates from rodent bioassays.
To estimate a range of candidate TCDD OSFs from the animal data, dose-response
modeling of the four chronic rodent bioassays identified in Section 2.4.3 was conducted. Dose-
response modeling was performed using whole blood concentrations, and BMDLhed values
(ng/kg-day) were derived for the 25 species/sex/endpoint data sets individually (see Table 5-14)
and for seven species/sex combined tumor data sets (see Table 5-15).
For each sex/species combination within a study, the combined tumor OSFs presented in
Table 5-15 represent the highest OSFs that have been derived from the animal cancer bioassay
multistage models. The most sensitive species and sex is male mice, for which the estimated
BMDLhed for combined tumors is 1.1 x 10"3 ng/kg-day. This result, which is derived under the
assumption that multiple tumor types occur independently in the exposed animals, is, as
expected, lower than the BMDLhed for the most sensitive individual tumor.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Based on these results, EPA believes that a credible value for the BMDLred derived from
the animal studies lies in the range shown in Table 5-15 between 3.1 x 10"2 and
1.1 x 10 3 ng/kg-day. These values, which correspond to oral slope factor values of 3.2 x io5
and 9.4 x io6 per mg/kg-day, respectively, encompass the range at which elevated cancer risks
can be detected for the most sensitive species, sex, and endpoints in the animal bioassay data.
As noted above in Sections 5.2.3.1.2.2 and 5.2.3.1.2.3, the OSFs shown in this section are
linear with blood concentration. The OSFs shown in Tables 5-14 and 5-15 correctly depict the
cancer risks generated from the multistage models for their specific associated BMDLreds.
However, ingested TCDD doses are not linear with the predicted TCDD blood concentrations
generated by the Emond pharmacokinetic model. Thus, the OSFs associated with the ingested
doses are not linear with ingested dose. If the OSFs derived in this section were to be used in
human health risk assessment, target risk levels (e.g., 1 x 10 3, 1 x 10 4, 1 x 10 5, 1 x 10 6, etc.)
would need to be identified. Then, the Emond et al. pharmacokinetic model could be used to
generate risk-specific doses of TCDD intake (ng/kg-day) corresponding to the target risk levels.
5.2.3.2.6. Qualitative uncertainties in slope factor estimates from rodent bioassays.
This section presents a qualitative discussion of the uncertainties associated with
calculating the OSF for TCDD from chronic animal bioassay data. Discussions on the feasibility
of conducting a quantitative uncertainty analysis for TCDD using dose-response methods are
provided in Section 6.4.2 of this document.
5.2.3.2.6.1. Quality of studies relied upon for determining POD.
EPA considers the overall quality and breadth of the studies used for the cancer dose-
response analysis to be excellent. All of the studies were published in the peer-reviewed
literature, and two of them were conducted by NTP (1982, 2006). Kociba et al. (1978) and Toth
et al. (1979) are older studies, but appear to have been conducted according to good laboratory
practice standards. The control and dose group sample sizes were relatively large, -50 animals
or more per group for all of the studies except for Toth et al. (1979), where the dose group sizes
were -40 animals per group. All four studies exposed the test animals via the oral route to
TCDD alone, as was stipulated in EPA's study inclusion criteria. Collectively, these four studies
reported development of numerous cancer endpoints (tumors) in both sexes in two strains of rats
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-40 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
(Sprague-Dawley and Osborne-Mendel) and two strains of mice (i.e., B6C3Fi, Swiss/H/Riop).
The overall high quality of these studies and the strong, positive association between TCDD
exposure and cancer suggests that study quality is not a major contributing factor to uncertainty
in the risk estimates.
5.2.3.2.6.2. Interpretation of results from studies relied upon for determining POD.
As discussed in Section 3.4.3.2.1, questions arose about the interpretation of liver tumor
responses in female rats in the Kociba et al. (1978) study. Three re-evaluations of the slides have
been reported (Kociba et al., 1978; Squire, 1980; Goodman and Sauer, 1992). The decision to
use the Goodman and Sauer (1992) evaluation was based on the fact that the authors used the
most current tumor classification procedures. The incidence of hepatocellular adenomas and
carcinomas (individually and combined), however, did vary (sometimes widely) for each dose
group across the three evaluations. Although the state-of-the-science is reflected in the incidence
data used for analysis, a small degree of additional uncertainty (i.e., above that associated with
data collection and reporting errors) could be associated with the female liver tumor risk
estimates from the Kociba et al. (1978) study due to this variability. No controversy has arisen
with regard to the interpretation of the NTP (1982, 2006), or Toth et al. (1979) tumor
identification and classification.
5.2.3.2.6.3. Consistency of results across chronic rodent bioassays.
The existence of four high-quality chronic bioassays for TCDD increases confidence and
reduces uncertainty in the cancer OSFs. Considered together, these studies tested two species
and both sexes of mice and rats, and a wide range of well-characterized tumor types. All four
studies were consistent in observing increases (at some dose level) in rates of liver tumors (in
both species and sexes). While tumors at other sites were observed (and those sites varied across
study, species, and sex), the liver tumors were consistently the most sensitive indicators of
carcinogenic response (with respect to BMDLred estimates). Lung tumors were also
consistently observed across three of the studies, in male mice in the NTP (1982) study and in
female rats in Kociba et al. (1978) and NTP (2006). As discussed above, the two most sensitive
single-tumor endpoints as judged by BMDL0i values were associated with elevated liver tumor
risks, followed by lung, lymphoma or leukemia, thyroid and adrenal cancers. The consistency of
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-41 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
tumor types and sensitivities across endpoints and studies lends confidence to the multistage
modeling results.
5.2.3.2.6.4. Human relevance of rodent tumor data.
There is some concordance in the tumor responses observed in the rodent test species and
humans, however, the most sensitive tumor site in the animals, the liver, has not been associated
with cancer from TCDD exposures in humans. On the other hand, lung cancer and leukemia
have been are found both in the animal studies and in epidemiologic studies of exposed workers.
The consistency across sex, species and strains in the animal studies suggests that the occurrence
of several of these tumors, in particular, liver and lung tumors is not an idiosyncratic response of
a particular combination of species, strain, or sex. As discussed in Section 5.2.1, the likely AhR
related carcinogenic mechanism is credible for humans as well as for rodent species.
5.2.3.2.6.5. Relevance of rodent exposure scenario.
The four chronic rodent bioassays exposed the test animals for ~2 years, the majority of
their lifespans. The exception is the Toth et al. (1979) study, where the animals were exposed
only for one year, but were kept on the study for a second year before they were evaluated for
cancer. These lifetime bioassays in animals have long been used by EPA to assess potential
lifetime exposures and effects in humans. However, in the case of TCDD, the half life of TCDD
in the body for rats, mice, and humans is very different (see Section 3). Thus, there is a
significant amount of uncertainty in the use of rat and mouse data to develop OSFs for human
cancer risk assessment of TCDD.
5.2.3.2.6.6. Impact of background TCDD exposures.
It is known that TCDD has been found in the feed used in animal bioassays, and that this
is a confounding factor, particularly in older studies. The effect of TCDD in the diets of test
species has the potential to be quite significant given the low levels of TCDD at which adverse
effects have been observed. Insofar as that is an issue, the risks associated with TCDD
exposures in the animal bioassays, and therefore the OSFs, would be biased high, which could be
the case for the NTP (1982), Kociba et al. (1978) and Toth et al. (1979) studies. The impact of
this issue is that the newer study, NTP (2006), accounted for TCDD exposures in the animal
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-42 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
feed. Thus, there is likely to be less uncertainty in the TCDD dose-response information
presented in NTP (2006) than in the other three studies conducted before 1990.
5.2.3.2.6.7. Choice of endpoint for POD derivation.
As noted above, the liver tumor PODs represent the most sensitive single-tumor endpoint
across the four cancer bioassays. Thus, the liver cancer endpoints must be seriously considered
for derivation of a TCDD OSF. As discussed in the previous section, EPA has also developed
Bayesian dose-response estimates for combined tumors, which yield BMDLoi values slightly
lower than those for any individual tumor type. Although it is the most conservative choice to
select the lowest combined tumor POD for OSF derivation, there are uncertainties associated
with the multiple tumor analysis. The assumption of independence of tumors across sites is
reasonable, particularly since the tumors from TCDD do not metastasize. However, the
independence assumption lacks hard evidence and needs further laboratory confirmation.
5.2.3.2.6.8. Choice of animal-to-human extrapolation method.
The analyses presented here have used the Emond human kinetic model for extrapolating
dose from animals to humans (as discussed in Section 3.4.2). The rationale for this choice is that
the blood concentration metric most accurately reflects the concentration of TCDD in the various
tissues. As discussed in Section 3.4.3.2.4, use of the blood concentration dose metric results in
critical dose estimates (HEDs) that are considerably lower (10- to more than 100-fold) than those
derived based on administered dose. This does not reflect bias in the blood-based measure;
rather it is a reflection of the highly nonlinear biokinetics of TCDD in the body. EPA has also
explored the impacts of using other dose metrics, including liver bound TCDD concentration
calculated based on the Emond model. As discussed in Section 3.4.3.2.6.2, this also results in
HED estimates much lower than those obtained based on administered dose.
5.2.3.2.6.9. Choice of model for POD and model uncertainty for POD derivation.
The bioassay-based cancer dose-response assessment in this section has used the
multistage model which is the standard model choice for such assessments and has been the basis
for most of EPA's cancer risk assessments. In that sense, there is no associated uncertainty for
model choice.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-43 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
There is some model choice uncertainty associated with instances of lack of fit. When
the multistage model does not adequately describe the observed pattern of responses (typically
determined by examining thep-walue for lack of fit), a decision must be made about possible
adjustments, including the dropping of higher dose groups thought to be less relevant to the
estimation of low-dose slopes. In this analysis, poorer fits (/> values less than 0.10) were
observed in four cases, all from NTP (1982) (see Table 5-13). The lowest BMDLoi associated a
low p-w alue (p = 0.09) was for the lung tumors in the NTP (1982) male mouse, the third lowest
POD behind the liver PODs in the individual tumor data sets. The other instances were for
adrenal cortex and thyroid follicular cell adenomas in male rats and for subcutaneous tissue in
female mice in the NTP (1982) study. In those instances, the ^-values were 0.06, 0.06, and 0.09,
respectively; the fits were considered adequate for describing the low-dose response patterns and
estimating slope factors. These poorly fit data sets contribute to uncertainty in the combined
tumor PODs. The lowest BMDLoi in the combined tumors is for the male mice combined liver
and lung tumors, thus estimates from this sex/species combination from NTP (1982) is highly
uncertain and impacts its choice as a POD.
5.2.3.2.6.10.Statistical uncertainty in model fits.
Every model fit to a data set is associated with some inherent statistical uncertainty. For
this reason, bounds were calculated and used for OSF derivation (e.g., lower bounds on
benchmark doses, in this case the BMDLois). Those bounds account for uncertainties associated
with finite samples of test animals, both in terms of the number of dose groups and of the
number of animals per dose group. Valid and accepted statistical procedures have been applied
to ascertain the impact of those limitations on the estimates of interest. That being the case, the
statistical uncertainties associated with finite samples have been adequately addressed.
5.2.3.2.6.11. Choice of risk level for POD derivation.
The BMR level that has been used for the POD in deriving the cancer OSF is one percent
extra risk. This choice is supported by the fact that risks need to be low enough so that the linear
extrapolation from them reflects low-dose behavior, reducing the impact of any higher-dose
changes (inflexions) in the dose-response. The multistage model has the characteristic that it can
be approximated by a linear model at low enough doses. Thus, a risk level in the range of one
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-44 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
percent is generally suitable for characterizing that approximate linear function, while not falling
too far below the observed range of the data.
5.2.3.3. EPA 'v Response to the NAS Comments on Choice of Response Level and
Characterization of the Statistical Confidence Around Low Dose Model Predictions
The NAS was concerned with the statistical power to determine the shape of the dose
response curve at low doses, well below observed dose-response information. EPA shares this
concern in that the shape of the dose-response curve in the low-dose region cannot be determined
with confidence when based on higher dose information.
When tumor data are used for dose-response modeling, a POD is obtained from the
modeled tumor incidences. When assessing carcinogenicity using a linear extrapolation
approach from a POD, a balance must be struck between staying within the range of the
observations and obtaining a representative estimate of the low-dose slope. Traditional cancer
bioassays, with approximately 50 animals per group, can typically support modeling down to an
increased incidence of 1-10%; epidemiologic studies, with larger sample sizes, below 1%. For
the TCDD animal cancer bioassay data, most of the low-dose tumor incidence responses are
under 10% (relative to controls), with some as low as 2%. After evaluating the magnitude of the
uncertainty in BMDLoiS against the impact of using BMDLi0s, EPA has chosen to use a 1%
BMR in all cases, determining that the uncertainty bounds on the BMDL0i values are reasonable
(see Appendix F for details). For comparison purposes, BMDLoiS, BMDL05S, and BMDLi0s are
presented for all modeled tumor incidences in Appendix F.
In the analysis of the animal cancer bioassays presented in this document, the multistage
cancer model was applied with a linear dose extrapolation to zero. EPA used a 1% excess risk
estimate, i.e., a BMDLoi, as the POD for development of candidate TCDD cancer oral slope
factors using a Bayesian multitumor approach (see Section 5.2.3.2. The advantage of a Bayesian
approach is that it produces a distribution of BMDs that allows better characterization of
statistical uncertainty.
Central tendency and lower bound slope factor estimates are part of the standard BMDS
multistage cancer model and are included in each output file for the animal bioassay single tumor
analyses in Appendix F. Central tendency BMDs are also reported for the results of the animal
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-45 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
bioassay multitumor analysis (see Table 5-13). Central tendency slope factor estimates are given
for all the qualifying epidemiological studies as well (see Tables 5-1 and 5-4).
5.2.3.4. EPA 'v Response to the NAS Comments on Model Forms for Predicting Cancer Risks
Below the POD
The NAS offered extensive comments on the cancer dose-response modeling in the 2003
Reassessment. Although epidemiologic and rodent bioassay data are useful for the evaluation of
the dose-response curve within the range of the observed response data, they have traditionally
not been useful sources of information for identifying a threshold or for estimating the shape of
the dose-response curve below the POD. Rather, mechanistic toxicological data have been the
evidentiary sources of choice for those types of analyses. As noted above, any quantitative
estimation of carcinogenic risk associated with TCDD exposure requires low-dose extrapolation
of experimental data. Unfortunately, the shape of the dose-response curve in the low dose region
is unknown.
Several of the analyses of epidemiological cohort data evaluated the fit of different dose-
response models to the data. Log-dose models accentuate the importance of low-dose low-
magnitude responses and can yield implausible results. The most relevant models used in these
studies are the untransformed-dose Cox regression models, which are the most similar to the
multistage (1-degree) model used for animal bioassay analysis. Better results have been obtained
in the cohort analyses when the flattening of the hazard-ratio curve is taken into account. The
latter has been modeled explicitly by Steenland et al. (2001), who use a piecewise linear model
and implicitly by Cheng et al. (2006), who drop out a percentage of the high-dose response data
and fit a linear model to the remainder. Importantly, the analyses of the epidemiologic cohorts
presented in Section 5.2.3.1 are limited to evaluation and reanalyses of published data as
reported by the study authors. EPA does not have access to the raw data from these
epidemiologic studies and, therefore, could not conduct de novo analyses.
5.2.3.4.1. Choice of extrapolation approach
5.2.3.4.1.1. TCDD and receptor theory.
TCDD is considered to be a receptor-mediated carcinogen in animals. Nearly all TCDD
experimental data are consistent with the hypothesis that the binding of TCDD to the AhR is the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-46 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
first step in a series of biochemical, cellular, and tissue changes that ultimately lead to toxic
responses observed in both experimental animals and humans (Part II, Chapter 2 of the 2003
Reassessment). Ligand-receptor binding, like any bimolecular interaction, obeys the law of mass
action as originally formulated by A.J. Clark (Limbird, 1996). The law of mass action predicts
the fractional receptor occupancy at equilibrium as a function of ligand concentration. Fractional
occupancy (Y) is defined as the fraction of all receptors that are bound to ligand:
[TCDD - AhR ] [TCDD - AhR ] [TCDD ]
[AhR ]TOT ~ [AhR ] + [TCDD - AhR ] ~ [TCDD ] + Kd (Eq 5_5)
where [TCDD] is the concentration of the ligand, [AhR] is the concentration of the receptor and
[TCDD-AhR] is the amount of liganded receptor. The equilibrium dissociation constant Kd
describes the affinity of the interaction and is the concentration of TCDD that results in 50%
receptor occupancy. This simple equation defines a rectangular hyperbola, which is the
characteristic shape of the vast majority of biological dose-response relationships.
In certain cases, no response occurs even when there is some receptor occupancy. This
suggests that there may be a threshold phenomenon that reflects the biological "inertia" of the
response (Ariens et al., 1960). In other cases, a maximal response occurs well before all
receptors are occupied, a phenomenon that reflects receptor "reserve" (Stephenson, 1956).
Therefore, the law of mass action cannot by itself fully explain the effect or response observed
after TCDD interacts with AhR. The ligand-receptor complex is associated with a signal
transduction or effector system. In the case of the AhR, this effector system can be considered to
be the transcriptional machinery itself. The key feature of this formulation is that a response is
proportional, or a function of, the number of receptors occupied.
Furthermore, for a ligand such as TCDD that elicits multiple receptor-mediated effects,
one cannot assume that the binding-response relationship for a simple effect (such as enzyme
induction) will necessarily be identical to that for a different and more complex effect (such as
cancer). The cellular cascades of events leading to different complex responses (e.g., altered
immune function, developmental effects, or cancer) are different, and other rate-limiting events
likely influence the final biological outcome resulting in different dose-response curves. Thus,
even though TCDD binding to AhR is assumed to be the initial event leading to a spectrum of
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-47 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
biological responses, TCDD-AhR binding data may not always correlate with the dose-response
relationship observed for particular effects.
A receptor-based mechanism would predict that, except in cases where the concentration
of TCDD is already high (i.e., [TCDD]~Kd), incremental exposure to TCDD will lead to some
increase in the fractional occupancy of AhR. However, as discussed above, it cannot be assumed
that an increase in receptor occupancy will necessarily elicit a proportional increase in all
biological response(s), because numerous molecular events contributing to the biological
endpoint are integrated into the overall response. That is, the final biological response could be
considered as an integration of a series of interdependent dose-response curves with each curve
dependent on the molecular dosimetry for each particular step. Dose-response relationships that
will be specific for each endpoint must be considered when using mathematical models to
estimate the risk associated with exposure to TCDD. It remains a challenge to develop models
that incorporate all the complexities associated with each biological response as the modes of
action for various toxicological endpoints appear to vary greatly. For TCDD, extensive
experimental data from studies using animal and human tissues indicate that cell- or tissue-
specific factors determine the quantitative relationship between receptor occupancy and the
ultimate biological response. This would suggest that the parameters for each mathematical
model might only apply to a single biological response within a given tissue and species, making
extrapolation to other systems challenging.
5.2.3.4.1.2. Low-dose extrapolation: threshold or no threshold?
As indicated in the 2005 Cancer Guidelines,30 toxicity reference values for human
noncancer endpoints have historically been estimated based on a no-observed-adverse-effect
level (NOAEL) or lowest-observed-adverse-effect level (LOAEL) from animal bioassay studies.
This terminology suggests a biological population threshold beneath which no harm is
anticipated. Reference values such as the oral reference dose (RfD) or inhalation reference
concentration (RfC) are derived by applying uncertainty factors (UFs) to a POD. Depending on
the nature of available data and modeling choice, a POD can be selected from values other than
30As stated in the 2005 Cancer Guidelines (U.S. EPA, 2005): "For effects other than cancer, reference values have
been described as being based on the assumption of biological thresholds. The Agency's more current guidelines
for these effects (U.S. EPA, 1996a, 1998b), however, do not use this assumption, citing the difficulty of empirically
distinguishing a true threshold from a dose-response curve that is nonlinear at low doses."
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
an NOAEL or LOAEL, such as an EDX, or a benchmark dose (BMD) or its BMDL. An RfD is
described as "likely to be without appreciable risk" but the probabilistic language has not as yet
been operationalized. There is no quantitative definition of "appreciable" and no mechanism to
compute risk as a function of dose, so as to ascertain that the risk is indeed not appreciable. The
risk at the RfD is not calculated, and it cannot be calculated within the current UF framework.
Instead, a hazard quotient is computed as the ratio of a given exposure to the RfD, or a margin of
exposure is estimated as the ratio of the POD to the human exposure level.
Cancer endpoints are predominantly thought to have no population biological threshold.
Although the terminology "threshold/nonthreshold" is still common in cancer dose-response
discussions, the 2005 Cancer Guidelines propose a different terminology, whereby "nonlinear
models" are those whose dose-response slope is zero at or above zero. In the natural language,
and indeed in data analysis, it is difficult to distinguish the following situations:
• The response approaches zero as dose goes to zero, versus
• The response slope goes to zero as dose goes to zero (nonlinear model).
This use of "nonlinear" is acknowledged to be idiosyncratic.31 The NAS review (NAS,
2006a) does not consistently apply the terminology from the 2005 Cancer Guidelines, nor does it
consistently distinguish the above two circumstances: ".. .the observed data are more consistent
with a sublinear response that approaches zero at low doses rather than a linear dose response"
(NAS, 2006a). The point of a nonlinear model in the sense of the 2005 Cancer Guidelines is that
the response slope approaches zero. Both linear and nonlinear responses approach zero at low
dose (in the absence of background). Since the terms "linear," "sublinear," and "nonlinear"
invite confusion in this context, the following terminology is used in this document:
31
From the 2005 Cancer Guidelines (U.S. EPA, 2005): "The term 'nonlinear' is used here in a narrower sense than
its usual meaning in the field of mathematical modeling. In these cancer guidelines, the term 'nonlinear' refers to
threshold models (which show no response over a range of low doses that include zero) and some nonthreshold
models (e.g., a quadratic model, which shows some response at all doses above zero). In these cancer guidelines, a
nonlinear model is one whose slope is zero at (and perhaps above) a dose of zero Use of nonlinear approaches
does not imply a biological threshold dose below which the response is zero."
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-49 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Threshold Model. There is some threshold T > 0 such that the probability of response for
any dose less than or equal to T is zero, and the probability is nonzero for any dose
greater than T.
Linear/ Linear above Threshold Model. For the linear model, the probability of response
is proportional to the dose. For the linear over threshold model, the probability of
response is zero for a dose below the threshold, and it is proportional to the excess dose
over the threshold otherwise.
Nonlinear Model. Any model that is not linear.
Supralinear/ Supralinear above Threshold Model. For the supralinear model, the slope of
the probability of response decreases as dose increases; in other words, the second
derivative of the response curve is negative. For the supralinear above threshold model,
the second derivative is negative above the threshold, and the response probability is zero
below the threshold.
Sublinear/Sublinear above Threshold Model. For the sublinear model, the slope of the
probability of response increases as dose increases; in other words, the second derivative
of the response curve is positive. For the sublinear above threshold model, the second
derivative is positive above the threshold, and the response probability is zero below the
threshold.
Zero Slope at Zero Model. The slope of the response curve is zero at or above dose zero.
All of these models may be understood in an individual or population sense. According
to the 2005 Cancer Guidelines, the trigger for applying the basic RfD methodology for cancer
endpoints is sufficient evidence for the "zero slope at zero" model for the population. By
definition, any sublinear, supralinear, or linear model above the threshold is a zero slope at zero
("ZS@Z") model.
The relation between individual and population models is not simple. Suppose for
purposes of illustration that each individual has a threshold, and that the threshold values are
uniformly distributed in the population, for some neighborhood of zero. (Note that "in the
neighborhood of zero" has the mathematical meaning that all values are sufficiently close to
zero; i.e., for some 8 >0, the thresholds are uniformly distributed over the interval [0,5].)
Then (see Text Box 5-1)
A. If each person's response function is linear for a dose above the threshold, then the
population dose-response curve is quadratic, and the slope of the population curve is zero
for dose zero (ZS@Z).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-50 DRAFT—DO NOT CITE OR QUOTE
-------
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
49
50
51
52
53
54
55
56
57
B. If each person's response function is log
linear, for a dose above the threshold,
then the population dose-response curve
is linear without a threshold, and the
slope of the population curve is positive
for dose zero.
Of course these are not the only
possibilities; in general, the population dose-
response curve depends on (1) the distribution of
individual thresholds in the neighborhood of
zero, (2) the dose-response curve for each
individual, and (3) the dose metric.
On the nature or the distribution of
individual thresholds, often referred to as the
population tolerance distribution, there is
ongoing debate as to how receptor kinetics
influence the shape of that distribution. Even
within an individual, there is a lack of consensus
as to whether receptor kinetics confer linear or
sublinear attributes to downstream events, or
whether receptor kinetics, themselves, are linear,
sublinear, or supralinear. Whatever the nature of
the form of receptor kinetics, it may have little or
no influence on the ultimate population response. The kinetics of receptors is in the domain of
the individual, rather than the population. As described previously, receptor kinetics are
governed by the law of mass action, which leads to a low-dose proportional (linear) response
model, generally modeled by some form of Hill function, the linear form being Michaelis-
Menten kinetics. Linearity can be violated by a number of inhibitory or stimulatory processes,
but is a fundamental conserved characteristic in living systems. There is no a priori reason to
believe that the shape of the dose-response curve in an individual has any relationship to the
shape of the population response, particularly for quantal endpoints. Lutz and Gaylor (2008)
present a strong argument for considering the population response in terms of the more
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-51 DRAFT—DO NOT CITE OR QUOTE
Text Box 5-1. Individual vs. Population
Thresholds
Suppose each individual has a threshold T. below
dose T the probability of response is zero, and
suppose that the values for T are uniformly
distributed in some neighborhood of zero.
A. Suppose that for dose 8 > T, the probability of
response is proportional to (8 - T). For dose 8
small enough to be in this neighborhood, the
population response probability is proportional
to:
It ei o,s i (S~ T)dT =
Sire(o,8) dT ~ k(0,5l TdT= 82 - 82/2 = S2/2.
The slope of the response function is found by
taking the derivative with respect to 8:
{(11(16) S2/2 = 8, which goes to zero as 6 0.
B. Suppose now that the response probability at
dose 8 is proportional to ln(S) - ln(7). all else
being the same. The population response
probability at dose 8 is:
J*Te(0,8) (ln(S)- ln(7)) dT.
To evaluate this, note that
lTe(o,8) ln(7) dT=8 ln(S) - 8.
Hence the integral is:
8 ln(S) - [ 8 ln(S) - 8] = 8.
The derivative of this with respect to 8 is 1,
which goes to 1 as 8 0.
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
traditional tolerance distribution, which is likely the result of more variable factors than the
shape of receptor kinetics. Perhaps more to the point, receptor activation is only the first of
many events in the path to the apical event (a tumor in this example). Because there are
undoubtedly numerous additional downstream events that must occur before the apical effect is
observed, there are many opportunities for interindividual variability to become manifest in the
tolerance distribution. Even at the first step, a more likely contributor to interindividual
variability than the shape of the response is the dose resulting in the response, as measured by the
ED50 (the aforementioned equilibrium dissociation constant^, Km in the Michaelis-Menten
formulation), which shifts the response curve. Factors that influence shifts in response curves
are generally modeled as normal or log-normal distributions and will most likely confer a log-
normal shape on the population tolerance distribution, particularly if there are a number of
dependent sequential steps or distinct subpopulations (Lutz, 1999; Hattis et al., 1999; Hattis and
Burmaster, 1994).
Pertaining to the shape of the dose-response curve, because the criterion for applying the
RfD model is a zero slope at dose zero, the role of dose metric must be fully appreciated. A
slope is simply the ratio of a change on the vertical axis (the probability of response) relative to a
change on the horizontal axis (the dose). Changing the dose metric from the dose to the
logarithm of the dose dramatically changes this ratio. As dose goes to zero, the rate of change of
log(dose) becomes infinite. Therefore, ANY dose-response relation with a finite slope at zero
will appear to have a zero slope when graphed against log(dose). Text Box 5-2 illustrates this
point for the mass action dose response, which has been proposed for receptor-mediated modes
of action.
To see how the discussion over threshold/nonthreshold might play out for TCDD,
consider the equilibrium dissociation constant Kd for TCDD, which measures the binding affinity
of TCDD to the AhR. Lower values indicate higher binding affinity and (other things being
equal) greater risk. For Han/Wistar rats, the value Kd = 3.9 is reported; human values are
reported as Kd =9.6 ±7.8(0.3 - 38.8 with 15 of 67 donors without detectable binding) (Connor
and Aylward, 2006).
If AhR binding is necessary for carcinogenesis, then the majority of a human population
may be much less susceptible than Han/Wistar rats, whereas a population threshold, if it exists,
might be well below the Han/Wistar rat threshold (see Section 4.4.2.9). The NAS contends that
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-52 DRAFT—DO NOT CITE OR QUOTE
-------
Text Box 5-2. Logarithmic Transform of Mass Action Hyperbolic Model
According to the operational model (Black and Leff, 1983; Motulsky and Christopoulos, 2003), the fractional
occupancy of receptors by ligands follows a mass action law; the concentration of ligand-occupied receptors
[LR] is equal to the product of the concentration of receptors [R] and the concentration of ligands [L], divided
by the ligand concentration plus the binding affinity. This is a hyperbolic function written as [LR] = [R] [L]/([L]
+ Kd).
The efficacy of a ligand can vary by tissue and endpoint;
this is usually expressed as a hyperbolic function of
[LR]; (TOP-BOTTOM) [LR]/([LR] + KE) where KE is the concentration of [LR] producing the effect halfway
between TOP and BOTTOM. Combining these expressions gives:
[L](TOP-BOTTOM) (x/(x+l))
Response([L]) = ,
[L] + KJ(z+l)
where x is the tissue-/endpoint-specific 'transducer constant' [R]/KE. Writing 8 for dose [L], this is proportional
to the hyperbolic equation f(8) = 8/(8 + K): K = KJ(t +1).
Switching the dose metric to 8* = log(S),
f(8) becomes f(exp(8*)) = exp(8*)/[exp(8*) + £].
f(8) and f(8*) are graphed below, for K = 1.
f(8) f(exp(8*))
"1 / • _
Although these represent the same dose-response relation, the graph of f(exp(8*)) has zero slope as dose goes to
zero (8* goes to - <). In fact, df/dS* = (df/dS)(dS/dS*), and dS/dS* = e5* ->¦ 0 as 8* ->¦ -®. The change in
response relative to the change in log dose becomes infinitesimal, since log dose changes infinitely fast as dose
goes to zero. f(8*) is not a zero slope at zero model.
an AhR-mediated mode of action indicates a threshold dose-response relation (NAS, 2006a).
Presumably, the value of the threshold, if it exists, depends on the AhR binding affinity.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-53 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Arguing for a population threshold in this case requires two types of information:
1. The distribution of the individual thresholds induced by, inter alia (among other things),
the individual Rvalues; and
2. The dose-response function for values above the threshold induced by Kd.
Without this information, the two possibilities {A,B} enumerated on Pages 5-50 through
5-51 cannot be distinguished, and the default linear relationship applies response probability is
modeled as a linear function of dose, for dose near zero. However, from the 2005 Cancer
Guidelines: "When adequate data on mode of action provide sufficient evidence to support a
nonlinear mode of action for the general population (emphasis added) and/or any subpopulations
of concern, a different approach—a reference dose/reference concentration that assumes that
nonlinearity—is used." In current terminology, the reference dose methodology applies if there
is sufficient evidence supporting a "zero slope at zero" model; otherwise, the linear nonthreshold
model applies by default.
In principle, the choice between the above models could fall within the purview of dose-
response modeling. However, standard statistical methods encounter well-known difficulties in
detecting thresholds. Without going into detail, suffice to say that the maximum likelihood
estimate of response probability when no responses are observed in a finite sample is always
zero. That said, some researchers have attempted to identify thresholds (Mackie et al., 2002;
Alyward et al., 2003) or nonlinearity (Hoel and Portier, 1994) by means of parameter estimation
of appropriate models. A review of 344 rodent bioassays on 315 chemicals led to the following
conclusion by Hoel and Portier (1994):
We have also found that the oft-held belief that genotoxic compounds typically
follow a linear dose-response pattern and that nongenotoxic compounds follow a
nonlinear or threshold dose response pattern is not supported by the data. In fact
we find the opposite with genotoxic compounds differing from linearity more
often than nongenotoxic compounds.
The choice between a linear and "zero slope at zero" model in current practice does not
fall under dose-response model fitting, it is made on the basis of a structured narrative as set
forth in the 2005 Cancer Guidelines (U.S. EPA, 2005):
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-54 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
In the absence of sufficiently, scientifically justifiable mode of action information,
EPA generally takes public health-protective, default positions regarding the
interpretation of toxicologic and epidemiologic data: animal tumor findings are
judged to be relevant to humans, and cancer risks are assumed to conform with
low dose linearity. ... The linear approach is used when: (1) there is an absence of
sufficient information on modes of action or (2) the mode of action information
indicates that the dose-response curve at low dose is or is expected to be linear.
Where alternative approaches have significant biological support, and no
scientific consensus favors a single approach, an assessment may present results
using alternative approaches. A nonlinear approach can be used to develop a
reference dose or a reference concentration.
5.2.3.4.1.3. Extrapolation method.
The 2005 Cancer Guidelines (U.S. EPA, 2005) emphasize that the method used to
characterize and quantify cancer risk from a chemical is determined by what is known about the
MOA of the carcinogen and the shape of the cancer dose-response curve.
The NAS was critical of EPA's decision to apply linear low-dose extrapolation for
TCDD cancer assessment in the 2003 Reassessment and encouraged the use of a nonlinear
approach. The 2005 Cancer Guidelines state that a nonlinear approach should be used when
"there are sufficient data to ascertain the mode of action and conclude that it is not linear at low
doses and the agent does not demonstrate mutagenic or other activity consistent with linearity at
low doses."
Receptor modeling theory (as outlined in the 2003 Reassessment, Part II, Chapter 8)
indicates that exogenous compounds which operate through receptor binding mechanisms, such
as TCDD, will follow a linear dose-response binding in the 1-10% receptor occupancy region.
This theory has been supported by empirical findings and suggests that the proximal biochemical
effects (such as enzyme induction) and transcriptional reactions for TCDD may also follow
linear dose-response kinetics. More distal toxic effects could take any one of multiple forms
(i.e., linear, sublinear, supralinear or threshold) depending on (1) the toxic mechanism;
(2) location on the dose-response curve; and (3) interactions with other processes such as
intracellular protein binding and cofactor induction/repression.
In the case of TCDD, many adverse effects experienced at low exposure levels have too
much data variability to distinguish on a statistical basis (goodness-of-fit) between dose-response
curve options, and whether the dose-response is linear, sublinear or supralinear. For tumor
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-55 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
responses, with the exception of squamous cell carcinoma of the oral mucosa and adenomas or
carcinomas of the pancreas, which were fit with a linear multistage model, the tumor endpoints
in the NTP (2006) study using female Sprague-Dawley (S-D) rats are all best fit with a sublinear
model (i.e., the multistage model fits to tumor incidence data were second or third degree; see
Table 5-13 and Appendix F). For all other tumor incidence data from all other cancer bioassays
that met the study inclusion criteria (NTP, 1982, Toth et al., 1979, and Kociba et al., 1978), the
multistage model fit was linear (first degree), when based on either administered dose or
modeled blood concentrations (see Appendix F).
Another issue of potential importance when evaluating the shape of the dose-response
curve for low dose effects is the concept of "interacting background." The concept of interacting
background refers to a pathological process in the exposed population that shares a causal
intermediate with the toxicant being evaluated. On this issue, a recent NAS committee (NAS,
2009) contended that
.. .the current EPA practice of determining "nonlinear" MO As does not account
for mechanistic factors that can create linearity at low dose. The dose-response
relationship can be linear at a low dose when an exposure contributes to an
existing disease process (Crump et al.1976, Lutz 1990). Effects of exposures that
add to background processes and background endogenous and exogenous
exposures can lack a threshold if a baseline level of dysfunction occurs without
the toxicant and the toxicant adds to or augments the background process. Thus,
even small doses may have a relevant biologic effect. That may be difficult to
measure because of background noise in the system but may be addressed through
dose-response modeling procedures. Human variability with respect to the
individual thresholds for a nongenotoxic cancer mechanism can result in linear
dose-response relationships in the population (Lutz 2001) (NAS, 2009, p. 130).
AhR activation could be considered a causal intermediate in several disease processes.
Recent studies have linked AhR activation in the absence of exogenous ligand to a multitude of
biological effects, ranging from control of mammary tumorigenesis to regulation of
autoimmunity (reviewed in Hahn et al., 2009). While the level of background activation of AhR
by endogenous compounds (and/or exogenous compounds other than TCDD) in the human
population is unknown, given the ubiquitous nature of several of the known endogenous and
exogenous AhR ligands, it is reasonable to assume that a certain baseline level of AhR activation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-56 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
exists in the population. The degree to which TCDD exposure augments this baseline level of
AhR activation is unknown.
The 2005 Cancer Guidelines (U.S. EPA, 2005) recommend that the method used to
characterize and quantify cancer risk from a chemical be determined by what is known about the
mode of action of the compound and the shape of the cancer dose-response curve. The linear
approach is used if the mode of action is not understood (U.S. EPA, 2005). In the case of
TCDD, (1) the mode of action of TCDD-induced carcinogenesis beyond potential AhR
activation is unknown; (2) information is lacking to determine confidently the shape of the dose-
response curves for various adverse endpoints (including cancer) in humans or experimental
animals; (3) there is undoubtedly a certain level of interacting background (i.e., AhR activation
by endogenous ligands) in the human population; (4) many of the rodent cancer dose-response
relationships (NTP, 1982; Toth et al., 1979; Kociba et al., 1978) are linear (first degree
multistage model fit) when based on either administered dose or modeled blood concentrations;
and (5) higher human interindividual variability compared to experimental rodents will tend to
shift the shape of the dose-response towards linear (relative to rodents). Therefore, a linear low-
dose extrapolation approach was used to estimate human carcinogenic risk associated with
TCDD exposure.
5.2.3.4.1.4. Consideration of nonlinear methods.
While the 2005 Cancer Guidelines deem linear extrapolation to be most appropriate for
TCDD, EPA has carefully considered the NAS recommendation to provide risk estimates using
both linear and nonlinear methods.
The 2005 Cancer Guidelines state
For cases where the tumors arise through a nonlinear mode of action, an oral
reference dose or an inhalation reference concentration, or both, should be
developed in accordance with EPA's established practice for developing such
values ... This approach expands the past focus of such reference values
(previously reserved for effects other than cancer) to include carcinogenic effects
determined to have a nonlinear mode of action.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-57 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
In this section, EPA presents two illustrative examples of RfD development for
carcinogenic effects of TCDD. Each of these examples focuses on data derived from animal
bioassays as described in Section 2.4.2.
5.2.3.4.1.4.1. Illustrative RfDs based on tumorigenesis in experimental animals.
TCDD has been shown to be a multisite carcinogen in both sexes of several species of
experimental animals. It also has been shown to be carcinogenic to humans. Most of the
available quantitative human epidemiologic data related to TCDD carcinogenesis are for all
cancer mortality. Mortality is a frank effect and is generally considered to be inappropriate for
RfD development, therefore, the illustrative example below utilizes available evidence from
experimental animals. Table 5-16 presents candidate PODs and RfDs for TCDD carcinogenicity
based on combined tumor responses from the animal bioassays described in Section 2.4.2. The
PODs from the NTP (1982, 2006) and Kociba et al. (1978) animal studies were derived from
Bayesian multitumor dose-response modeling (as described in Section 5.2.3.2, Table 5-15) using
a BMR of 1%. Because only TCDD-induced liver tumors were reported by Toth et al. (1979),
the BMR of 1% (POD) from that study was generated using a first degree linear multistage
model (see Table 5-14). Following BMD modeling, BMDLhedS were then estimated using
alternative dose metrics from the Emond model as described in Section 3. The illustrative RfDs
were derived by dividing the BMDLhedS by appropriate uncertainty factors. In each instance, a
total UF of 30 was applied, comprising factors of 3 for the toxicodynamic component of the
interspecies extrapolation factor (UFA) and a factor of 10 for human interindividual variability
(UFh).
As shown in Table 5-16, the illustrative RfDs for TCDD-induced tumors range from
3.6E-11 for liver and lung tumors in male mice (NTP, 1982) to 1.0E-9 for adrenal cortex, tongue
and nasal/palate tumors in male rats (Kociba et al., 1978). This illustrative RfD range for TCDD
tumorigenesis falls within the range of candidate RfDs for noncancer TCDD effects presented in
Table 4-5.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-58 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
5.2.3.4.1.4.2. Illustrative RfDs based on hypothesized key events in TCDD 'sMOAs for liver
and lung tumors.
As described in Section 5.1, most evidence suggests that the majority of toxic effects of
TCDD are mediated by interaction with the AhR. EPA considers interaction with the AhR to be
a necessary, but not sufficient, event in TCDD carcinogenesis. The sequence of key events
following binding of TCDD to the AhR and that ultimately leads to the development of cancer is
unknown. While the mode of action of TCDD in producing cancer has not been elucidated for
any tumor type, the best characterized carcinogenic actions of TCDD are in rodent liver, lung,
and thyroid. The hypothesized sequence of events following TCDD interaction with the AhR is
markedly different for each of these three tumor types. Additionally, no detailed hypothesized
mode of action information exists for any of the other reported tumor types.
The endpoints selected for this illustration were evaluated to provide insight into the
quantitative relationships between tumor development and precursor events in TCDD-induced
carcinogenesis. The endpoints described below may or may not be biologically adverse in
themselves; the intent herein was to consider TCDD-induced biochemical and cellular changes
that could lead to subsequent tumor development.
In the following exercise, illustrative RfDs were derived for key events in TCDD's
hypothesized modes of action in the liver and lung. No appropriate dose-response data were
identified for key events in TCDD's hypothesized MOA for thyroid tumors in a
sex/species/strain that has been shown to develop thyroid tumors (i.e., female B6C3F1 mice and
male and female Osborne-Mendel rats [NTP, 1982]).
As this is an illustrative exercise only, only studies that were originally identified in
Section 2 for potential noncancer dose-response modeling were evaluated here (see Section 2.4.2
for study details). There may be additional studies available in the literature that would further
inform the dose-response assessment of these endpoints.
Additionally, for animal model consistency, only results from studies conducted in
female S-D rats are presented here. The majority of the available information on TCDD
carcinogenicity (and TCDD carcinogenic precursor events) comes from studies conducted in
female S-D rats and the most recent TCDD carcinogenicity study was conducted in female S-D
rats (NTP, 2006). While both Kociba et al. (1978) and NTP (2006) have conducted TCDD
carcinogenicity studies in female S-D rats, different substrains were used; this difference in
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-59 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
substrain may have resulted in the different carcinogenic responses reported from the two
studies. While the carcinogenicity of TCDD in female S-D rats has been well characterized, this
animal model does not exhibit the full suite of tumor responses reported for TCDD (for instance,
female S-D rats have not been shown to develop thyroid tumors). Additionally, the most
sensitive single tumor response in female S-D rats from NTP (2006) is squamous cell carcinoma
of the oral mucosa (see Section 5.2.3.2), a tumor type for which no mode of action information
exists. Therefore, the illustrative RfDs described below may not be protective against all tumor
types.
For each endpoint, PODs for illustrative cancer RfD development were identified as
described for the noncancer RfD derivation in Section 4. Briefly, for the endpoints identified
below, the NOAELredS and/or LOAELHedS were determined based on EPA analysis of the
original data presented by the study author (see Section 2.4.2 for details) and by application of
the Emond PBPK models as described in Section 3.3.4. BMDLreds were determined as
described in Section 4.2 for all data sets amenable to BMD modeling. Modeling outputs for the
endpoints are presented in Appendices E and G as noted in Table 5-17. The illustrative RfDs
were derived by dividing the POD by appropriate uncertainty factors as indicated in Table 5-17.
5.2.3.4.1.4.2.1. Liver tumors.
Figure 5-4 presents one hypothesized mode of action for TCDD-induced liver tumors in
rats. TCDD activation of the AhR leads to a variety of changes in gene expression, including
increased CYP1A1 mRNA and subsequent increases in CYP1A1 activity. These alterations in
gene expression are hypothesized to lead to hepatotoxicity, followed by compensatory
regenerative cellular proliferation and subsequent tumor development. The details of the
mechanism of TCDD-induced hepatotoxicity have not been fully determined but both CYP
induction and oxidative stress have been postulated to be involved (Maronpot et al., 1993;
Viluksela et al., 2000). Additionally, oxidative DNA damage has been implicated in liver tumor
promotion (Umemura et al., 1999). The enhanced cell proliferation arising from either altered
gene expression or hepatotoxicity, or both, could be the principal factor leading to promotion of
hepatocellular tumors (Whysner and Williams, 1996).
A dose-response relationship exists for TCDD-mediated hepatotoxicity, and this parallels
the dose-response relationship for tumor formation (or formation of foci of cellular alteration as a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-60 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
surrogate of tumor formation). However, the dose-response relationship for other
TCCD-induced responses such as enhanced gene expression is different from the dose-response
for tumor formation in terms of both efficacy and potency (see Popp et al. [2006] for review).
A representative endpoint for each of the hypothesized key events following AhR
activation for TCDD-induced liver tumors was identified and is shown in Figure 5-4. Illustrative
RfDs based on each representative endpoint are shown in Table 5-17.
5.2.3.4.1.4.2.2. Lung tumors.
Far less is known about TCDD's mode of action in the lung. Figure 5-5 presents two
hypothesized modes of action for TCDD-induced lung tumors in rats. The first hypothesized
mode of action of TCDD in the lung involves disruption of retinoid homeostasis in the liver.
Retinoic acids and their corresponding nuclear receptors, the RARs and the RXRs, work together
to regulate cell growth, differentiation, and apoptosis. It is hypothesized that TCDD, through
activation of the AhR, can affect parts of the complex retinoid system and/or other signaling
systems regulated by, and/or cross-talking with, the retinoid system (reviewed in Nilsson and
Hakansson, 2002). These effects are then hypothesized to lead to lung tumor development,
however the mechanisms underlying this hypothesis are not well-defined. The second
hypothesized mechanism for the carcinogenic action of TCDD in the lung is through induction of
metabolic enzymes. Through activation of AhR and subsequent induction of metabolizing
enzymes (such as CYP1A1), TCDD may enhance bioactivation of other carcinogens in lung
(Tritscher et al., 2000). However, there are few studies to support this hypothesis.
Representative endpoints could only be identified for two of the hypothesized key events
following AhR activation for TCDD-induced lung tumors. These endpoints are presented in
Figure 5-5. Illustrative RfDs based on each of these two representative endpoints are shown in
Table 5-17. There is insufficient information to form any conclusions on the quantitative
progression to tumorigenicity or on the relative protection afforded by preventing the key events
shown.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-61 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
5.2.3.4.1.4.2.3. Limitations of illustrative RfDs based on hypothesized key events in TCDD's
MO As for liver and lung tumors.
A trend for increasing RfD values that follows the progression of endpoints towards the
production of tumors is evident. However, there are a number of factors that prevent making
strong conclusions based on this exercise. These limitations include the following
• This example addresses only two tumor types in one species, strain and sex (female S-D
rats), with little information available on the hypothesized mode of action for lung
tumors. No mode of action information is available for the most sensitive tumor type in
this animal model (squamous cell carcinoma of the oral mucosa). Therefore, it is
possible that the illustrative RfDs presented in this example would not be protective
against all tumor types in female S-D rats. Importantly, other animal models have been
shown to be more sensitive to TCDD-induced carcinogenesis based on combined tumor
analysis (see Section 5.2.3.2); an RfD based on tumorigenesis in this animal model may
not be protective against tumorigenesis in other, more sensitive, animal models (or, by
extension, in humans).
• Several of the BMDLs are based on poorly-fitting models, such that the RfD is based on
a LOAEL, which is not a particularly good measure for comparison across endpoints
(e.g., LOAELs are subject to influences of dose spacing in bioassays). Furthermore, the
hepatotoxicity BMDL based on a dichotomous 10% BMR, is not directly comparable to
all the other BMDLs based on a continuous 1 standard-deviation BMR (Crump, 2002).
• The endpoints selected as representative of each hypothesized key event may not be the
most appropriate choices. These particular endpoints were chosen because they were the
most sensitive indicator (i.e., lowest POD) from the available data or were the only
available choice based on a lack of data for other effects related to the hypothesized key
event.
• The optimum timing of these events may not be reflected in the endpoints selected.
Almost certainly, changes in gene expression are early events, such that a single
exposure should be relevant, as in the mRNA changes reported after a single TCDD
exposure (Vanden Heuvel et al., 1994), although it is not known whether the magnitude
of these changes would be altered after longer-term exposure, or whether longer-term
exposure would be more relevant to downstream events. Similarly, single exposures for
induction of CYP enzymes would seem to relevant as a measure of the immediate effect,
but it may be longer-term repeated CYP activity that is important for longer-term
downstream events; Table 5-17 shows a nominal order-of-magnitude difference in effect
levels for similar effect magnitudes (ca. 20-fold) from single exposures (Kitchin and
Woods, 1979) and long-term exposures (53-weeks; NTP, 2006). The relevant exposure
durations for oxidative stress and later effects are longer term, so a measurement of
oxidative stress at 90-days in a rodent may be appropriate; Wyde et al. (2001) suggest
that induction of 8-oxo-dG DNA adducts are a result of longer-term oxidative stress
because of the lack of effect of single exposures. Hepatotoxicity and hepatocellular
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-62 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
proliferation events would appear at successively later times, but the effective exposure
levels would depend heavily on the endpoints chosen to represent those events and the
time at which they were measured. The toxic hepatopathy endpoint reported in NTP
(2006), is a general measure of mild to moderate liver toxicity, but is measured only at
the end of the study when tumors have already appeared. Hepatocyte hypertrophy,
measured at 31 weeks may be more duration-relevant, but may not indicate actual
hepatocellular toxicity.
• The lowest of the tested doses may well be much higher, given that all animal diets are
contaminated to a certain extent by TCDD, resulting in initial TCDD body burdens in all
animals. Vanden Heuvel et al. (1994) reported TCDD liver concentrations in control
animals almost as high as for the low-dose group, which could equate to a significant
increase in the actual exposure experienced by the low-dose group. A similar effect on
the low-dose group (0.45 ng/kg) in Kitchin and Woods (1979) is possible, although they
did not report control animal tissue concentrations. Higher exposure levels or longer-
term exposures would not be affected to the same degree, as administered TCDD levels
would likely be large compared to initial body burden or low-level feed stock exposure.
Given the limitations described above, establishing an unambiguous progression of
effects is extremely problematic given the lack of sufficient data. Identifying a RfD that could
be considered to be protective against tumorigenesis in humans based on these data and models
is subject not only to the determination of effective low doses for the RfDs in Table 5-17 but also
to the determination of effective exposures that could be considered to be protective of all other
tumor types in female S-D rats as well as all other animal models. The latter would entail
identifying precursors that are sufficient in themselves for progression to tumorigenesis for all
tumor types. Given the disparate sequence of hypothesized key events following TCDD-induced
AhR activation for the tumor types for which some information is available, AhR
binding/activation is the only key event that is likely to be shared across tumor types. No
appropriate quantitative data on AhR binding/activation by TCDD in relevant animal models
were located; therefore, an illustrative RfD based on TCDD AhR activation could not be
developed.
Simon et al. (2009) present a similar analysis for the liver tumors observed in the NTP
(2006) study, showing a progression of effects from early biochemical events to irreversible liver
toxicity, culminating in tumorigenesis. While illustrative of the putative tumor-promoting MOA
for TCDD, the limitations of using such an approach within the context of an assessment of the
overall carcinogenic risk of TCDD as detailed above still apply. Simon and colleagues also
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-63 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
present RfDs for liver tumors and several precursor endpoints. All the RfDs presented in Simon
et al. (2009) are essentially equivalent and are 1 to 3 orders of magnitude higher than the RfDs
for equivalent endpoints presented in Table 5-17. These discrepancies are partly due to the fact
that the Emond PBPK models (Emond et al., 2004, 2005, 2006; see also Section 3.3.4) used in
this document predicts lower TCDD intakes for similar tissue concentrations than the CADM
kinetic model (Carrier et al., 1995a, b; Aylward et al., 2005b) used by Simon and colleagues.
However, a larger contributor to these discrepancies is the use of a chemical-specific adjustment
factor (CSAF) of 0.1 for the toxicodynamic component of the interspecies uncertainty factor by
Simon et al. (2009), while EPA used an uncertainty factor of 3. EPA does not find that the in
vitro evidence presented by Simon et al. in support of a CSAF of 0.1 for interspecies
toxicodynamics meets the burden of proof necessary for a reduction in this uncertainty factor.
5.2.3.4.1.4.3. Effect of dose metric on linearity of response.
EPA agrees that there is evidence for sublinearity in the TCDD carcinogenic response in
rats. The lung and liver tumor endpoints in the NTP (2006) study using female S-D rats are best
fit with a sublinear model. Except for squamous cell carcinoma of the oral mucosa and
combined carcinomas and adenomas of the pancreas, the multistage model fits to tumor
incidence data were second or third degree (see Table 5-13 and Appendix F). Figure 5-6 shows
the multistage model fit for cholangiocarcinomas as a representative example. For all other
tumor incidence data in all other cancer bioassays, the multistage model fit was linear (first
degree), when based on either administered dose or modeled blood concentrations (see
Appendix F). In further evaluating the combined liver tumor incidence for female S-D rats in
Kociba et al. (1978) (as re-evaluated by Goodman and Sauer, 1992; "Kociba/G&S data"), using
an unconstrained dichotomous Hill model, a progression from supralinearity to linearity to
sublinearity is obtained when fitting the model to the data based on progressively more relevant
dose metrics. Figure 5-7 shows the Hill model fits to the Kociba/G&S data based on
administered dose, modeled blood concentrations and modeled liver AhR-bound
concentrations.32 Details of the model fits are presented in Table 5-18. The Hill coefficient
parameter values are also given in the caption to Figure 5-7. A Hill coefficient of 1 indicates
32The modeled TCDD compartmental concentrations were obtained using the Emond rodent PBPK model described
in Section 3.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-64 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
linearity, while values less than 1 indicate supralinearity and values greater than 1 indicate
sublinearity. The TCDD blood concentration metric is used in this document as it was
determined to be the most relevant toxicokinetically-equivalent metric across species. However,
as discussed in Section 3, the AhR-bound TCDD concentration may be the most relevant metric
for liver effects, but this metric is also the most uncertain. Uncertainty aside, the tendency for
the model fits to move towards sublinearity with increasing relevance of dose metric suggests
that the "true" tumorigenic response may be sublinear in female S-D rat liver. The progression
towards sublinearity is likely to be due to the failure of the less-relevant metrics to reflect the
dose-dependent elimination of TCDD in the liver. The range of the administered doses in
Kociba et al. (1978) is much larger than that of the effective internal TCDD concentrations. This
"stretches out" the response relative to dose, resulting in a supralinear model fit. The blood-
concentration metric (as determined using the Emond PBPK model) partially accounts for dose-
dependent elimination, but does not account for the partition of effective (AhR-bound) and
ineffective (CYP1A2-bound) TCDD in the liver33, thus yielding only a partial correction to the
shape of the dose-response curve (see Section 3). Thus, hypothesizing that the "true" response
for liver tumors is sublinear in female S-D rats and assuming that the same mechanisms are
operant in humans, a zero-slope at zero dose relationship could be proposed for liver tumors,
allowing for an RfD approach for liver tumorigensis (as described above). This approach,
however, would not be considered to be protective against all tumors given the lack of
appropriate mode of action information for many of the known tumor types, particularly
squamous cell carcinoma of the oral mucosa in this animal model.
On balance, however, the evidence is only suggestive and the arguments are speculative.
Even if a sublinear response could be shown for rodent tumors, the greater interindividual
variability in the human population would tend to shift the response towards linear (Lutz, 1999;
Hattis, 1996). Hattis (1996), however, points out that, assuming a lognormal distribution of
susceptibilities, increasing the variance may result in a linear or supralinear shape for a large
portion of the response curve but never brings the relationship to a true low-dose linear form.
How close to zero the linearity extends and how much it encompasses the risks of concern
depends on the specifics of the particular human variability scenario.
33 Modeling the response against the whole liver concentrations results in a supralinear fit similar to administered
dose (Hill coefficient = 0.71).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-65 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
5.3. DERIVATION OF THE TCDD ORAL SLOPE FACTOR AND CANCER RISK
ESTIMATES
EPA was able to derive candidate OSFs for all cancer mortality from human
epidemiologic studies as well as for individual and combined tumor incidence from rodent
cancer bioassays. Each of these studies were selected for TCDD dose-response modeling using
the study inclusion criteria outlined in Section 2. The derivation of these OSFs can be found for
the epidemiologic data in Section 5.2.3.1 and for the rodent bioassay data in Section 5.2.3.2.
The OSFs based on epidemiologic studies from three cohorts ranged from 3.75 x io5 to
2.5 x 106 per mg/kg-day (see Tables 5-1 and 5-3). For the animal data, OSFs based on
individual tumors were developed for 25 study/sex/endpoint combinations, and the results ranged
from 1.8 x 104 to 5.9 x 106 per mg/kg-day (see Table 5-14). The OSFs based on combined
tumors were developed for 7 study/sex combinations, and the results ranged from 3.2 x 105 to
9.4 x io6 per mg/kg-day (see Table 5-15).
As recommended by expert panelists at EPA's 2009 Dioxin Workshop (U.S. EPA,
2009c) and in the 2005 Cancer Guidelines (U.S. EPA, 2005), EPA has chosen to give higher
consideration to the human epidemiologic data rather than the animal bioassay data in
developing an OSF for TCDD. Candidate OSFs derived from the human data are consistent with
the animal bioassay OSFs; specifically, the human OSFs fall within the same range as the animal
bioassay OSFs. Because all the human and animal studies were considered to be of high quality
and yielded similar ranges of OSFs, EPA has chosen to rely on the epidemiologic data for OSF
derivation.
The strengths and limitations of the five epidemiological studies meeting the inclusion
criteria for cancer dose-response modeling are summarized in Table 5-19. Among the human
studies, the occupational TCDD exposures in the NIOSH and Hamburg cohorts are assumed to
be reasonably constant over the duration of occupational exposure. In contrast, the TCDD
exposure patterns in the Seveso and BASF cohorts are associated with industrial accidents; as a
consequence, the exposure patterns are acute, high dose followed by low-level background
exposure. Such exposure patterns similar to those experienced by the BASF and Seveso cohorts
have been shown to yield higher estimates of risk when compared to constant exposure scenarios
with similar total exposure magnitudes (Kim et al., 2003; Murdoch and Krewski, 1988; Murdoch
et al., 1992). Thus, EPA has judged that the NIOSH and Hamburg cohort response data are more
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-66 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
relevant than the BASF and Seveso data for assessing cancer risks from continuous ambient
TCDD exposure in the general population.
The NIOSH (Steenland et al., 2001; Cheng et al., 2006) and Hamburg (Becher et al.,
1998) cohort studies report cumulative TCDD levels in the serum for cohort members. The most
significant difference among the Cheng et al. (2006) analysis and those of Steenland et al. (2001)
and Becher et al. (1998) is the method used to back-extrapolate exposure concentrations based
on serum TCDD measurements. Steenland et al. (2001) and Becher et al. (1998) back-
extrapolated exposures and body burdens using a first-order model with a constant half-life. In
contrast, Cheng et al. (2006) back-extrapolated body burdens using a kinetic modeling approach
that incorporated concentration- and age-dependent elimination kinetics.
Although all three of these are high-quality studies, the kinetic modeling used by Cheng
et al. (2006) is judged to better reflect TCDD pharmacokinetics, as currently understood, than the
first-order models used by Steenland et al. (2001) and Becher et al. (1998). EPA believes that
the representation of physiological processes provided by Cheng et al. (2006) is more realistic
than the assumption of simple first-order kinetics and this outweighs the attendant modeling
uncertainties. Furthermore, the use of kinetic modeling is consistent with recommendations both
by the NAS and the Dioxin Workshop panel.
However, as discussed in Section 3.3.2, the kinetic model that they employed does have
certain limitations, including the fact that it has been calibrated based on a relatively small
number of human subjects. In addition, their kinetic model does not allow body mass index
(BMI; and hence fat content) to vary with age, which may bias the model results. Nonetheless,
EPA prefers the increased technical sophistication of the dose estimates used in the cancer
mortality risk estimates derived from Cheng et al. (2006) to those derived from Steenland et al.
(2001).
EPA, therefore, has decided to use the results of the Cheng et al. (2006) study for
derivation of the TCDD OSF based on total cancer mortality as calculated by EPA using
data and models from the Cheng et al. (2006) study as described in Section 5.2.3.1.2.
Table 5-3 shows the oral slope factors at specific target risk levels (OSFRLs) which range
from 3.7 x 10s to 1.3 x 106 per (mg/kg-day). EPA recommends the use of an OSF of
1.3 x 106 per (mg/kg-day) when the target risk range is 10~5 to 10~7.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-67 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
5.3.1. Uncertainty in Estimation of Oral Slope Factors From Human Studies
A substantial degree of uncertainty is associated with the estimation of slope factor
values and cancer risk specific doses for TCDD based on the epidemiological studies. In some
instances, the influence of a given factor is theoretically amenable to analysis, but such
investigation is limited by the availability of sufficiently detailed data to support such an
analysis. In other cases, only very broad ranges can be placed on the uncertainty associated with
a given feature of the analysis, or uncertainties must be discussed qualitatively.
5.3.1.1. Uncertainty in Exposure Estimation
The major technical challenge within each of the epidemiological studies was developing
relevant and precise estimates of exposure. While Warner et al. (2002) collected blood samples
relatively close to the time of the Sevesso accident and could reasonably estimate peak exposures
based on these collected samples, in the case of the Becher et al. (1998), Ott and Zober (1996),
Steenland et al. (2001), and Cheng et al. (2006) studies, the major exposure issues included the
following
• Selecting (an) appropriate dose metric(s) for dose-response modeling,
• Estimating serum TCDD levels for the entire cohort based on measurements from a
smaller number of the subjects in the cohort collected long after the occupational
exposures had occurred, and then assigning exposures to the remaining members of the
cohort based on qualitative job classifications.
• Estimating time-weighted average tissue doses (e.g., lipid-average serum concentration
over time) based on single samples taken at one point in time. (Except for the Becher et
al. [1998] analysis where one of the study strengths was their estimate of TCDD half life,
which utilized repeated measurements from a subset of their cohort).
In the Becher et al. (1998), Steenland et al. (2001), and Cheng et al. (2006) studies, dose-
response modeling was performed using ppt-years lipid-adjusted serum concentration as the
primary dose metric for TCDD. The choice of serum concentration was natural in the sense that
the serum TCDD was the only direct measurement of exposure or dose that was available. In
addition, as discussed in Section 3.3.4, serum concentration is a reasonable index of total tissue
concentration (target organ dose), and lipid-adjusted serum concentration provides a reasonable
index of TCDD in the fatty components of tissues. Ott and Zober (1996) used ng/kg body
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-68 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
weight at the time of the accident as the primary dose metric, and EPA later converted these to
units of ppt-years lipid-adjusted serum concentration.
The decision to use cumulative serum concentrations (ppt-years) as the primary dose
metric for carcinogenicity is based on the understanding that time weighted concentrations (over
a chronic exposure period) are the most appropriate dose measures for cancer risk assessment.
This may not be strictly true if cancer induction by TCDD is considered to be a "threshold
process." However, as discussed in Section 5.2, there are reasonable grounds to believe that the
assumption of low-dose linearity is reasonable for TCDD, especially when calculating
population risks where the effects of interindividual variability must be taken into account.
In addition to the issue of low-dose thresholds, the rationale for using cumulative dose
metrics also can fail at high doses if the adverse response in question involves a step that is
saturable (e.g., where there is a maximum level of response that cannot be exceeded owing to a
rate-limited process). There is some evidence for such a phenomenon in the NIOSH cohort
where cancer risks in the highest exposure group (>50,000 ppt-years) appear to saturate, and the
response decreases at this level. Steenland et al. (2001) suggest that the apparent saturation of
dose-response in this cohort may be due, at least partially, to exposure misclassification, rather
than to an actual reduction in response per unit exposure.
The uncertainty associated with differences in the exposure patterns is important to
consider across the five epidemiologic studies. Steenland et al. (2001), Cheng et al. (2006), and
Becher et al. (1998) studied cohorts exposed to elevated TCDD levels over a long period of time,
while Ott and Zober (1996) and Warner et al. (2002) studied cohorts exposed to TCDD levels
significantly above background at one point in time but the exposures and likely the TCDD body
burdens declined significantly following these periods of elevated exposure. Both these chronic
and acute exposures can be analyzed in terms of cumulative exposure to TCDD. Use of such a
metric requires an assumption that the "actual" cancer potency associated with a cumulative dose
where much of the dose is received at a single point in time and then gradually eliminated would
be similar to the cancer potency of the same cumulative dose received over a longer period of
time and also gradually eliminated. While EPA believes that such an assumption is not
unreasonable, the experiment of Kim et al. (2003), which showed statistically significant
increase in liver effects due to a peak TCDD dose when compared to chronically-dosed Sprague-
Dawley rats administered the same levels of TCDD when measured as a cumulative dose,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-69 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
suggests that additional analyses of cumulative and peak TCDD dose measures may need to be
conducted.
There are uncertainties associated with the approaches used to estimate TCDD exposures
in the members of the occupational epidemiologic studies for which no measurement data were
available. To impute TCDD levels for workers without measured samples, all four occupational
epidemiologic studies matched workers for whom measured TCDD samples had never been
reported to workers with measured TCDD levels based on job histories. The NIOSH cohort is
used to illustrate some of the uncertainties. In the NIOSH cohort, the subset of workers (roughly
five percent of the total cohort) with blood serum data comprised surviving members of the
cohort (in 1988), and therefore, their age distribution would have differed from the rest of the
cohort. For each worker in this subset, the following data were available: (1) job classification
information, (2) employment history, and (3) serum TCDD measures. All of the workers in this
subset were employed at a single plant where the work histories were less detailed than at other
plants, and many of the workers at this plant had the same job title and were employed during the
same calendar period. There is an assumption that workers with same job title and work history
were exposed to the same TCDD levels within a plant and across plants; this obviously does not
account for exposure heterogeneity.
Both Steenland et al. (2001) and Cheng et al. (2006) addressed the potential for exposure
measurement error in TCDD estimates and possible exposure misclassification. For the highest
exposure workers, Steenland et al. (2001) and Cheng et al. (2006) found weak, "noisy," and/or
negative exposure-response relationships. Steenland et al. (2001) suggests that possible
explanations for this observation include the saturation of effects at the upper end of the dose-
response curve, instability of the TCDD exposure estimates based on the limited number of
highly exposed individuals, and the increased probability of exposure misclassification for
workers whose job histories indicate the highest exposures. As Steenland et al. (2001) reported,
some of the highest exposures might have been inaccurately estimated because they occurred in
workers exposed to short-term, high-dose exposures during spill clean-up. Cheng et al. (2006)
used sensitivity analyses to examine this measurement error issue and evaluated the potential for
exposure misclassification by using ln-transformed TCDD ppt-years. The authors also removed
all observations with exposures within the lower and upper 1, 2.5, or 5th percentiles of the
TCDD ppt-year distribution and also removed observations within just the upper 1, 2.5, or 5th
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-70 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
percentile of TCDD ppt-years. These sensitivity analyses provided similar results. An
additional concern is that exposure errors might distort the exposure distribution in the
population, which generally spreads the response out over a wider dose range. This serves to
increase the variance of the regression model, altering both the POD and the corresponding OSF.
Becher et al. (1998) only considered workers from a single plant but their analysis
included workers employed in five different job locations within the plant. The influence of
worker location on slope factor estimates does not appear to be further explored and may
represent a source of uncertainty.
To estimate long-term body burden metrics from the serum TCDD measurements,
Steenland et al. (2001) employed simple first order kinetic elimination rate model with a half-life
of 8.7 years. Limitations of this approach include (1) the average elimination half-life among the
study subjects may not be 8.7 years given differences between the study population and the
Ranch Hand population from which the value was estimated, (2) use of a single-value estimate
fails to take into account the inherent variability in elimination half life among the individual
workers, and (3) it fails to take into account variations in elimination kinetics throughout the
lifetime of the exposed worker due to change in body fat, age, etc. The impact of these potential
sources of bias on the estimates of time-integrated body burden cannot be quantitatively
assessed. However, Steenland et al. (2001) noted that modest changes in elimination half-life (to
7.1 years) had only a very small impact on risk estimates.
Cheng et al. (2006) estimated past body burdens using the CADM approach (described in
Section 3) (Aylward et al., 2005a, b) rather than a half-life estimate. As noted above, the
incorporation of concentration- and age-dependent elimination into this approach has significant
advantages over the use of a constant elimination half-life. However, as discussed in
Section 3.3, the CADM has only been subject to limited testing against human validation data
sets, so the degree to which its advantages are realized in practice cannot be easily assessed.
There are no available human data in the low dose region, the region of interest to this
assessment, to compare with the CADM (or Emond) model predictions.
Becher et al. (1998) developed half life estimates based on multiple TCDD blood
measures in 48 individuals from this cohort. These half life estimates were then used to back
calculate TCDD concentrations at the end of each worker's employment, accounting for age and
percentage of body fat. This cohort-specific information may provide a better exposure estimate
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-71 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
than Steenland et al. (2001) or Ott and Zober (1996) who used similar kinetic approaches.
However, the comparison of the accuracy of the exposure estimates across the cohorts is not
easily assessed. There are several assumptions and important uncertainties involved in modeling
TCDD exposures in these cohorts. The study authors have invoked different kinetic assumptions
when extrapolating measured levels of TCDD in sera backward in time to estimate higher
chronic or peak dosage (i.e., there is uncertainty in these back-calculations that includes
assumptions regarding elimination kinetics). There is also uncertainty in applying such estimates
to other members of the cohort based on similar characteristics (e.g., job category).
5.3.1.2. Uncertainty in Shape of the Dose-Response Curve
Another source of uncertainty is the nature of the dose-response curve in the low dose
region of interest for risk assessment for environmental exposures (e.g., 5-10 ng/kg-day). Most
of the dose-response curves appear reasonably linear in this region. These epidemiologic data,
however, are based on occupational studies in which exposures were often several orders of
magnitude higher than environmental exposures. Data from these studies are quite sparse in the
low dose region, and only one study examined uncertainty due to the low dose region. Steenland
and Deddens (2003) attempted to analyze this region specifically by fitting threshold curves to
the NIOSH data in which there was no extra risk from exposure until some specific level.
However, this model did not fit as well as models without a threshold. In general, the usual
assumption of linearity in the low dose region seems reasonable when using epidemiologic data
given the lack of data in this region that precludes the rejection of linearity.
There is uncertainty in the extrapolation of the OSF to the low dose region (e.g.,
<1 ng/kg-day). EPA developed the cancer assessment in this document assuming the slope in the
low-dose region of the dose-response curve is linear; the decision was made due to the lack of
sufficient evidence to support an assumption of nonlinearity as outlined in the 2005 Cancer
Guidelines (U.S. EPA, 2005). Similarly, there is uncertainty as to whether a threshold exists for
TCDD-induced toxicity leading to tumorigenesis and the dose associated with such a threshold,
if it exists, is unknown. EPA chose to model this dose-response without a threshold because
there is insufficient evidence to support an assumption of a threshold.
It also is noteworthy that the shapes of the exposure-response in several of these studies,
based on the published statistical models, is indicative of a response that tends to tail off or
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-72 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
"plateau" at high cumulative exposures to TCDD. This phenomenon has been seen in many
studies of occupational carcinogens, and may reflect a number of things including exhaustion of
people susceptible to cancer, saturation of biological pathways which are part of the pathway to
cancer, and increased error measurement of dose at high levels biasing dose-response towards
the null (Stayner et al., 2003).
5.3.1.3. Uncertainty in Defining the Reference Population
Another source of uncertainty using human epidemiologic data is due to the lack of
completely unexposed populations. Epidemiologic data in TCDD studies is based on comparing
health outcomes from populations which experienced elevated TCDD exposures to outcomes
from populations which experienced lower exposures, rather than to a population with zero
TCDD exposure (They also compared outcomes in members of the cohort that were more highly
exposed to TCDD to outcomes in members that were less exposed to TCDD). This lack of a
completely unexposed population in epidemiology studies is inevitable given that truly
unexposed human populations do not exist. Therefore, the extra risk calculated in the form of
EDois, is inevitably the extra risk above a low background exposure due to dioxins in the general
environment. Typically, the general population in western countries where the epidemiologic
studies have been done have had serum levels on the order of 5 ppt. Hence, the extra risks may
be considered as those incurred by added exposure above these background doses. For example,
an EDoi of 18 ng/kg, which yields an excess risk of cancer mortality of 1%, would mean
18 ng/kg above a typical background level of 5 ng/kg (5 ppt).
5.3.1.4. Uncertainty in Cancer Risk Estimates
It is important to remember the overall uncertainty associated with cancer risk estimates
for TCDD can be approximately bounded by evaluating the spread of estimates from the several
epidemiological studies that have been conducted. In the 2003 Reassessment, EPA noted that
the range of ED0i/lower bound of the 95% confidence interval on the dose that yields a 1% effect
(LEDoi) values identified in the then-available dose-response studies (Steenland et al., 2001;
Becher et al., 1998; Ott and Zober, 1996) implied an approximate range of all-cancer mortality
slope factors between 1.2 to 2.5 per ng/kg-day. The equivalent oral cancer slope factors resulting
from the analysis of the newer Cheng et al. (2006) study are below and at the low end of this
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-73 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
range (0.37-1.2 per ng/kg-day for fatal cancer risks between 1 x 10 3 and 1 x 10 7), This
suggests that the highest quality epidemiological studies all yield risk estimates that agree to
within about a factor of 7. This relatively close agreement between the studies of the different
cohorts lends additional confidence to the slope factor derivation. EPA notes that each of these
estimates is derived from an occupational cohort, and significant differences likely exist (e.g.,
exposure levels and routes) between these cohorts and those exposed to TCDD environmentally
(i.e., primarily through the food chain).
5.3.2. Other Sources of Uncertainty in Risk Estimates From the Epidemiological Studies
There are a number of other aspects of the Steenland et al. (2001), Cheng et al. (2006),
Becher et al. (1998), and Ott and Zober (1996) studies that may contribute uncertainty to the
cancer risk estimates. First, all studies that meet the criteria (with the exception of Warner et al.,
2002) measure cancer mortality rather than cancer incidence. This presumably biases the slope
factor downward relative to that which would be calculated for cancer incidence, which would
give a truer picture of the total health impacts of TCDD exposures on the general population. In
the NIOSH cohort, roughly one-third of the fatal cancers were from lung cancer. Because of the
high case mortality rate associated with lung cancer during the period of cohort evaluation, the
slope factor estimated for cancer mortality might not be much lower than that calculated for
cancer incidence. Estimation of cancer incidence in the general population associated with
TCDD exposure would require assumptions related to the relative survival and age-specific
cancer risks in the exposed population compared to the NIOSH cohort or the Hamburg cohort;
insufficient data are available to support such an analysis.
The general issues associated with potential confounding effects were noted and
discussed in the 2003 Reassessment. In addition to smoking and lifestyle factors that might
affect cancer risks, intra-individual variation in TCDD kinetics and susceptibility could also
affect the relationship between exposure and cancer risk. One specific example of such a
confounding factor would be variation in TCDD elimination half-lives. For example, if a large
proportion of the 256 observed cancer deaths occurred in NIOSH workers with longer half lives,
this could bias the slope factor downward because higher doses would be associated with cancer
cases in the deceased workers.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-74 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Another source of uncertainty in the exposure/dose assessment in the Steenland et al.
(2001), Cheng et al. (2006), Becher et al. (1998), and Ott and Zober (1996) studies is the limited
of information related to exposure to dioxin-like compounds (DLCs). In addition, the Hamburg
cohort was also exposed to hexachlorocyclohexane (HCH) and lindane. The oral slope estimates
derived by these studies attribute all the observed excess cancer mortality solely to TCDD
exposures. This assumes that occupational exposure was entirely to TCDD, with no other
occupational exposure to dioxins and furans. However, because TCDD typically occurs as a
component of a mixture with other DLCs, this assumption could leads to a positive bias in the
slope factors estimates derived from these epidemiologic studies if confounding is present. The
magnitude of the potential bias can be estimated in a general way through the estimation of risks
for plausible mixtures of DLCs and TCDD exposures in the cohort with the same composition as
the Steenland et al. (2001) and Cheng et al. (2006) studies, but the detailed data required to
perform such an analysis on the NIOSH cohort are not available.
Cheng et al. (2006) assessed the impact of possible confounding by conducting excluding
individual plants in the modeling. If the estimated cancer risks as a function of exposure did not
change too much when specific facilities were left out, then confounding was deemed unlikely.
Cheng et al. (2006) likewise found little variation in risks based on these analyses. In addition to
the slope factor estimated for TCDD, Becher et al. (1998) also evaluated the slope based on
TEQs. They found a dose-response effect for TCDD but not for TEQ (excluding TCDD) which
suggests that confounding by DLCs did not occur.
There is adequate evidence to believe age, gender, and body fat content all can have a
significant impact on elimination kinetics and consequent cancer risks associated with TCDD
exposure (U.S. EPA, 2003). The Hamburg cohort accounted for impacts in their kinetic analysis.
There may be gender differences that affect susceptibility to TCDD exposure, and the NIOSH
cohort and the Becher et al. (1998) analysis were comprised almost exclusively of men, so these
differences were not systematically addressed. There are few quantitative data on which to
examine the potential variation of these modifying factors on exposure estimates; however
several of the studies have reported similar results based on variable half-life estimates.
Finally, none of these cancer cohorts contained any children, and the unique sensitivities
of infants, toddlers, and children were not addressed. Aside from differences in exposure
patterns and body fat content, the unique developmental status of children may result in a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-75 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
substantially different profile of health risks (and magnitudes of those risks) than can be
addressed by simply compensating on the basis of differences in body weight, food intake, etc.
Further, because EPA could not develop an estimate from the Warner et al. (2002) cohort, none
of the studies for cancer dose-response analyzed contained a significant percentage of women.
Thus, the generalizability of the slope factor estimates to women and children is uncertain.
5.3.3. Approaches to Combining Estimates From Different Epidemiologic Studies
Meta-analyses and pooled analyses are two common approaches for combining
epidemiologic study data. Meta-analyses are a useful way to combine epidemiologic data from
different studies and derive a common estimate of effect, particularly when there are a large
number of comparable studies that are fairly homogenous as to make them impossible to
combine. A meta-analysis often involves a weighted average of effect measures, dose-response
coefficients, orEDoiS.
Unlike a meta-analysis, a pooled analysis combines the original exposure and health
outcome data across multiple studies, enabling a fit of new models to the data which were not
used in the original publications. Whereas a pooled analysis of the four different cohorts
considered here would be useful to explore the functional form and fit of models (either
statistical or multistage) across all four cohorts, this would entail a lengthy undertaking and is not
being contemplated here, due in part to concerns about the confidence in the results of such an
undertaking.
5.3.3.1. The Crump et al (2003) Meta-analysis
Crump et al. (2003) published a meta-analysis that incorporated data from the three
studies EPA used in the quantitative dose-response modeling presented in the 2003
Reassessment (U.S. EPA, 2003). These three study populations were the NIOSH (Steenland et
al., 2001), the Hamburg (Becher et al., 1998), and the BASF (Ott and Zober, 1996) cohorts. The
data for the NIOSH study included six additional years of follow-up and improved TCDD
exposure estimates that had not been applied to EPA's dose-response modeling in the 2003
Reassessment. This study examined the relationship between TCDD exposure and all-cancer
mortality. SMR statistics that had been used in all three studies were applied.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-76 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
The Crump et al. (2003) analysis was based on published data, and therefore, selection of
the dose metric was limited to how aggregated data had been presented in the publications. For
the NIOSH component of the analysis, the exposure data were based on worker-specific data and
specific processes performed at each plant (Steenland et al., 2001). The previous approach
assigned workers that had broad categories of exposure duration with the same cumulative serum
level, and did not take into account the particular plant or the job assignment within the plant.
The Crump et al. (2003) approach did take into account when exposure occurred in relation to
the follow-up interval. The TCDD exposure metric used was a cumulative serum lipid
concentration (CSLC). For the Hamburg cohort, Crump et al. (2003) used an average value from
the exposure ranges provided in Flesch-Janys et al. (1998). For the BASF cohort, arithmetic
averages for the dose categories were converted to TCDD CSLC intakes by dividing them by
0.25 (average body fat of 25%) and a decay rate that corresponded to a half-life of 7 years.
The outcome variable for the dose-response modeling was all cancer mortality, and
CSLC was the independent variable. Crump et al. (2003) performed a series of trend tests to
determine the lowest dose for which a statistically significant trend in SMR could be shown and
all other lower doses. These tests also examined the highest dose in which there was no
statistically significant trend using data from this dose and all other lower doses. Estimates of
EDio, EDos, and ED0i for TEQ with respect to the lifetime probability of dying from cancer were
calculated. This calculation assumed a first-order elimination process with a half-life of
7.6 years, a 50% systemic uptake of ingested dioxin, that dioxin concentration in serum lipid is a
suitable measure for dioxin concentration in all lipid, and that all dioxin is sequestered in lipid
(which comprises 25% of body weight). Age-specific mortality rates in the presence of dioxin
exposure were then generated. Life-table methodology was used to calculate lifetime risks of
cancer mortality.
Based on the modeling results, the hypothesis of a baseline SMR of 1.0 was rejected, and
the linear model produced an SMR estimate of 1.17 (95% CI = 1.04-1.30) from these studies.
The dose-response curves for the three studies were not homogeneous. Namely, the points from
the BASF cohort fell below the predicted curve. Because the heterogeneity was not judged to be
extreme by different statistical tests, however, the investigators used a common model in a
combined analysis of the data from the three studies. The linear model provided an adequate fit
of the data, and the slope associated with CSLC-ppt was 6.3 x io~6 (95% CI = 8.8 x io~7 to
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-77 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
1.3 x 10 5). Based on goodness of fit analysis, the preferred estimate of ED0i was 7.7 pg/kg/day,
which was six times higher than the estimate derived Steenland et al. (2001).
5.3.3.2. EPA 'v Decision Not to Conduct a Meta-analysis
From a statistical perspective, meta-analyses may not be very reliable when applied to a
small number of studies. Crump et al. (2003) used only three studies. Had EPA undertaken a
meta-analysis for the studies that met its criteria, most of the weight would come from the two
large studies on the NIOSH and Hamburg cohorts. However, such an analysis relies on an
assumption of a normally distributed between-study effect. This normality assumption cannot be
assessed with only three observations, yet the meta-analysis estimate is highly sensitive to this
distributional assumption (Higgens et al., 2009). Because of this limitation and the imprecision
of the between-study variance estimate, statisticians often recommend forgoing meta-analysis in
favor of discussing the individual studies when few studies are available (Cox, 2006; Higgens et
al., 2009). Based on these considerations, EPA decided not to undertake a meta-analysis in this
document.
As noted previously, Crump et al. (2003) has conducted a meta-analysis of the three
cohorts considered here, i.e., the NIOSH, Hamburg, and BASF cohorts. However, Crump et al.
modeled SMR data in which the cohorts were compared to the general population, rather than on
internal exposure-response analyses as relied upon in this document. Their analysis included a
total of 15 different SMRs from the three studies. A prior analysis of the dose-responses by
Becher et al. (1998) was used (i.e., the categorical SMR analysis by Flesch-Janys et al. [1998]).
Additionally, a prior analysis of the NIOSH cohort (Steenland et al., 1999) in which SMRs were
calculated was used. Crump et al. (2003) found that a linear dose-response gave a good fit to the
data, and used that for deriving an ED0i. However, they found that a supra-linear dose-response
provided a better fit to the data, but rejected the supra-linear model (a power model) because of
an infinite slope at zero dose. In the original publications by Becher et al. (1998) and Steenland
et al. (2001), both observed a supra-linear dose-response trend. Crump et al. (2003) concluded
that the ED0i was 45 pg/kg-day, similar to the ED0i of 8 pg/kg-day calculated by Steenland et al.
(2001) using the same dietary units (pg/kg-day).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-78 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Table 5-1. Cancer slope factors calculated from Becher et al. (1998),
Steenland et al. (2001) and Ott and Zober (1996) from 2003 Reassessment
Table 5-4
Study
EDoi (LEDoi)
(ng/kg)
Cancer slope factor per
ng/kg-day above
background3 (UCL)
Hamburg cohort
Power model
Becher et al. (1998)
6 (N.A.)
5.1 (N.A.)
Hamburg cohort
Additive model
Becher et al. (1998)
18.2 (N.A.)
1.6 (N.A.)
Hamburg cohort
Multiplicative model
Becher et al. (1998)
32.2 (N.A.)
0.89 (N.A.)
NIOSH cohort
Piecewise linear model
Steenland et al. (2001)
18.6 (11.5)
1.5 (2.5)
BASF cohort, from Ott
and Zober (1996),
multiplicative
50.9 (25.0)
0.57 (1.2)
aAssumes 25% of body weight is lipid; in humans 80% of dioxin dose is absorbed from the normal
diet; the TCDD half-life is 7.1 years in humans. Background all cancer mortality rate calculated
through lifetable analysis to 75 years. Summary results are for male all cancer risk, because the
male lifetime (to 75 years) all cancer risk is greater than for females, leading to correspondingly
higher cancer slope factors. As detailed in Part III, Chapter 8, RelRisk(ED0i) = 0.99 +
0.01/Risk(0 dose)- Based on the manner in which the dose-response data were calculated using Cox
regression rate ratio analyses, risks are given as cancer slope factors for 1 pg/kg-day above
background, assumed 5 ppt TCDD in lipid.
UCL = upper confidence limit.
Source: U.S. EPA (2003; Part III, Chapter 5, Table 5-4).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-79 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 5-2. Cox regression coefficient estimate and incremental risk for
2 NIOSH cohort data as presented in Cheng et al. (2006) compared with
3 Steenland et al. (2001)
4
Model
Cox regression
coefficient estimate
(ppt-year)"1
Incremental risk3
Ro = 0.124
Ro = 0.112
Steenland et al. (2001) result
Piecewise linearb
1.5 x 10 5
7.0 x 10 4
6.3 x 10 4
Unlagged exposure, Cheng et al. (2006)
Piecewise linearc
1.4 x l(T6
6.5 x 10~5
5.9 x 10~5
Linear, lower 95% of observations
1.6 x l(T6
7.4 x 10~5
6.7 x 10~5
Linear, full data
-8.9 x l(T9d
<0
<0
Lagged exposure (15 years), Cheng et al. (2006)
Linear, lower 95% of observations
3.3 x 10 6
1.2 x 10 4
1.1 x 10 4
Linear, full data
1.7 x I0~8d
6.3 x 10~7
5.7 x 10~7
5
6 "Assumes 5 ppt serum lipid TCDD concentration for 75 years (unlagged) or 60 years (taking into account a 15-year
7 lag).
8 bCox regression coefficient from the Steenland et al. (2001) analysis as reported in the 2003 Reassessment, Part III,
9 pp. 5-34, males, no lag. The value of this coefficient for the piecewise linear modeling was not explicitly reported
10 by Steenland et al. (2001).
11 °Piecewise linear with cutpoint set at maximum likelihood for model fit, 452,000 ppt-years.
12 dNot statistically significantly different from 0.
13
14 Source: Cheng et al. (2006; Table IV).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-80 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 5-3. Comparison of lipid-adjusted serum concentrations, fat
2 concentrations, risk specific dose estimates and equivalent oral slope factors
3 based on upper 95th percentile estimate of regression coefficient3 of all fatal
4 cancers reported by Cheng et al. (2006) for risk levels of 1 x 10~3,1 x 10~4,
5 lx 10"5,1 x 10"6, and 1 x 10"7
6
Risk level
(RL)
AUCrl,
(ppt-yr)
FATrL
(ng/kg)
Risk specific dose
(Drl) (ng/kg-day)
Equivalent oral slope
factors (OSFrl) per
(mg/kg-day)
1 x 1(T3
1.26 x l(T3
1.803 x 101
2.73 x 10~3
3.7 x 105
1 x l(T4
1.26 x 102
1.803 x 10
1.23 x 10~4
8.1 x 105
1 x l(T5
1.26 x 101
1.803 x 10_1
8.57 x 10~6
1.2 x 106
1 x l(T6
1.26 x 10
1.803 x 10~2
7.77 x 10~7
1.3 x 106
1 x l(T7
1.26 x KT1
1.803 x 10~3
7.62 x 10~8
1.3 x 106
7
8 a Based on regression coefficient of Cheng et al. (2006; Table III), excluding observations in the upper 5% range
9 (>252,950 ppt-year lipid adjusted serum TCDD) of the exposures; where reported (3 = 3.3 x 10 6 ppt-years and
10 standard error =1.4 / 10 6, Upper 95%tile estimate of regression coefficient (p95) calculated to be
11 6.04 x 10~6 = (3.3 x 10~6) + 1.96 x (1.4 x 10~6).
12
13
14 Table 5-4. Comparison of lipid-adjusted serum concentrations, fat
15 concentrations, risk specific dose estimates and equivalent oral slope factors
16 based on best estimate of regression coefficient3 of all fatal cancers reported
17 by Cheng et al. (2006) for risk levels of 1 x 10"3,1 x 10"4,1 x 10"5,1 x 10"6,
18 and 1 x 10"7
19
Risk level
(RL)
AUCrl,
(ppt-yr)
FATrL
(ng/kg)
Risk specific dose
(Drl) (ng/kg-day)
Equivalent oral
slope factors
(OSFrl) per
(mg/kg-day)
1 x 10~3
2.31 x io3
3.303 x 101
6.63 x 10~3
1.5 x 105
1 x 10~4
2.31 x io2
3.303 x 10
2.63 x 10-4
3.8 x 105
1 X 10~5
2.31 x io1
3.303 x 10_1
1.67 x 10~5
6.0 x 105
1 X 10~6
2.31 x 10
3.303 x 10-2
1.44 x 10-6
6.9 x 105
1 X 10-7
2.31 x 10_1
3.303 x 10~3
1.40 x 10~7
7.1 x 105
20
21 aBased on regression coefficient of Cheng et al. (2006; Table III), excluding observations in the upper 5% range
22 (>252,950 ppt-year lipid adjusted serum TCDD) of the exposures; where reported (3 = 3.3 x 10 6 ppt-years.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-81 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 5-5. Kociba et al. (1978) male rat tumor incidence data" and blood
2 concentrations for dose-response modeling
3
Morphology: topography
Vehicle control
(ng/kg)
Low dose
(ng/kg)
Medium dose
(ng/kg)
High dose
(ng/kg)
0
860
3,945
21,334
Stratified squamous cell
carcinoma of hard palate or nasal
turbinates
0/85
0/50
0/50
4/5 0b
Stratified squamous cell
carcinoma of tongue
0/85
1/50
1/50
3/50b
Adenoma of adrenal cortex
0/85
0/50
2/50
5/5 0b
4
5 aSource: Kociba et al. (1978; Table 4).
6 Statistically significant by Fischer Exact Test (p < 0.05).
7
8
9 Table 5-6. Kociba et al. (1978) female rat tumor incidence data3 and blood
10 concentrations for dose-response modeling
11
Morphology: topography
Vehicle control
(ng/kg)
Low dose
(ng/kg)
Medium dose
(ng/kg)
High dose
(ng/kg)
0
853
3,942
21,246
Hepatocellular adenoma(s) or
carcinoma(s)
2/86
1/50
9/5 0a
18/45b
Stratified squamous cell
carcinoma of hard palate or
nasal turbinates
0/86
0/50
1/50
4/49b
Keratinizing squamous cell
carcinoma of lung
0/86
0/50
0/50
7/49b
12
13 aSource: Kociba et al. (1978; Table 5). Incidence for Hepatocellular adenomas or carcinomas is from Goodman and
14 Sauer (1992; Table 1); EPA calculated statistical significance as the study authors did not provide this.
15 Statistically significant by Fischer Exact Test (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-82 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 5-7. NTP (1982) female rat tumor incidence data3 and blood
2 concentrations for dose-response modeling
3
Morphology: topography
Vehicle control
(ng/kg)
Low dose
(ng/kg)
Medium dose
(ng/kg)
High dose
(ng/kg)
0
1,072
3,111
16,207
Subcutaneous tissue: fibrosarcoma
0/75
2/50
3/50
4/49b
Liver: neoplastic nodule or
hepatocellular carcinoma
5/75c
1/49
3/50
14/49b
Adrenal: cortical adenoma, or
carcinoma or adenoma, NOS
1l/73c
9/49
5/49
14/46b
Thyroid: follicular-cell adenoma
3/73c
2/45
1/49
6/47
4
5 "Source: NTP (1982; Table 10).
6 Statistically significant by Fischer Exact Test (p < 0.05).
7 Statistically significant trend by Chochran-Armitage test (p < 0.05).
8
9
10 Table 5-8. NTP (1982) male rat tumor incidence data3 and blood
11 concentrations for dose-response modeling
12
Morphology: topography
Vehicle control
(ng/kg)
Low dose
(ng/kg)
Medium dose
(ng/kg)
High dose
(ng/kg)
0
1,072
3,116
16,272
Liver: neoplastic nodule or
hepatocellular carcinoma
0/74b
0/50
0/50
3/50
Thyroid: follicular-cell adenoma or
carcinoma
l/69b
5/48c
8/5 0C
1 l/50c
Adrenal cortex: adenoma
6/72
9/50
12/49b
9/49
13
14 "Source: NTP (1982; Table 9).
15 Statistically significant trend by Chochran-Armitage test (p < 0.05).
16 Statistically significant by Fischer Exact Test (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-83 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 5-9. NTP (1982) female mouse tumor incidence data3 and blood
2 concentrations for dose-response modeling
3
Morphology: topography
Vehicle control
(ng/kg)
Low dose
(ng/kg)
Medium dose
(ng/kg)
High dose
(ng/kg)
0
1,064
3,184
17,406
Subcutaneous tissue: fibrosarcoma
l/74b
1/50
1/48
5/47c
Hematopoietic system: lymphoma
or leukemia
18/74b
12/50
13/48
20/47c
Liver: hepatocellular adenoma or
carcinoma
3/73b
6/50
6/48
11/47°
Thyroid: follicular-cell adenoma
0/69b
3/50
1/47
5/46c
4
5 "Source: NTP (1982; Table 15).
6 Statistically significant trend by Chochran-Armitage test (p < 0.05).
7 Statistically significant by Fischer Exact Test (p < 0.05).
8
9
10 Table 5-10. NTP (1982) male mouse tumor incidence data3 and blood
11 concentrations for dose-response modeling
12
Morphology: topography
Vehicle control
(ng/kg)
Low dose
(ng/kg)
Medium
dose (ng/kg)
High dose
(ng/kg)
0
420
1,240
6,118
Lung: alveolar/bronchiolar adenoma
or carcinoma
10/7 lb
2/48
4/48
13/50
Liver: hepatocellular adenoma or
carcinoma
15/73b
12/49
13/49
27/50c
13
14 "Source: NTP (1982; Table 14).
15 Statistically significant trend by Chochran-Armitage test (p < 0.05).
16 Statistically significant by Fischer Exact Test (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-84 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 5-11. NTP (2006) female rat tumor incidence data" and blood
2 concentrations for dose-response modelingb
3
System:
morphology:
topography
Vehicle
control
(ng/kg)
Low
dose
(ng/kg)
Low-med
dose (ng/kg)
Median
dose
(ng/kg)
Med-high
dose (ng/kg)
High dose
(ng/kg)
0
1,408
3,137
5,393
9,129
16,361
Liver:
cholangiocarcinoma
0/49c
0/48
0/46
1/50
4/49
25/53°
Liver:
hepatocellular
adenoma
0/49c
0/48
0/46
0/50
1/49
13/53c
Oral mucosa:
squamous cell
carcinoma
l/49c
2/48
1/46
0/50
4/49
10/53°
Pancreas: adenoma
or carcinoma
0/48c
0/48
0/46
0/50
0/48
3/51
Lung: cystic
keratinizing
epithelioma
0/49c
0/48
0/46
0/49
0/49
9/52c
4
5 "Source: NTP (2006; Table A3 a).
6 bIncidence adjusted for animals <365 days on study.
7 Statistically significant by Poly-3 Test (p < 0.05).
8
9
10 Table 5-12. Toth et al. (1979) male mouse tumor incidence data3 and blood
11 concentrations for dose-response modeling
12
Morphology: topography
Vehicle control
(ng/kg)
Low dose
(ng/kg)
Medium dose
(ng/kg)
High dose
(ng/kg)
0
315
7,814
50,105
Liver tumors
7/38
13/44
21/44b
13/43
13
14 aSource: Toth et al. (1979; Table 1).
15 Statistically significant by Chi2 Test (p < 0.01).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-85 DRAFT—DO NOT CITE OR QUOTE
-------
Table 5-13. Comparison of multi-stage modeling results across cancer bioassays using blood concentrations
Study
Species
Sex
Morphology: topography
Multi-stage modeling*:
stage, GoF />-value, LL difference
BMD01
(ng/kg)
BMDL01
(ng/kg)
Kociba
et al.,
1978
Rat
Male
Stratified squamous cell carcinoma of hard palate or nasal turbinates
\,p = 0.81
3,175
1,540
Stratified squamous cell carcinoma of tongue
\,p = 0.47
3356
1433
Adenoma of adrenal cortex
\,p= 0.78
1,793
1,020
Combined tumors Bayesian analysis
849
528
Female
Hepatocellular adenoma(s) or carcinoma(s)
\,p = 0.24,
387
277
Stratified squamous cell carcinoma of hard palate or nasal turbinates
\,p = 0.97
2,484
1,289
Keratinizing squamous cell carcinoma of lung
\,p = 0.63
1,730
984
Combined tumors Bayesian analysis
280
206
NTP,
1982
Rat
Female
Subcutaneous tissue: fibrosarcoma
\,p = 0.18
1,700
751
Liver: neoplastic nodule or hepatocellular carcinoma
\,p = 0.22
638
402
Adrenal: cortical adenoma, or carcinoma or adenoma, NOS
\,p = 0.34
878
444
Thyroid: follicular-cell adenoma
\,p = 0.57
1,840
846
Combined tumors Bayesian analysis
251
172
Male
Liver: neoplastic nodule or hepatocellular carcinoma
\,p = 0.85
3,345
1,472
Thyroid: follicular-cell adenoma or carcinoma
l,p = 0.06
657
380
Adrenal cortex: adenoma
\,p = 0.06
2,161
665
Combined tumors Bayesian analysis
410
243
Mouse
Female
Subcutaneous tissue: fibrosarcoma
\,p = 0.93
1,849
916
Hematopoietic system: lymphoma or leukemia
\,p = 0.98
622
331
Liver: hepatocellular adenoma or carcinoma
\,p = 0.34
807
449
Thyroid: follicular-cell adenoma
1, p = 0.09, no improvement with
higher orders
1,653
779
Combined tumors Bayesian analysis
244
162
-------
0
>3*
&
to
s
S
to
s
>;*
a
a,
Sf
^s
^s
TO
TO'
*
^5
«
0
^ is
1 TO
00 ^
« o
s
1
a,
2 §-
7? ^
J> ^5
Tl S
H ©
3 §
2 §
O ^
H r?
<*^
hh Oq
H to
W|
O £?
o
H
W
Table 5-13. Comparison of multi-stage modeling results across cancer bioassays using blood concentrations
(continued)
Study
Species
Sex
Morphology: topography
Multi-stage modeling*:
stage, GoF />-value, LL difference
BMD01
(ng/kg)
BMDL01
(ng/kg)
NTP,
1982
cont.
Mouse
cont.
Male
Lung: alveolar/bronchiolar adenoma or carcinoma
l,p = 0.09
354
191
Liver: hepatocellular adenoma or carcinoma
l,p = 0.93
115
75
Combined tumors Bayesian analysis
88
58
NTP,
2006
Rat
Female
Liver: cholangiocarcinoma
3,/? = 0.99, dLL = 2.93
4,173
2,277
Liver: hepatocellular adenoma
3,/? = 0.93, dLL = 2.10
5,631
3,596
Oral mucosa: squamous cell carcinoma
l,p = 0.27
1,214
765
Pancreas: adenoma or carcinoma
l,p= 0.64
5,794
2,551
Lung: cystic keratinizing epithelioma
2,p = 0.51, dLL = 3.55
4,575
2,890
Combined tumors Bayesian analysis
651
431
Toth et
al,
1979
Mouse
Male
Liver: tumors
l,p = 0.29
203
115
*Analysis uses a chi-square goodness of fit statistic for differences in the log likelihoods (p > 0.05).
-------
1 Table 5-14. Individual tumor points of departure and slope factors using
2 blood concentrations
3
Study
Tumor Site (Sex/Species)
BM DI-iiid
(ng/kg-day)
OSF
(per mg/kg-day)
NTP, 1982
Liver: adenoma or carcinoma (male mice)
1.7E-03
5.9E+6
Tothetal., 1979
Liver tumors (male mice)
1.9E-03
5.2E+6
NTP, 1982
Lung: adenoma or carcinoma (male mice)
6.6E-03
1.5E+6
Kocibaetal., 1978
Liver: adenoma or carcinoma (female rats)
1.2E-02
8.6E+5
NTP, 1982
Hematopoietic: lymphoma or leukemia (female mice)
1.5E-02
6.6E+5
NTP, 1982
Thyroid: follicular cell adenoma (male rats)
1.9E-02
5.3E+5
NTP, 1982
Liver: neoplastic nodule or hepatocellular carcinoma
(female rats)
2.1E-02
4.9E+5
NTP, 1982
Adrenal: cortical adenoma or carinoma or adenoma,
NOS (female rats)
2.4E-02
4.2E+5
NTP, 1982
Liver: adenoma or carcinoma (female mice)
2.4E-02
4.1E+5
NTP, 1982
Adrenal cortex: adenoma (male rats)
4.4E-02
2.3E+5
NTP, 1982
Subcutaneous fibrosarcoma (female rats)
5.3E-02
1.9E+5
NTP, 2006
Oral mucosa: squamous cell carcinoma (female rats)
5.5E-02
1.8E+5
NTP, 1982
Thyroid: adenoma (female mice)
5.6E-02
1.8E+5
NTP, 1982
Thyroid: follicular cell adenoma (female rats)
6.4E-02
1.6E+5
NTP, 1982
Subcutaneous fibrosarcoma (female mice)
7.2E-02
1.4E+5
Kocibaetal., 1978
Lung: carcinoma (female rats)
8.0E-02
1.2E+5
Kocibaetal., 1978
Adenoma of adrenal cortex (male rats)
8.5E-02
1.2E+5
Kocibaetal., 1978
Nasal/Palate: carcinoma (female rats)
1.2E-01
8.2E+4
Kocibaetal., 1978
Tongue: carcinoma (male rats)
1.4E-01
7.0E+4
NTP, 1982
Liver: neoplastic nodule or hepatocellular carcinoma
(male rats)
1.5E-01
6.7E+4
Kocibaetal., 1978
Nasal/Palate: carcinoma (male rats)
1.6E-01
6.3E+4
NTP, 2006
Liver: cholangiocarcinoma (female rats)
2.9E-01
3.5E+4
NTP, 2006
Pancreas: adenoma or carcinoma (female rats)
3.4E-01
2.9E+4
NTP, 2006
Lung: cystic keratinzing epithelioma (female rats)
4.1E-01
2.4E+4
NTP, 2006
Liver: hepatocellular adenoma (female rats)
5.6E-01
1.8E+4
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-88 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 5-15. Multiple tumor points of departure and slope factors using blood
2 concentrations
3
Study
Sex/species: tumor sites
BMDLhed
(ng/kg-day)
OSF
(per mg/kg-day)
NTP, 1982
Male mice: liver adenoma and carcinoma, lung
1.1E-03
9.4E+6
NTP, 1982
Female mice: liver adenoma and carcinoma, thyroid
adenoma, subcutaneous fibrosarcoma, all lymphomas
5.3E-03
1.9E+6
NTP, 1982
Female rats: liver neoplasitc nodules, liver adenoma
and carcinoma, thyroid follicular cell adenoma,adrenal
cortex adenoma or carcinoma
5.7E-03
1.8E+6
Kocibaetal., 1978
Female rats: liver adenoma carcinoma, oral cavity,
lung
7.3E-03
1.4E+6
NTP, 1982
Male rats: thyroid follicular cell adenoma, adrenal
cortex adenoma
9.6E-03
1.0E+6
NTP, 2006
Female rats: liver cholangiocarcinoma, hepatocellular
adenoma, oral mucosa squamous cell carcinoma, lung
cystic keratinizing epithelioma, pancreas adenoma,
carcinoma
2.3E-02
44E+5
Kocibaetal., 1978
Male rats: adrenal cortex adenoma, tongue carcinoma,
nasal/palate carcinoma
3.1E-02
3.2E+5
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-89 DRAFT—DO NOT CITE OR QUOTE
-------
Table 5-16. Illustrative RfDs based on tumorigenesis in experimental animals
Study
Species, strain
(sex)
Protocol
Endpoint
BMDLhed3
(ng/kg-day)
RID1'
(mg/kg-day)
NTP, 1982
Mouse, B6C3F1,
male
2-year gavage;
n = 50
Liver adenoma and carcinoma, lung
1.1E-3
3.6E-11
NTP, 1982
Mouse, B6C3F1,
female
2-year gavage;
n = 50
Liver adenoma and carcinoma, thyroid adenoma,
subcutaneous fibrosarcoma, all lymphomas
5.3E-3
1.7E-10
NTP, 1982
Rat, Osborne-
Mendel, female
2-year gavage;
n = 50
Liver neoplasitc nodules, thyroid follicular cell adenoma,
liver adenoma and carcinoma, adrenal cortex adenoma or
carcinoma
5.7E-3
1.9E-10
Tothetal., 1979
Mouse, Swiss/
H/Riop, male
1-year gavage
(1-year average);
n = 38-44
Liver tumors
6.1E-3
2.0E-10
Kociba et al.,
1978
Rat, S-D, female
2-year dietary;
n = 50
Liver adenoma carcinoma, oral cavity, lung
7.3E-3
2.4E-10
NTP, 1982
Rat, Osborne-
Mendel, male
2-year gavage;
n = 50
Thyroid follicular cell adenoma, adrenal cortex adenoma
9.6E-3
3.2E-10
NTP, 2006
Rat, S-D, female
2-year gavage;
n = 53
Liver cholangiocarcinoma, hepatocellular adenoma, oral
mucosa squamous cell carcinoma, lung cystic keratinizing
epithelioma, pancreas adenoma, carcinoma
3.1E-2
1.0E-9
Kociba et al.,
1978
Rat, S-D, male
2-year dietary;
n = 50
Adrenal cortex adenoma, tongue carcinoma, nasal/palate
carcinoma
3.1E-2
1.0E-9
2! »
O ^
H r?
<*^
HH CfQ
H
W|
O
o
H
W
aBMR = 0.01.
bUF = 30; UFa = 3, UFH = 10.
-------
Table 5-17. Illustrative RfDs based on hypothesized key events in TCDD's MOAs for liver and lung tumors
Key event
Endpoint and exposure duration
NO(A)ELhed
(ng/kg-day)
LO(A)ELhed
(ng/kg-day)
BMDLhed3
(ng/kg-day)
RID1'
(mg/kg-day)
Study
Liver tumors
Changes in gene expression
CYP1A1 mRNA,
1 day
1.8E-05
3.4E-04
1.1E-010
(Appendix G)
6E-13 d
Vanden Heuvel et al.,
1994
Changes in gene expression
Benzo(a)pyrene hydroxylase (BPH)
activity (CYP1A1), 1 day
9.2E-04
6.0E-03
4.6E-040
(Appendix G)
2E-lld
Kitchin and Woods,
1979
EROD (CYP1A1), 53 weeks
none
1.4E-01
9.5E-03C
(Appendix G)
3E-10d
NTP, 2006
Oxidative stress
DNA single-strand breaks,
90 days
none
3.3E-02
2.2E-02C
(Appendix G)
7E-10d
Hassoun et al., 2000
TBARS, 90 days
-
-
4.4E-02
(Appendix G)
2E-09d
Hassoun et al., 2000
Cytochrome C reductase, 90 days
-
-
8.8E-02
(Appendix G)
3E-09 d
Hassoun et al., 2000
Hepatotoxicity
Toxic hepatopathy,
2 years
none
1.4E-01
1.8E-010
(Appendix E)
5E-09e
NTP, 2006
Hepatocyte hypertrophy, 31 weeks
9.3E-02
3.3E-01
8.8E-03
(Appendix E)
3E-10 d
NTP, 2006
Hepatocellular proliferation
Labeling index,
31 weeks
none
1.4E-01
6.6E-020
(Appendix G)
2E-09 d
NTP, 2006
Lung tumors
Metabolic enzyme induction
EROD (CYP1A1), 53 weeks
none
1.4E-01
2.9E-04
(Appendix G)
1E-Ild
NTP, 2006
Retinoid homeostatsis
Hepatic retinol and retinyl
palmitate, 90 days
none
1.1E+00
1.7E-010
(Appendix E)
6E-09 d
VanBirgelen etal.,
1995
"BMR for continuous endpoints - 1 standard deviation; for quantal endpoints - 10%.
bBolded NOAEL, LOAEL, or BMDL is selected POD; poorly-fitting BMDLs above the LOAEL not used.
°Poor BMD model fit or no good model fit.
dUF = 30; UFa - 3, UFH - 10.
eUF = 300; UFA - 3, UFH - 10; UFL - 10.
-------
1 Table 5-18. Dichotomous Hill model fits to combined adenoma and
2 carcinoma data from Kociba et al. (1987) as re-evaluated by Goodman and
3 Sauer (1992)a
4
TCDD dose metric
Background
response
EDsob
Hill
coefficient
Chi-square GOF
p-value
Administered dose
0.0193
173
0.682
0.185
Blood concentration
0.0193
57.1
0.951
0.155
Liver AhR-bound concentration
0.0193
79.2
1.365
0.270
Whole liver concentration
0.0193
2.51 x io4
0.714
0.172
5
6 "Calculations performed in S-PLUS® 6.2 for Windows®.
7 bng/kg tissue concentration except for administered dose, which is ng/kg BW per day (ng/kg-day).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-92 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 5-19. Comparison of principal epidemiological studies
2
Strengths
Weaknesses
Study
Cumulative TCDD levels in the serum were
estimated on an individual-level basis for
the 3,538 workers.
Evaluated effect of lag periods (0 and 15
years).
Measured and back-extrapolated TCDD
concentrations to refine and quantify job
exposure matrices, which were then used to
estimate dioxin cumulative dose for each
member of their entire cohort.
Internal cohort comparisons (Cox regression
model).
Background exposure estimated.
• Exposure to other chlorinated hydrocarbons
(dioxin like compounds).
• Extrapolation of dose from a small subset
(roughly 5%, n = 170) of the cohort.
• Serum fat or body fat levels of TCDD were
back-calculated using a simple first-order
model. Half-life of TCDD is variable but
simulated as a constant. Changes in the lipid
fraction of body weight or presence/absence
of genetic differences in humans that alter the
distribution and metabolism of TCDD were
not considered.
• Serum lipid levels of TCDD in 1988 were
measured only at one of the eight plants in
the study. No follow-up measures. The
estimates of dose are based on blood samples
taken decades after exposure.
NIOSH cohort
Steenland et al.
(2001)
Cumulative TCDD levels in the serum were
estimated on an individual-level basis for
the 3,538 workers.
TCCD dose estimates were simulated with a
kinetic model that included considerations
of exposure intensity and age-dependent
body weight and fat levels.
Evaluated effect of lag periods (0 and 15
years).
Background exposure estimated.
Stratified risk estimates for smoking and
nonsmoking.
Race and age adjustments.
Internal cohort noted an inverse-dose
response for high-exposure groups and thus
excluded the data resulting in better
associations.
• Extrapolation of dose from a small subset
(roughly 5%, n = 170) of the cohort.
• The authors reported the CADM model
provided an improved fit over the one-
compartmental model, but no evidence was
reported regarding any formal test of
statistical significance.
• Serum lipid levels of TCDD in 1988 were
measured only at one of the eight plants in
the study. No follow-up measures. The
estimates of dose are based on blood samples
taken decades after exposure.
• Exposure to other chlorinated hydrocarbons
(dioxin like compounds).
• No consideration for recent exposures to
TCDD, changes in the lipid fraction of body
weight or presence/absence of genetic
differences in humans that alter the
distribution and metabolism of TCDD could
cause misclassification.
NIOSH cohort
Cheng et al.
(2006)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-93 DRAFT—DO NOT CITE OR QUOTE
-------
1
Table 5-19. Comparison of principal epidemiological studies (continued)
Strengths
Weaknesses
Study
• Repeated TCDD measures in serum in 48
individuals. Used to estimate half-life for
study cohort. Took into account the age
and body fat percentage of the workers.
Measured and back-extrapolated TCDD
concentrations to quantify exposures for the
remaining cohort members using 5 different
working areas of the plant
• Evaluated effect of lag periods up to 20
years.
• Multiple statistical models used to evaluate
fatal cancer slope estimates.
• Background exposure estimated.
• Exposure to other chlorinated hydrocarbons
(dioxin like compounds), HCH, and lindane.
• Extrapolation of dose from a small subset
(roughly 4%, n = 1,189) of the cohort.
• Serum fat or body fat levels of TCDD were
back-calculated using a simple first-order
model. Presence/absence of genetic
differences in humans that alter the
distribution and metabolism of TCDD were
not considered.
• Serum lipid levels of TCDD for only 275
workers.
Becher et al.
(1998);
Hamburg
Cohort
• Both internal and external analyses.
• Adjustment for age, BMI, and smoking.
• Both cancer incidence and cancer mortality
data available, although results somewhat
discordant, with steeper dose-response seen
for cancer mortality.
• Acute dose due to accident may not be
comparable to chronic dose accumulated over
a long time, as in most environmental
exposures.
• Relatively small number of cancer deaths
compared to NIOSH and Hamburg cohorts
(n = 31).
• Serum TCDD levels measured 30 years after
accident, requiring extrapolation back in time
to estimate cumulative dose over time.
• Serum TCDD levels measured only on a
sample of the cohort (138 out of 243),
requiring assumptions about similarities in
exposure scenario for other workers to
estimate their exposure
Ott and Zober
(1996)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-94 DRAFT—DO NOT CITE OR QUOTE
-------
Table 5-19. Comparison of principal epidemiological studies (continued)
Strengths
Weaknesses
Study
• TCDD levels measured in all 891 members
of this female cohort.
• Most TCDD measurements based on
observed levels in stored serum at the time
of the accident in 1976, no extrapolation
needed to estimate past levels.
• Internal analyses.
• Evaluates female cancer incidence, other
studies evaluate male cancer mortality.
• Presumed adjustment for age and potential
breast cancer confounders (15 of 21 cancers
were breast cancer).
• Acute dose due to accident may not be
comparable to chronic dose accumulated over
a long time, which is typical of most
environmental exposures.
• Did not evaluate different lag periods.
• Not clear if any adjustment for confounders.
• Small number of cancers (n = 21).
• Doses known in 1976, require assumptions
about excretion over time to estimate
cumulative dose (9 year half life assumed),
presumed metric of primary interest. No
more recent TCDD concentration data used.
• Reported logio transformation of the exposure
estimates in their regression analysis.
Warner et al.
(2002)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-95 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
Cell-type specific
factors (kinases,
cofactors etc.)
AHR
TCDD
Ligand
Binding
Arnt
Target
Gerie
Expression
Altered
Cell Cycle
Regulation
Altered
Metabolism
Post-translational
Heterodimerization
modification
Binding to other
transcription factors
(Rb, ERa)
Binding
toDRE
Figure 5-1. Mechanism of altered gene expression by AhR. The regulation of
gene expression by TCDD in mammalian cells requires binding of the xenobiotic
to the aryl hydrocarbon receptor (AhR). The AhR is part of a multi-protein
complex that includes heat shock proteins and various kinases and other post-
translational modifying factors. Upon ligand binding, the AhR heterodimerizes
with the aryl hydrocarbon receptor nuclear translocator (Arnt) and binds to dioxin
response elements (DREs) found in target genes. Alternatives to DRE-dependent
gene expression exist whereby the AhR complex associates with other
transcription factors and results in a cross-talk between these systems. The
culmination of regulation of AhR targets genes (both increases ad decreases in
transcription) results in an alteration in cellular phenotypes, including changes in
intracellular metabolism and changes in cell cycle regulation.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-96 DRAFT—DO NOT CITE OR QUOTE
-------
§-
§•
TO
s
TO
3
m
>;*
ss»
2r
%
>4
TO
jg
TO'
86
^3
5S
§
0
WN is
i TO
\Q W
-J o
1
Su
§ §•
J> ^
Tj S
H O
. . <-*.
O
a
. Co
^ S3:
9 |
o
o
HH Oq
H <«
W|
O g?
o
H
W
Liver
TCDD
AhR
Changes in
Gene expression
Lung
TCDD
AhR
Co-carcinogens
Oxidative
Stress
Changes in
Gene expression
Hepatoxicity
Hepatocellular
proliferation
<--*
Retinoid
homeostasis
S—
i
I
*
>J y
Metabolic
Enzymes
(Cyps, COX2)
Adenoma
and
Carcinoma
Toxicity
Proliferation
i
i
i
i
~
Adenoma
and
Carcinoma
Thyroid
TCDD
~
AhR
Hepatic UGT1
i
Decreased T4
1
Increased TSH
Liver
Proliferation
Adenoma
and
Carcinoma
Thyroid
Figure 5-2. TCDD's hypothesized modes of action in site-specific carcinogenesis. See text for details. In each
instance, the solid arrows depict pathways that are well-established and are associated with low uncertainty. The
dashed arrows represent connections that are less established and are associated with higher uncertainty.
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Yes
No
Yes
No
Is the
derived OSF the lowest for
a species/sex combination
in a given study?
For each
_^^"species/s ex/tumor combination7~~~~\
were increases in tumor incidence with
dose statistically significant either pair-wise
or by a trend test?
Include as candidate OSF
Exclude as candidate OSF
Include tumor data set
Exclude tumor data set
Using the animal kinetic model, estimate blood concentrations
corresponding to average daily administered doses
Using the human kinetic model, estimate a Human Equivalent Dose (BMDLheds) for
each BMDL0i and derive a candidate OSF by dividing 0.01 by the BMDLheds
Conduct linear multi-stage dose-
response modeling using kinetic
doses and derive a BMDL01 for
each species/sex/tumor
combination within a study
Assuming independence of
tumors, apply multiple-tumor modeling
using kinetic doses and derive
a BMDL01 for each species/sex
combination within a study
Final list of key cancer animal bioassay studies for quantitative
dose-response analysis of TCDD
Figure 5-3. EPA's process to select and identify candidate OSFs from key
animal bioassays for use in the cancer risk assessment of TCDD.
For each cancer study that qualified for TCDD dose-response assessment using the study inclusion criteria,
EPA first selected the species/sex/tumor combinations with statistically significant increases in tumor
incidence by either a pair-wise test between the treated group and the controls or by a trend test showing
increases in tumors with increases in dose. Next, EPA used an animal kinetic model to estimate blood
concentrations corresponding to the study average daily administered doses for use in dose response
modeling. BMDL0i's were then estimated for the blood concentrations by, (1) using the linearized
multistage model for each species/sex/tumor combination within each study, and (2) using the linearized
multistage model within a Bayesian Markov Chain Monte Carlo framework that assumes independence of
tumors and modeling all tumors together for each species/sex combination within each study. Using the
human kinetic model, human equivalent doses (BMDLheds) were then estimated for each of the BMDL0iS
and oral slope factors were calculated by OSF = 0.01/BMDLhed- The lowest OSF for a species/sex
combination for either a single tumor type or all tumors combined was selected as a candidate OSF for
TCDD risk assessment.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-98 DRAFT—DO NOT CITE OR QUOTE
-------
Liver
TCDD
1
AhR
I
Changes in
gene ex
oression
Oxidative
stress
Hepatotoxicity***'
^•....^Hepatocellular
proliferation
CYP1A1 mRNA, 1 day
Vanden Heuvel etal., 1994
EROD (CYP1A1), 53 weeks
NTP, 2006
TBARS, 90 days
Hassoun et al., 2000
Toxic hepatopathy, 2 years
NTP, 2006
Labeling index, 31 weeks
NTP, 2006
Adenoma
and
carcinoma
Figure 5-4. Representative endpoints for each of the hypothesized key events
following AhR activation for TCDD-induced liver tumors.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-99 DRAFT—DO NOT CITE OR QUOTE
-------
Lung
Hepatic retinol, 90 days
Van Birgelen et al., 1995a
TCDD
I
AhR
Cocarcinogens
Changes in
gene ex
— Retinoid
homeostatsis
pression
Metabolic
enzymes
(Cyps, COX2)
•~Tox
city*
EROD (CYP1A1), 53 weeks
NTP, 2006
'^Proliferation
Adenoma
and
carcinoma
Figure 5-5. Representative endpoints for two hypothesized key events
following AhR activation for TCDD-induced lung tumors.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-100 DRAFT—DO NOT CITE OR QUOTE
-------
o
Multistage model fit
Linear fit
CD
o
^r
o
CM
o
o
o
1
5
10
50
100
dose (ng/kg-day)
Figure 5-6. Multistage model fit to cholangiocarcinoma response data (NTP,
2006) with comparison to linear model fit.
o
administered dose (ng/kg-day)
blood concentration (ng/kg)
AHR-bound concentration (ng/kg)
CD
o
^r
o
CM
o
o
o
0
20
40
60
80
100
Exposure measure
Figure 5-7. Weibull model fits to Kociba/G&S liver tumor response data
with alternative dose metrics. Weibull powers are 0.68, 0.95, and 1.4 for the
administered dose, blood concentration and AhR-bound concentration fits,
respectively.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 5-101 DRAFT—DO NOT CITE OR QUOTE
-------
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
6. FEASIBILITY OF QUANTITATIVE UNCERTAINTY ANALYSIS
FROM NAS EVALUATION OF THE 2003 REASSESSMENT
6.1. INTRODUCTION
This section focuses on the third area for improvement in the 2003 Reassessment that was
identified by the National Academy of Sciences (NAS) review committee (NAS, 2006a), i.e.,
improving transparency, thoroughness, and clarity in quantitative uncertainty analysis.
Although the NAS committee summarized the shortfalls in the 2003 Reassessment categorically,
the elaborations within their report often contain the qualification "if possible" and do not take a
position with regard to the feasibility of many of its suggestions. With appreciation for the
extent of information available for dioxin, the goal of this section is to circumscribe the
feasibility of a data-driven quantitative uncertainty analysis for
2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) dose-response assessment. Following brief
highlights of the evolution of quantitative uncertainty analysis for such applications, this section
lays out definitions of key terms, reviews U.S. Environmental Protection Agency (EPA)'s
position regarding cancer and noncancer endpoints, summarizes the NAS critique, and evaluates
the feasibility of quantitative uncertainty analysis for TCDD within the framework of EPA's
reference dose (RfD) and cancer dose-response methodologies.
6.1.1. Historical Context for Quantitative Uncertainty Analysis
The basic methods of probabilistic risk assessment were developed in the aerospace
program in the 1960s, and they found their first full-scale application in the Nuclear Regulatory
Commission's Reactor Safety Study of 1975—including accident consequence analysis and
uncertainty analysis (U.S. NRC, 1975). This study, commonly referred to as the Rasmussen
Report after its lead author, is considered to be the first modern probabilistic risk assessment. In
the aftermath of the 1979 Three Mile Island accident, a new generation of probabilistic risk
assessments (PRAs) appeared in which some of the methodological problems of the 1975 study
were avoided. These advances were reflected in the Commission's Fault Tree Handbook
(U.S. NRC, 1981) and PRA guide (U.S. NRC, 1983), which shored up and standardized much of
the risk assessment methodology. An extensive chapter of the latter was devoted to uncertainty
and sensitivity analysis. These documents formed the basis for standards and guidelines
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
established by other agencies, including the U.S. Department of Energy (U.S. DOE, 1992) and
National Aeronautics and Space Administration (NASA, 2002).
In 1991, a set of U.S. Nuclear Regulatory Commission (U.S. NRC) studies known as
NUREG 1150 used structured expert judgment to quantify uncertainty and set new standards for
uncertainty analysis, in particular with regard to expert elicitation (U.S. NRC, 1991). This was
followed by a joint U.S.-European Union (EU) program for quantifying uncertainty in accident
consequence models. Expert judgment methods were further elaborated in those evaluations, as
well as screening, dependence modeling and sensitivity analysis (EC, 2009). Studies building
off of this work have performed a large-scale uncertainty analysis of European consequence
models and provided extensive guidance on identifying important variables; selecting,
interviewing and combining experts; propagating uncertainty; inferring distributions on model
parameters; and communicating results, as documented by Goossens et al. (1996, 1997, 1998,
2001a-g) and others (Brown et al., 1997; Harper et al., 1995, 2002).
The National Research Council (NRC) has been a persistent voice in urging the
government to enhance its risk assessment methodology since its report on risk assessment in the
federal government (NRC, 1983). The Council's 1989 report, Improving Risk Communication,
inveighed against minimizing the existence of uncertainty and noted the importance of
considering the distribution of exposure and sensitivities in a population (NRC, 1989). The issue
of uncertainty was a clear concern in subsequent reports, including those assessing human
exposure to airborne pollutants (NRC, 1991). Building on its evaluation of Issues in Risk
Assessment (NRC, 1993), the landmark study Science and Judgment in Risk Assessment (NRC,
1994) gathered many of these themes in a plea for quantitative uncertainty analysis as "the only
way to combat the 'false sense of certainty,' which is caused by a refusal to acknowledge and
(attempt to) quantify the uncertainty in risk predictions." A subsequent report Estimating The
Public Health Benefits of Proposed Air Pollution Regulations (NRC, 2002) identified three
barriers to the broad acceptance of recent EPA health benefit analyses: (1) the large amount of
uncertainty inherent in these analyses, (2) the manner in which EPA deals with this uncertainty,
and (3) "... projected health benefits are often reported as absolute numbers of avoided death or
adverse health outcomes without a context of population size or total numbers of outcomes."
The Council encouraged EPA to "explore alternative options for incorporating expert judgment
into its probabilistic uncertainty analyses."
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
In 2006, the Office of Management and Budget released a draft bulletin proposing
technical guidance for risk assessments produced by the federal government (OMB, 2006). An
NRC (2007) review found many shortfalls in this proposal and recommended its retraction, and a
revision was undertaken. In an early 2009 report, Science and Decisions: Advancing Risk
Assessment, the NRC committee on improving risk analysis encouraged EPA to harmonize
approaches for cancer and noncancer dose-response assessment (NRC, 2009), which involves
uncertainty issues discussed in this section. Even more recently, EPA released a draft white
paper, Using Probabilistic Methods to Enhance the Role of Risk Analysis in Decision Making
(U.S. EPA, 2009d). Although not focused specifically on quantitative uncertainty analysis, there
is overlap with the issues treated here, and relevant insights are anticipated from ongoing efforts
in this area.
6.1.2. Definition of Terms
For purposes of this study, the following definitions are adopted:
Uncertainty Characterization: This consists of & Structured Uncertainty Narrative and, if
the uncertainty is supported by quantitative models, Quantitative Uncertainty Analysis.
Structured Uncertainty Narrative: This identifies assumptions conditional on which
uncertainty is to be characterized and delineates the type of arguments with supporting
evidence that buttress these assumptions.
Joint Distribution/Marginal Distribution: For a set of uncertain quantities, a joint
distribution is an assignment of probabilities (or probability densities) for each possible
combination of values of these quantities. Each uncertain quantity has a marginal
distribution, that is, an assignment of probabilities (or probability densities) to each
possible value of that quantity. Assigning a marginal distribution to each quantity is not
equivalent to assigning a joint distribution to the set of quantities, unless the quantities
are independent; in this case the joint distribution is just the product of the margins.
Quantitative Uncertainty Analysis: This is a quantification of the uncertainty attending
the use of quantitative models. It applies to a mathematical model of physical
phenomena, some of whose parameter values are not known with certainty. A joint
distribution is assigned to uncertain model parameters and propagated through the model
to yield a joint distribution over the model output. Thus, a quantitative uncertainty
analysis always has a joint distribution over model outputs as its result.
Qualitative/Informal Uncertainty Analysis: This assembles the arguments and evidence
and provides an assessment of their plausibility in terms of verbal modifiers. The
meaning of verbal modifiers such as "likely/unlikely" or "plausible/implausible" in the
natural language is indeterminate and context dependent. The way in which these
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
qualifiers combine in the natural language requires critical attention from a quantitative
viewpoint (if A is likely and B is likely and C is likely, is A and B and C likely?). Before
the advent of personal computers, various shorthand techniques were developed for
computing system risk. In control theory schemes of 'interval probabilities' were
proposed which could be propagated through a system to yield bounds on system
reliability. Whereas these bounds originally reflected accuracy of shorthand
approximations of complex formulae, their offspring have been proposed as
quantifications of uncertainty. Alternative notions of uncertainty are also proposed with
the goal of simplifying the assessment and computational burden or capturing putative
features of uncertainty which are overlooked in probability theory. These include
possibility theory, fuzzy numbers, qualitative algebra, imprecise probabilities, belief
functions, certainty factors, and the like. Nonmonotonic reasoning systems attempt to
capture reasoning about knowledge, or reasoning from partial knowledge; they include
default logic, defeasible logic, abductive logic, autoepistemic logic, to name a few. This
is not the place to discuss foundational issues, except to remark that the practitioner
should carefully explore the whole range of alternatives and critically examine the
operational meaning of the primitive notions in each alternative.
Sensitivity Analysis: If a quantitative model uses "nominal values" for various input
parameters, a sensitivity analysis is performed by choosing different values for these
parameters and re-running the model to assess the impact of changes in these parameters
on model output. Applicable methods include one- and two-at-a-time methods, design of
experiments and Morris's method. They aim at estimating first-, and perhaps
higher-order effects with a minimal number of model runs, by systematically varying the
nominal values. In large uncertainty analyses, sensitivity analysis is used to screen
variables for in-depth uncertainty quantification, and thus is part of a quantitative
uncertainty analysis. As a note, the NAS committee report does not distinguish between
uncertainty and sensitivity analysis. In fields which have not developed a tradition in
uncertainty quantification, the spread of values generated by a sensitivity analysis is
sometimes presented as a representation of uncertainty (Murphy et al., 2004). That may
or may not be the case; the question is moot so long as the uncertainty on model input
parameters is not quantified. Systematically varying input values is not the same as
sampling input parameter values from their uncertainty distributions. In any event, a
systematic approach to parameter variation is essential; simply changing a few values
happenstance and generating different output does not serve a scientific purpose, and
inevitably raises questions of selection bias. That said, if alternative values recommend
themselves, then running these through the models can help sensitize users to parameter
variations.
Cognitive Uncertainty. This concerns uncertainty regarding what is the case. This may
be conceived as uncertainty over the set of possible worlds. Uncertainty over possible
worlds may be represented as probability. Two interpretations or operationalizations of
the probability formalism are current, the objective or frequentist interpretation and the
subjective or Bayesian interpretation. These interpretations are not mutually exclusive,
as subjective probabilities can and often do track relative frequencies.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Volitional Uncertainty. This concerns uncertainty regarding what to do. In the natural
language, being unsure which course of action to choose is also called "uncertainty."
Insofar as uncertainty on the best course of action can be translated into a claim about the
state of the world, volitional uncertainty can be translated into cognitive uncertainty. A
regulatory body charged with setting a speed limit is obliged to make a decision. The
decision may be cautious or reckless, well or poorly motivated, wise or foolish; but it
cannot be true or false. Since the decision makes no claim about the state of the world, it
cannot be uncertain in the cognitive sense. The uncertainty cannot be analyzed by
sampling from some distribution. However, if the decision is based on the claim that the
chosen speed limit minimizes accidents while maintaining a prescribed traffic volume,
that claim may be uncertain and may be subjected to quantitative uncertainty analysis. A
discretionary decision of a regulatory body may entrain cognitive uncertainty, but it
becomes amenable for quantitative uncertainty analysis only when it is linked to a claim
about the state of the world.
Aleatoric/Epistemic Uncertainty. This distinction is also termed Variability/Uncertainty.
A variable whose uncertainty is aleatoric for a given population takes different, uncertain,
values for each member of the population. If its uncertainty is epistemic, it takes the
same uncertain value for all members of the population. Issues involving uncertainty and
variability or epistemic and aleatory uncertainty translate into issues of dependence, when
conducting a quantitative uncertainty analysis (see Section 6.1.3.3). In its Science and
Judgment report, NRC (1994) correctly remarks that "the amount of variability is
generally itself an uncertain parameter." It is natural to ask whether a given uncertainty
is aleatoric or epistemic, whereas it is awkward to ask whether this uncertainty is
uncertain or variable—which explains the preference for the epistemic/aleatoric
terminology.
6.1.3. Basic Requirements of a Quantitative Uncertainty Analysis
The uncertainty propagation can be performed by some rough estimation, as for example
the delta method (Oehlert, 1992) or in rare cases it can be performed analytically, as in simple
error propagation. Most often, however, it will be performed using Monte Carlo simulation. A
joint distribution is assigned to the parameters of a quantitative model and then propagated
through the model by sampling repeatedly from this joint distribution, computing model output
and generating a distribution of model output. Every uncertainty analysis is conditional on initial
assumptions. A "complete" uncertainty analysis is an unattainable goal; the best that can be
done in practice is to identify and motivate the assumptions that are used. This section is not a
how-to guide, but a to-do list of key tasks involved in any quantitative uncertainty analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
6.1.3.1. Quantitative Model
The starting point of any quantitative uncertainty analysis is a mathematical model or
prescription for calculating quantities of interest. A structured narrative explains the choice of
quantitative models. If some values of input parameters for this calculation are not known with
certainty, then the question arises: "What is the uncertainty attending the use of this model?"
This is the question a quantitative uncertainty analysis seeks to answer.
6.1.3.2. Marginal Distributions over Model Parameter
If the model parameters are directly measurable with sampling error, then the sampling
distribution may itself be used in the quantitative uncertainty analysis. If the model parameters
are fit to data that are sampled from a known or hypothesized distribution, then by resampling
this distribution and refitting the model, distributions over the model parameters may be
constructed. Physically-based simulation models, such as pharmacokinetic models or
environmental transport models, may be solved analytically if equilibrium reaction rates (the
transfer coefficients) are constant. If these rates are not constant, as when concentrations are
near saturation levels, then simulation is indicated. Structured expert judgment has been applied
for uncertainty quantification within the engineering community since the time of the Rasmussen
Report. More recently, this approach has been "test-driven" by EPA in assessing health effects
of fine particulates (Walker et al., 2009), and its potential application has been identified in the
Agency's Guidelines for Carcinogen Risk Assessment, commonly referred to as the cancer
guidelines (U.S. EPA, 2005).34
6.1.3.3. Dependence Between Parameter Uncertainties: Aleatoric and Epistemic (Uncertainty
and Variability)
Two uncertain quantities are independent if knowledge about one of them does not alter
our uncertainty regarding the other. The quantities are dependent if they are not independent.
The role of dependence modeling in quantitative uncertainty analysis must be addressed. To
illustrate, cigarette smoking and body fat are both thought to influence biomarkers for toxic
34The U.S. EPA (2005) cancer guidelines state: "In many of these scientific and engineering disciplines, researchers
have used rigorous expert elicitation methods to overcome the lack of peer-reviewed methods and data.... These
cancer guidelines are flexible enough to accommodate the use of expert elicitation to characterize cancer risks, as a
complement to the methods presented in the cancer guidelines. According to NRC (2002), the rigorous use of
expert elicitation for the analyses of risks is considered to be quality science."
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
response to dioxin exposure, such as ethoxyresorufin-(9-deethylase (EROD) activity (Pereg et al.,
2002). Both cigarette smoking and percent body fat in a random individual sampled from a
target population are uncertain.35 However, these uncertainties are not independent, inasmuch as
smokers tend to have less body fat (Vanni et al., 2009).
Issues involving uncertainty and variability or epistemic and aleatory uncertainty
translate into issues of dependence, when conducting a quantitative uncertainty analysis. For
example, a constant used to estimate the biokinetic behavior of dioxin may be uncertain. If it is
believed to be the same for every member of the population, the uncertainty is termed
"epistemic." In a quantitative uncertainty analysis, this factor would be sampled from its
uncertainty distribution on each Monte Carlo run and used for all members of the population.
Body fat, in contrast, is aleatoric. We do not sample one value from the body fat distribution and
use this value for all members of the population on each Monte Carlo run. Instead we sample a
body fat value for each individual on each run. Because body fat is correlated with other
relevant variables (e.g., smoking, sex, age, and socioeconomic status), all of these variables
should be sampled in a manner that reflects their dependences. Kinetic constants whose
uncertainty is epistemic are completely correlated across individuals: if the value is 0.5 for one
individual, it is 0.5 for everyone. Body fat values vary from individual to individual, and they
are correlated through a host of other variables.
6.1.3.4. Model Uncertainty
All models, being idealizations, are false; on this there is no uncertainty to quantify.
However, the choice of model may constrain the ability to represent uncertainty in observable
phenomena. Thus, in a linear low-dose model, uncertainty over a cancer slope factor may be
quantified, but uncertainty regarding changes in slope at distinct low-dose regimes cannot be
captured. When the model choice imposes severe and potentially unwelcome constraints on
uncertainty quantification, this must be addressed. Distributions over model parameters may be
35Because dioxins generally distribute to body fat/lipid, the percent body fat is often used to estimate body burden; a
default value of 25% is common (Connor and Aylward, 2006). However, body fat percentage varies widely
between individuals, from a minimum essential level (e.g., 2% for men, 10% for women) to obesity (e.g., 38% or
more for men, 42% for women). Considering that current estimates suggest 30% of the U.S. population are obese,
an uncertainty analysis of dioxin risk in this population should sample individuals from their gender/body fat
distribution and correlate this with other known or suspected covariates influencing toxic response (such as diet,
smoking, natural and endogenous ligands, disease, and age).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
selected and evaluated based on their ability to reflect uncertainty distributions over observable
phenomena predicted by the models. In such cases, the uncertainty propagated through the
quantitative model is not strongly model-dependent. In other cases, multiple model alternatives
may be applied, whose "probability of being the true model" is known or assumed. Since
different models can always be regarded as specializations of more general models, the
distinction between parameter and model uncertainty is sometimes more apparent than real. The
multistage and Weibull dose-response models both contain the model Pr(x) = y + (1 - y)
(1 - e ^l x) as a sub model, to which they collapse if other parameters are zero (multistage) or one
(Weibull). Recalling that the function 1/(1 + x) is first order equivalent to (1 - x) for small x, the
same may be said for logistic models as well. In this case, these models could easily be
parameterized within one family, rendering the choice between them a choice of parameter
values. Similarly, the choice between sub-, supra-, and linear models is sometimes reduced to
parameter estimation within a more general class of model (Hoel and Portier, 1994).
In other cases, the reduction of model uncertainty to parameter uncertainty is less natural.
For example, according to the "chemoprotection model" of Greenlee et al. (2001), dioxin's
binding to the aryl hydrocarbon receptor (AhR) inhibits proliferation in tumor cells and thus
suppresses mammary tumors. Dose-dependent protection and cancer induction can both be true,
each applying to different tissues. Although mathematical models exhibiting these twin features
are under development (Kohn and Melnick, 2002), these models are not yet readily estimable
from data, and the choice between them is referred to the structured narrative.
6.1.3.5. Sampling Method
All sampling on a computer is "pseudo random." Significant issues arise in choosing a
method for sampling high-dimensional distributions with dependence. If evaluating the
quantitative model is very time consuming, various 'quasi random' schemes may be applied,
including Latin hypercube sampling, importance sampling, and Hammersley sampling. These
methods involve trade-offs between economy and accuracy of the dependence modeling.
6.1.3.6. Method for Extracting and Communicating Results
When a large quantitative uncertainty analysis has been performed, the method for
identifying important contributors and communicating this information to users is not
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
straightforward. Analysts have proposed many ways to quantify the uncertainty contribution of
one variable, or set of variables, on others,36 and his/her decision at this juncture may strongly
impact the "take-home" message from the study. An importance measure that averages over an
entire sample space may obscure the features of real interest. For example, the drivers of cancer
induction at low doses might be different from the drivers at high doses. If the drivers of
low-dose cancer induction are of interest, then importance measures that average over all doses
should not be considered.
6.2. EPA APPROACHES FOR ORAL CANCER AND NONCANCER ASSESSMENT
EPA typically develops different types of toxicity information in its oral cancer and
noncancer dose-response assessments, although efforts to harmonize these approaches continue
to be made. Noncancer endpoints are usually assessed using the RfD methodology to derive a
threshold below which there is likely no appreciable risk. (Note "risk" is used here simply as a
general term indicating the potential for adverse effects.) In contrast, cancer endpoints are
commonly assessed using a dose-response function with the probability of excess risk above
background modeled as a linear function of dose, for doses down to zero. The RfD method
relies on a point of departure (POD), and the dose-response method uses a POD if the linear
model is chosen by default. From the U.S. EPA (2005) guidelines, cancer endpoints can also be
assessed using the RfD methodology if the proof burden is satisfactorily met (previously
described in Section 5.2.3.3).
Toxicity reference values have generally been derived for human noncancer endpoints
based on a no-observed-adverse-effect level (NOAEL) or lowest-observed-adverse-effect level
(LOAEL) from animal bioassay studies. This terminology suggests a biological population
threshold beneath which no harm is anticipated. Reference values such as the RfD or reference
36A few examples may suffice. The standard Pearson correlation coefficient measures the linear dependence
between two variables, positive or negative. The rank or Spearman correlation coefficient measures the monotone
dependence. The correlation ratio measures the (unsigned) variance contribution of an explanatory variable on a
target variable. The regression coefficient measures the expected change in standard (not natural!) units of a target
variable, per standard unit change in an explanatory variable and assumes this expected change is independent of the
values of the explanatory variables. Multiple correlation measures the correlation between a given variable and its
best linear predictor based on another set of variables. The partial correlation of two variables given a set of other
variables is their correlation after discounting the influence of the other variables. The correlation ratio, multiple
correlation and the regression coefficient are not symmetric, the correlation multiple regression are always
non-negative (Saltelli et al., 2000; Kurowicka and Cooke, 2006).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
concentration are derived by applying uncertainty factors (UFs) to a POD. Depending on the
nature of available data and modeling choice, a POD can be selected from values other than a
NOAEL or LOAEL, such as an EDX (effective dose eliciting x percent response), or a benchmark
dose (BMD) or its lower confidence bound (BMDL). The BMD is the dose that induces a
benchmark response (BMR), which is often chosen to represent a 5 or 10% increase in excess
risk above background. The POD is divided by one or more uncertainty factors that represent
knowledge gaps (see Section 6.4.1.2 for details on specific types of UFs).
An RfD is described as "likely to be without appreciable risk" but the probabilistic
language has not as yet been operationalized. There is no quantitative definition of
"appreciable" and no mechanism to compute risk as a function of dose, so as to ascertain that the
risk is indeed not appreciable. In addition, different participants in discussions over
threshold/nonthreshold models for dioxin may have different perspectives regarding how to
define "appreciable risk." Under the current POD/UF framework, dose-response functions are
not provided for calculating the actual risk at or above the RfD. Instead, to provide a "risk
indicator" for use in screening for health hazards, a hazard quotient (HQ) is computed as the
ratio of a given exposure to the RfD, or a margin of exposure (MOE) is estimated as the ratio of
the POD to the human exposure level.
For the cancer endpoint, an oral cancer slope factor may be derived for human health risk
assessment, typically based on tumor incidence data from an animal bioassay or on cancer
incidence or deaths from an epidemiologic study. In U.S. EPA's Cancer Guidelines, cancer is
predominantly thought to have no population biological threshold and a linear extrapolation to
zero is applied from the POD based on extra risk above background, i.e., the probability of the
endpoint decreases linearly in dose from the POD to zero or to a population background level. In
the absence of sufficient evidence supporting low-dose cancer induction, the linear model is
applied as a default. Cancer endpoints can also be evaluated using a "nonlinear" model. In this
case, the proof burden clearly rests on the nonlinear model; it must have a preponderance of
evidence to override the health-protective default choice, as described in U.S. EPA Cancer
Guidelines. U.S. EPA (2005) cancer guidelines state, "When adequate data on mode of action
provide sufficient evidence to support a nonlinear mode of action for the general population
(emphasis added) and/or any subpopulations of concern, a different approach—a reference
dose/reference concentration that assumes that nonlinearity—is used." In current terminology,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
the RfD methodology applies to the cancer endpoint if there is sufficient evidence supporting a
"zero slope at zero" model; otherwise, the linear nonthreshold model applies by default. (See
Section 5.2.3.3 for a detailed discussion of linear vs. nonlinear extrapolations below the observed
data, population vs. individual thresholds, and for the how U.S. EPA (2005) Cancer Guidelines
are applied in choosing dose-response model forms for risk assessment.)
6.3. HIGHLIGHTS OF NAS REVIEW COMMENTS ON UNCERTAINTY
QUANTIFICATION FOR THE 2003 REASSESSMENT
The NAS (2006a, b) identified a number of uncertainty characterization issues for the
2003 Reassessment; key sources of uncertainty for which quantification is suggested are
highlighted in Table 6-1. The discussion in this section focuses on comments related to dose
response.
There are several nuances in the NAS position relative to the need for substantial
improvement in transparency, thoroughness, and clarity in quantitative uncertainty analysis for
the 2003 Reassessment. These nuances concern whether the nonlinear model (note that the NAS
committee uses "sublinear" and "nonlinear" interchangeably) is scientifically better supported
than the linear model, and if the sublinear model is better supported, whether this is based on
data or on apodictic knowledge (knowledge without uncertainty) of the mode of action (MO A).
The NAS committee does not distinguish between individual and population dose-response
models; however the criteria from U.S. EPA Cancer Guidelines clearly apply to population
models. Assuming that the AhR-mediated MOA implies a threshold for each individual, the step
to a population "zero slope at zero" model requires the following, as identified and discussed in
detail in Section 5.2.3.3:
1. The distribution the individual thresholds induced by the MOA, and
2. The dose-response function for values above the thresholds.
This information can either come from data, or from apodictic knowledge of the MOA,
but the burden of proof clearly rests on the nonlinear model. This section summarizes the NAS
committee's overall positions. Responses to specific suggestions are given in Section 6.4 and
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-11 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
summarized in Section 6.5. Several excerpts of specific comments from NAS (2006a) illustrate
key issues.
The NAS committee favors the nonlinear model with a threshold:
.. .the committee concludes that, although it is not possible to scientifically prove
the absence of linearity at low doses, the scientific evidence, based largely on
mode of action, is adequate to favor the use of a nonlinear model that would
include a threshold response over the use of the default linear assumption.
(p. 126)
The committee does not state whether the threshold applies to the population, or whether each
individual has his/her own threshold.
The NAS also comments on whether the nonlinear model should be used to compare with
the health-protective linear default:
Because the committee concludes that the data support the hypothesis that the
dose-response relationship for dioxin and cancer is sublinear, it recommends that
EPA include a nonlinear model for cancer risk estimates, but that EPA also use
the current linear models for comparative purposes, (p. 12)
The committee does not suggest what the (population/individual) threshold might be, nor how it
might be supported on the basis of data. Rather, the apodictic knowledge that there is a
(population/individual) threshold places the dioxin risk assessment within the RfD framework,
using a HQ or MOE as the basis for indicating the potential risks from exposure. The committee
further asks for a quantitative characterization of the range of uncertainty:
The committee determined that the available data support the use of a nonlinear
model, which is consistent with receptor-mediated responses and a potential
threshold, with subsequent calculations and interpretation of MOEs. EPA's sole
use of the default assumption of linearity and selection of ED0i as the only POD
to quantify cancer risk does not provide an adequate quantitative characterization
of the overall range of uncertainties associated with the final estimates of cancer
risk. (p. 17)
Regarding the preponderance of evidence, the committee indicates that quantitative
evidence will not decide the linearity/nonlinearity (nonthreshold/threshold) issue, but knowledge
(without uncertainty) of the MOA should be used:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Quantitative evidence of nonlinearity below the POD, the ED0i, will never be
available because the POD is chosen to be at the bottom end of the available
dose-response data. ... EPA should give greater weight to knowledge about the
mode of action and its impact on the shape of the dose-response relationship.
(p. 126)
The comment continues, with the committee implicitly acknowledging that there is no
evidence arguing against linearity, but that the lack of evidence should not justify using the linear
model:
The committee considers that the absence of evidence that argues against linearity
is not sufficient justification for adopting linear extrapolation, even over a dose
range of one or two orders of magnitude or to the assumption of linearity through
zero, which would not normally be applied to receptor-mediated effects, (p. 126)
Whereas EPA has applied its own guidance on cancer risk assessment and adopted
linearity as a health-protective default in the absence of sufficient evidence (volitional
uncertainty), the NAS committee views the absence of evidence as imposing a burden of proof
on the linear model (cognitive uncertainty). (As a note, terminology issues and conceptual
inconsistencies within the NAS report illustrate the importance of clarity and are not unexpected,
given the complexity of these issues and the nature of a committee process.)
The thrust of the NAS remarks regarding transparency, thoroughness and clarity in
quantitative uncertainty analysis relevant to dose-response can be summarized as follows:
1. The uncertainty of cancer risks due to dioxin exposure should be quantified.
2. Dioxin cancer risk should be treated either as a threshold phenomenon, thus following the
basic RfD methodology, or should be modeled using a sublinear dose-response function
below the observed data, with the linear model used for comparison.
3. The POD should be subjected to quantitative uncertainty analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
A similar point of view has been indicated by others.37 Detailed suggestions regarding specific
improvements for quantitative uncertainty analysis in the 2003 Reassessment are outlined in the
next section and summarized in Section 6.5.
6.4. FEASIBILITY OF CONDUCTING A QUANTITATIVE UNCERTAINTY
ANALYSIS FOR TCDD
This section focuses on uncertainty analysis for TCDD dose response, which involves a
range of issues as highlighted in Table 6-1.
6.4.1. Feasibility of Conducting a Quantitative Uncertainty Analysis under the RfD
Methodology
This discussion applies to all noncancer endpoints of TCDD, and to cancer endpoints
insofar as they fall under the RfD methodology. An RfD is obtained through the following steps
1. Choose a POD, then
2. Apply uncertainty factors (UFs) to account for knowledge shortfalls.
The method of uncertainty factors harkens back to the engineering practice of safety
factors (Lehman and Fitzhugh, 1954). To illustrate, if the reference load for an engineered
structure is X, then engineers might design the structure to withstand load 3X, using a safety
factor of 3 to create a margin of safety. If the structure functions in a corrosive environment,
another factor could be multiplied to guarantee safety for that condition, and another factor could
be applied for heat, another for vibrations, and so on. The choice of values is simply based on
good engineering practice, i.e., reflecting what has worked in the past. Although safety factors
are still common in engineering, they are giving way to probabilistic design in many
applications. The reason is that compounding safety factors quickly leads to overdesigning.
Compounding safety margins for spaceflight systems may render them too heavy to fly. As our
understanding of a system increases, it becomes possible to guarantee the requisite safety by
leveraging our scientific understanding of the materials and processes. That of course requires
37For example, from Popp et al. (2006): "Overall, the evidence indicates that (1) TCDD causes cancer via a
receptor-mediated process; (2) this dose-response is non-linear; and (3) a threshold region exists for TCDD-induced
cancer below which adverse effects are unlikely to occur."
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
formulating clear probabilistic safety goals and developing the techniques to demonstrate
compliance.
The engineering community has never sought to account for uncertainty by treating
safety factors as random variables and assigning them (marginal) distributions; such an approach
would not counteract the overdesigning inherent in safety factors. Many authors, including the
recent national committee for Science and Decisions (NRC, 2009), have advocated just such a
probabilistic approach to the apparent "overdesigning" of the RfD when multiple UFs are used in
its derivation.
The NAS committee that evaluated the 2003 Reassessment does not discuss how to
perform uncertainty analysis. But their call for substantial improvement in quantitative
uncertainty analysis with TCDD falling under the RfD framework entails examining the
feasibility of quantitative uncertainty analysis within this framework. (Note that the EPA
Integrated Risk Information System database uses uncertainty factors without probabilistic
interpretations; some context for this is offered in Section 6.4.1.2.)
6.4.1.1. Feasibility of Conducting a Quantitative Uncertainty Analysis for the Point of
Departure
The POD plays a role in both the noncancer RfD methodology and the cancer
dose-response methodology. The POD can be selected from various options, such as a NOAEL
or LOAEL, a BMDL, or an EDX. The feasibility of quantitative uncertainty analysis for each of
these three options is considered below.
As described by Swartout et al. (1998), "The NOAEL is the highest of the tested doses in
a toxicological experiment that is judged not to have caused an adverse effect" (with dose
expressed as a dose rate, in mg/kg-day). A quantitative uncertainty analysis for a NOAEL or
LOAEL encounters the following problem. In an experiment involving a small, positive
response, the probability of seeing no response can be calculated using a binomial model with
the number of exposed animals and the observed number of responses. However, in an
experiment with no response, the probability of having observed a response cannot be calculated
without assuming a response probability. Such an assumption could not be based on observed
data. If the analysis is restricted to experiments showing a positive response, the results will be
biased. The probability of a higher NOAEL or higher LOAEL can be computed, but not that of a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
lower NOAEL or LOAEL. In other words, the probability that an experiment with a positive
result may have yielded a null response can be estimated, but not the probability that an
experiment with a null response might have yielded a positive response.
In addressing uncertainty quantification for a BMDL or EDX, two questions must be
distinguished regarding the response:
1. What is the distribution of possible doses that causes an x% increase over background?
2. What is the distribution for possible values of increase over background caused by a
given dose?
The BMD is defined as the dose that realizes a BMR. It is an estimate from bioassay data
that requires choosing a BMR and fitting a dose-response curve. The BMR, being a choice, is
not amenable to quantitative uncertainty analysis, but the choice can be motivated in a structured
narrative. The BMDL is the lower confidence limit (e.g., 5%) that can be found based on the
uncertainty in the parameters of the dose-response relationship. Thus, the BMDL is addressed to
the first question above, and represents in this case the 5% lower confidence band of the
distribution of possible doses causing an x% increase over background. In the standard
approach, the uncertainty captured by the BMDL is sampling uncertainty conditional on the truth
of the dose-response model. Different models might fit the data equally well yet lead to different
BMDLs.
The BMDL is also influenced by the constraints imposed on the parameter fitting.
Suppose that the slope is expected to be greater than one, and that the maximum likelihood
estimate of the slope is slightly greater than one. Since the constraint is not binding, the
constrained and unconstrained model would have the same Akaike Information Criterion and
would be equivalent in this sense. However, computing the BMDL with the slope constraint can
lead to very different values than without this constraint. In the latter case, slope values less than
one contribute to the uncertainty in the dose causing the BMR (see Cooke, 2009).
The EDX can also be taken as a POD. It is similar in spirit to the BMD; however, as used
here, the term EDX applies when the dose causing an x% extra risk over background has actually
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
been observed, not estimated from a fitted dose-response model.38 The observations are subject
to sample fluctuations, and if the experiment on which the EDX is based were repeated, different
values might be found. It is helpful to consider a numerical example. Suppose a background
response rate of 10% is established based on many observations of nonexposed individuals. In a
given experiment, involving say 100 individuals given dose 14 individuals responded. The
percent increase x over background (extra risk) is found by solving 14/100 = 10/100 + x
x 90/100, or x = 4.4%. We conclude that d = ED4 4. We may assume that if the experiment
were repeated with 100 new individuals sampled independently from the whole population, the
response would be given by a binomial distribution with parameters (14, 100). The number of
responses might be greater or smaller than four, there is a 16% chance of observing 10 or fewer
responses. The response to dose d would not be distinguished from the background in that case,
and a higher dose would be used for the POD.
The uncertainty analysis of EDX as the POD involves addressing the second question
above, without a quantitative dose-response model. A quantitative uncertainty analysis is
hampered, however, by the possibility that dose d would produce a response less than or equal to
the background, in which case the POD is indeterminate—another experiment with a different
dose would be chosen as the POD. A true quantitative uncertainty analysis of EDX as the POD
would thus require a full bioassay experimental design, with binomial sampling of response rates
at each dose level in the assay. Absent that, quantitative uncertainty analysis is not possible.
The interplay of choice and estimation ingredients in the POD depends on the type of
POD. The main features of the above discussion are captured in Table 4-2. This table notes that
the BMDL captures the uncertainty caused by sampling fluctuations given that the dose-response
model is true. Other methods are available to compute the BMDL using (1) model-independent,
observable uncertainty; (2) nonparametric Bayesian dose-response models; or (3) Bayesian
model averaging (Cooke, 2009). Only the EDX can be subject to a quantitative uncertainty
analysis, and then only if a full bioassay data set is available.
38This definition of EDX is adopted to distinguish the modeled response (BMD) and the observed response (EDX),
and it is more restrictive than usages common in the literature.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
6.4.1.2. Feasibility of Conducting a Quantitative Uncertainty Analysis with Uncertainty
Factors
Uncertainty factors are chosen based on a structured narrative characterizing knowledge
shortfalls involving
1. Interspecies extrapolation, (UFA: from animal data to humans).
2. Intraspecies extrapolation (UFH: to account for human interindividual variability,
considering sensitive subgroups).
3. LOAEL to NOAEL extrapolation (UFL: to estimate the dose corresponding to no adverse
effect from a reported LOAEL).
4. Subchronic to chronic extrapolation (UFS: to estimate effects of chronic exposures from a
subchronic study).
5. Database deficiency (UFD: to extrapolate from an incomplete data set, e.g., in terms of
endpoints assessed or study design, i.e., from a poor to a sufficient or rich data context).
The standard chronic RfD can represent a sensitive human (H) response to a toxic
substance under chronic (C) exposure conditions. Suppose a BMDL POD were based on animal
(A) data from a subchronic (S) study. The database for that chemical could be rich (R), e.g.,
involving multiple (and at least sensitive) species/strains, both genders, multiple life stages, with
multiple endpoints observed under sound study designs. Or the data could be poor (P), with
limited measurements from only a subchronic animal study (ASP) forming the basis for
estimating a general reference value for humans (including sensitive subgroups) under chronic
exposure conditions. For that case, the UF method would be applied as follows
A VP
RfD = — (Eq. 6-1)
UFaxUFsxUFdxUFh
where UFA, UFS, UFD, and UFH are the uncertainty factors for extrapolating from animals to
humans (UFA), subchronic to chronic exposure conditions (UFS), without adequate endpoint
coverage (UFD), and considering sensitive human subpopulations (UFH). It is possible to assign
distributions to the UFs in Eq. 6-1, and to perform a Monte Carlo analysis to produce a
quantitative uncertainty distribution over the dose or value likely to be without appreciable risk
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
to humans for chronic exposures. Many authors have proposed such an approach,39 and the
recent NRC (2009) report on science and decisions emphatically counsels EPA to move in this
direction.
The idea of using a Monte Carlo analysis to develop quantitative uncertainty distributions
for the RfD is simple, but the data on which the UFs are based and the assumptions that would
need to be made should be further explored. For example, it is assumed that the extrapolation
from subchronic to chronic exposure (UFS) is the same whether applied to animals or humans,
and whether applied to sufficient (rich) or deficient (poor) data contexts. Swartout et al. (1998)
write "Within the current RfD methodology, UFS does not consider differences among species,
endpoints, or severity of effects; the same factor is applied in all cases." In addition, due to the
paucity of relevant human data, the same authors suggest the use of other endpoints as surrogates
in estimating the extrapolation from animals to humans, UFA. Further, few data exist in humans
to accurately portray the interindividual variability in humans represented by UFH. Much of the
data gathered to date on distributions of UFs have aggregated across other extrapolations; that is,
data from subchronic to chronic ratios are aggregated over different species and different data
contexts. Finally, it may be noted that an important issue is the data on which empirical
distributions of response ratios are based. Brand et al. (1999, 2001) studied the sampling
behavior of response ratios and raise concerns with regard to their informativeness.
Detailed analyses of the data underlying a Monte Carlo uncertainty analysis of Eq. 6-1
would afford the possibility of verifying at least some of the assumptions and numerical
estimations such an analysis must make. Even if the assumption that the same UFS is applicable
for all species, endpoints, and effect severities is thought to be biological plausible, the question
of whether a given set of chemicals reflect this assumption, and hence are suitable for Monte
39There has been considerable work on giving a probabilistic interpretation of the UFs, including by Abdel-Rahman
and Kadry (1995), Vermeire et al. (1999), Baird et al. (1996), Swartout et al. (1998), Slob and Pieters (1998), Evans
and Baird (1998), Calabrese and Gilbert (1993), Calabrese and Baldwin (1995), Hattis et al. (2002), Kang et al.
(2000), and Pekelis et al. (2003). These evaluations can be considered to frame what might be called a random
chemical approach. Several authors adduce properties based on log normal distributions. Insightful studies by
Kodell and Gaylor (1999; Gaylor and Kodell, 2000) suggest that uncertainty factors are independent log normal
variables. Combining uncertainty factors involves multiplying the median values, and combining the "error factors"
according to the formula KSxH = exp[1.6449 x V(as2 +
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Carlo analysis of Eq. 6-1, can only be decided by data evaluation. Data are the ultimate arbiter
of whether quantitative uncertainty analysis with uncertainty factors, as currently envisioned, has
sufficient evidentiary support.
6.4.1.3. Conclusion on Feasibility of Quantitative Uncertainty Analysis with the RfD
Approach
A quantitative uncertainty analysis of the POD is not feasible for PODs based on
NOAELs or LOAELs. For the BMDL, such an analysis is not appropriate because the BMDL is
already a quantile from an uncertainty distribution of the BMD. However, this uncertainty
distribution can be obtained in different ways that capture different aspects of uncertainty.
Quantitative uncertainty analysis is feasible if the POD is based on the EDX (as defined above)
and is supported by a full set of bioassay data. A quantitative uncertainty analysis based on a
probabilistic interpretation of uncertainty factors in their present form invokes strong
assumptions. The data on which the distributions of uncertainty factors are based could be used
to check at least some of these assumptions.
6.4.2. Feasibility of Conducting a Quantitative Uncertainty Analysis for TCDD Under the
Dose-Response Methodology
Quantitative uncertainty analysis starts with a mathematical model and seeks to quantify
the uncertainty attending the use of this model. Dose-response relations are mathematical
models expressing the probability of response as a mathematical function of dose. For several
decades, the uncertainty attending the use of dose-response models has been an abiding concern
in many sectors, including the chemical and nuclear industries as well as the public health sector.
Given a set of animal bioassay data, quantifying dose-response uncertainty may be approached in
different ways. The differences reflect different types of uncertainty that are captured. A recent
evaluation enumerates the following possible methodologies (Bussard et al., 2009):
Benchmark Dose Modeling (BMD): Choose the 'best' model, and
assess uncertainty assuming this model is true. Supplemental results can compare
estimates obtained with different models, and sensitivity analyses can investigate
other modeling issues.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-20 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Probabilistic Inversion with Isotonic Regression (PI-IR): Define
model-independent 'observational' uncertainty, and look for a model that captures
this uncertainty by assuming the selected model is true and providing for a
distribution over its parameters.
Non-Parametric Bayes (NPB): Choose a prior mean response (potency)
curve (potentially a "non-informative prior") and a precision parameter to express
prior uncertainty over all increasing dose-response relations, and update this prior
distribution with the bioassay data.
Bayesian Model Averaging (BMA) (as considered here): Choose an
initial set of models, and then estimate the parameters of each model with
maximum likelihood. Use classical methods to estimate parameter uncertainty,
given the truth of the model. Determine a probability weight for each model
using the Bayes Information Criterion, and use these weights to average the model
results.
The first of the above methods involves standard classical statistical methods and
captures sampling uncertainty conditional on the truth of the model used. The other methods are
"exotic" in the sense that they attempt to capture uncertainty that is not conditional on the truth
of a given model. All have been subjected to peer review and published, but they do not enjoy
the market penetration of the standard classical methods. The Bayesian models involve
subjective choices of prior distributions. Insofar as the final result is largely independent of the
choice of prior, these methods conform to the current starting point of focusing on data-driven
methods and not appealing to structured expert judgment. (Structured expert judgment can also
be considered an exotic method; an explanation of this approach falls outside the scope of this
report.)
It should be emphasized that the four approaches above were illustrated as part of an
integrated bench-testing exercise using simplified data for demonstration purposes. Application
to actual, complicated data sets would provide additional insights into the feasibility of each per
the nature of those data.
Many steps are involved in arriving at such a starting point, to frame the extrapolation of
data from animal bioassays to human reference values together with consideration of
epidemiological data from studies of workers (routine exposures) or the general public (including
dietary exposures and those via discrete poisonings or accidental releases). Major issues are
summarized below.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-21 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
6.4.2.1. Feasibility of Quantitatively Characterizing the Uncertainties Encountered when
Determining Appropriate Types of Studies (Epidemiological, Animal, Both, and
Other)
The risk assessor must choose the data set(s) that will serve as a starting point for
dose-response modeling. With respect to TCDD, a wealth of animal bioassay data exist in the
scientific literature, across species ranging from rats, mice, guinea pigs, and hamsters to mink,
dogs and monkeys, and a variety of tissues, organs, and systems. In addition, there are historical
human data such as from clinical/case reports, accidental releases, and occupational exposures,
including epidemiological data for several cohorts. Historical data are usually attended with
large uncertainties regarding the doses actually received by individuals. Both clinical/case
reports and epidemiological data pose issues with regard to possible confounders. Different
TCDD studies have assessed various endpoints, ranging from chloracne to hepatic enzyme
induction, abnormal fasting glucose, diabetes, lung and cardiovascular effects, neurological
effects, developmental delay, dental defects, sex ratio, epididymal sperm count, endometriosis,
non-Hodgkin's lymphoma, susceptibility to infection, depression, hostility, and anger, among
others. Confounders that have been evaluated in the epidemiological studies include gender,
body mass index, age, cigarette and alcohol consumption, and hair and eye color (Eskenazi et al.,
2002a, b; Pereg et al., 2002; Baccarelli et al., 2005, 2006).
There is disagreement in the literature over the nature, scope, and quality of the historical
and evolving epidemiological data for TCDD. Popp et al. (2006) put forth a position similar to
that taken by the NAS:
According to IARC [1997], the strongest overall evidence for the carcinogenicity
of TCDD is for all cancers combined, rather than for any specific site. The
relative risk for all cancers combined in the most highly exposed and
longer-latency subcohorts is 1.4. Although IARC indicated that this relative risk
does not appear likely to be explained by confounding, relative risks of this
magnitude for studies of other substances, particularly with a lower confidence
interval at or near one, are generally found to be the result of confounders. Few
examples (perhaps only smoking and ionizing radiation in atomic bomb
survivors) exist of agents known to cause an increase in cancers at many sites.
This lack of precedent for a multi-site carcinogen without particular sites
predominating, combined with the very small excess relative risks, means that the
epidemiological findings must be treated with caution.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
On the other hand, U.S. EPA Cancer Guidelines express a clear preference for
epidemiological studies over animal data. The question here is whether quantitative uncertainty
analyses based on bioassay data and different epidemiological data can be combined in some
overall uncertainty assessment. Diverse studies are sometimes combined in meta-analyses.
Different studies have different protocols, and ignoring these differences can distort results. To
illustrate, placental EROD activity was 4-fold higher in a group of highly poly chlorinated
biphenyl (PCB)-exposed Inuit women compared to a group of women from southern Quebec
with lower exposure. However the differences in EROD activity disappeared when smoking was
taken into account. Had tobacco consumption been unknown in one of the groups, spurious
conclusions would have been drawn (Pereg et al., 2002). Can uncertainty arising from
combining studies with different protocols be taken into account in quantitative uncertainty
analysis? The question is similar to that of accounting for uncertainty due to missing covariates
in Cox regression (see Section 6.4.2.2).
Standard statistical tools will be of no avail, as they quantify uncertainty in model
parameters estimated from data. However, exotic methods, such as Bayesian methods,
probabilistic inversion, or structured expert judgment may be applicable. These methods can be
applied when a quantitative model predicts other phenomena, even though these phenomena
could not be used to estimate the model. The question of whether such methods could remain
sufficiently tethered to data, or whether structured expert judgment is unavoidable, is a subject
for future research.
6.4.2.2. Uncertainty in TCDD Exposure/Dose in Epidemiological Studies
There are many types of epidemiological studies including case controls, longitudinal and
cross-sectional cohorts, and simple exposed/unexposed studies. The variety of mathematical
models includes Cox proportional hazard, Poisson regression, and logistic regression. The
outputs vary accordingly: relative risk ratios, odds ratios, and standardized mortality ratios
(SMR, ratio of observed to expected deaths). Exposure uncertainties often concern back-casted
exposures based on current serum lipid concentrations, estimated/self reported dietary habits,
fish consumption, placenta lipid concentrations, and other measures.
Uncertainty in exposure is often compensated by coarsening the output, e.g., SMRs
instead of dose-dependent risk factors. Packages computing the outputs routinely produce
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
confidence intervals that reflect sampling fluctuations, assuming truth of the model. Additional
uncertainty could be factored in with exotic methods. A significant issue in epidemiological
studies is the effect of omitted covariates. Omitted covariates in Cox regression will bias the
estimates of effects of included covariates. If the omitted covariates are independent of the
included covariates, the bias is toward zero in absolute value (Bretagnolle and Huber-Carol,
1988); if the omitted covariates are not independent, little can be said.
With regard to individual studies, it might be possible to identify specific opportunities
for uncertainty quantification. This is illustrated with the study of Steenland et al. (2001) of
more than 3,500 male workers exposed to TCDD-contaminated products at eight U.S. chemical
plants. Each worker was assigned an exposure score based on estimated level of contact with
TCDD, the degree of TCDD contamination of product at each plant over time, and the fraction of
a workday in contact with the product. For 170 workers, the serum TCDD levels were also
measured. The serum levels were back-extrapolated to the last time of exposure using a constant
biological half life, and regressed on the exposure scores. This regression model was used to
predict the dose in all workers, and predicted dose was correlated with cancer mortality.
Figure 6-1 shows a scatter plot of back-casted versus predicted TCDD serum levels for the
170 workers on which the regression was based.
Given a predicted TCDD level, the uncertainty on the back-casted TCDD value could be
inferred from such data by various techniques. A key question is whether the actual cancer
mortalities among 170 back-casted workers are randomly placed in the conditional distribution
given predicted TCDD. Imagine, in other words, that the mortalities among the 170 back-casts
are colored red in the above picture. At any given level of TCDD prediction, are the red points
evenly distributed, or are they shifted to the right? In principle, the correlation between mortality
and back-casted TCDD level, given the predicted level, could be estimated. This amounts to
estimating heteroscedasticity in the regression model. Then, for each of the 3,538 workers, given
his predicted TCDD level, we could sample a back-casted TCDD level, appropriately correlating
with mortality, and recompute the dose response analysis. Repeating this many times we could
build up a distribution for excess lifetime cancer mortality risk.
It is instructive to step through similar issues with regard to biological half life,
background and body fat. The Steenland et al. (2001) analysis assumed a constant TCDD
biological half life (8.7 years). A distribution over this half life could plausibly be developed
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
from published sources. Assuming this half life is constant for all workers, but uncertain
(epistemic uncertainty), this distribution could easily supplement the previous distribution: first
sample a half life (to be applied to all workers), then estimate the regression model for this half
life, and sample back-casted TCDD levels given each worker's exposure score, taking account of
correlation with mortality. This works if the half life uncertainty is epistemic. However, since
the half life is estimated from data, it is more reasonable to suppose that the half life varies from
worker to worker (aleatoric uncertainty). Here again the correlation with mortality must be taken
into account, indeed it seems reasonable to suppose that the 256 cancer deaths involved workers
with longer half lives. However, there is no way ex post of determining the biological half life in
the deceased workers.
Background exposure and body fat are similar to half life. The study authors held
background level constant at the median level (6.1 ppt, range 2.0-19.7) for 79 nonexposed
workers from whom blood was also drawn (see also Section 6.4.2.4). The full distribution of
TCDD levels for these nonexposed workers could be used as well. Is it reasonable to suppose
that responsive workers have background levels that are sampled randomly from this
distribution, or might they not plausibly come from the high end of the distribution? The
analysis also assumed a constant percentage of body fat (30%), whereas body fat percentage
varies in the general population, e.g., for men this has been reported to range from 2 to 38% or
more (see Footnote 32). The body fat distribution in the worker population could have been
ascertained, but again the question arises, are the responsive workers sampled randomly from
this distribution?
These three factors, variable half life, variable background, variable body fat percentage,
might conspire to make the effective dose level among the responsive workers significantly
higher than would appear in a study that assumes these factors to be constant. However, such
concerns cannot be addressed in a quantitative uncertainty analysis, unless cancer mortality can
be correlated with these variables. In an optimal study design, this information could be
retrieved from the data, but optimality is a luxury that is seldom afforded. The optimal should
not be the enemy of the good; it might be possible to estimate these correlations in some other
defensible manner, in which case the effect of exposure uncertainty could be quantified and
propagated. Such an analysis would involve substantial effort and should not be undertaken
under assumptions that are themselves implausible. Experimental protocols do not currently
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
require such uncertainty quantification; perhaps they should and perhaps someday they will. In
any event Steenland et al. (2001) must be applauded for conscientiously identifying these key
issues.
6.4.2.3. Uncertainty in Toxicity Equivalence (TEQ) Exposures in Epidemiological Studies
Toxicity equivalence factors (TEFs) are used to infer the health effects of dioxin-like
compounds based on their relative potencies compared to TCDD. These factors are not known
with certainty, and the question arises whether uncertainty in TEFs can be incorporated into a
quantitative uncertainty analysis. The process of deriving TEFs applied by the World Health
Organization (WHO, 2005) is described in Van den Berg et al. (2006). Distributions of relative
potencies (REPs) were developed from the scientific literature, with preference for in vivo
studies, as supplemented by in vitro studies. An expert panel used a consensus process to select
a TEF value for each congener, in half log steps "Thus, the TEF is a central value with a degree
of uncertainty assumed to be at least ± half a log, which is one order of magnitude. However, it
should be realized that TEF assignments are usually within the 50th to 75th percentile of the REP
distribution, with a general inclination toward the 75th percentile in order to be health protective"
(Van den Berg et al., 2006) (see Figure 6-2).
The WHO considers the uncertainty in TEFs to span one order of magnitude (presumably
log uniformly distributed). It would be tempting to use the distributions in Figure 6-2 to quantify
uncertainty in the TEFs in a quantitative uncertainty analysis. However, the issue of dependence
in this case is daunting. For example, should values of 1,2,3,7,8,-pentachlorodibenzofuran and
2,3,4,7,8-pentachlorodibenzofuran be sampled independently? The choice of dependence
structure will have a large effect. As described by Van den Berg et al. (2006), the differences in
REPs reflects differences in dosing regimens, species, endpoints, mechanisms, and calculation
methods. In a quantitative uncertainty analysis one must insure that these are not double
counted.
Reasons for significant differences in REPs for the same congener can be caused
by the use of different dosing regimens (acute vs. subchronic), different endpoints,
species, and mechanisms (e.g., tumor promotion caused by at least two different
mechanisms as for mono-o/7/?o-substituted PCBs), as well as different methods
used for calculating REPs. Thus, different methodological approaches used in
different studies clearly provide uncertainties when deriving and comparing REPs.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-26 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
If future study designs to derive REPs were more standardized and similar, the
variation in REPs when using the same congener, endpoint, and species might be
expected to be smaller (Van den Berg et al., 2006).
Although the TEFs themselves and the distributions underlying them are based on expert
judgment, it is possible to incorporate these into a quantitative uncertainty analysis; however, it
is not simply a matter of taking the distributions in Figure 6-2 to predict the results, with
uncertainty, of exposure to dioxin-like compounds. The issues of dependence and double
counting must first be addressed. Inasmuch as the distributions are the result of expert judgment,
this would reasonably involve structured expert judgment as well. (Procedures for this type of
assessment have been developed and applied, and it would entail a significant level of effort.)
6.4.2.4. Uncertainty in Background Feed Exposures in Bioassays
TCDD is not produced intentionally but rather is formed as a byproduct of volcano
eruptions, forest fires, manufacturing of steel and certain chemicals (including some pesticides
and paints), pulp and paper bleaching, exhaust emissions, and incineration. It enters the food
supply primarily via aerial transport and deposition of emissions, and it bioaccumulates in animal
fat. In general, food of animal origin contributes to about 80% of the overall human exposure.
For example, Schecter et al. (1997) measured dioxins in pooled food samples collected in 1995
from supermarkets across the United States. Reported as parts per trillion (ppt) toxicity
equivalences (TEQs), fresh water fish had the highest level (1.43); followed by butter (1.07);
hotdog/bologna (0.54); ocean fish (0.47); cheese (0.40); beef (0.38); eggs (0.34); ice cream
(0.33); chicken (0.32); pork (0.32); milk (0.12); and vegetables, fruits, grains, and legumes
(0.07). (More recent exposure studies by Lorber et al. [2009] and others have updated these
older estimates of dietary levels.)
These results illustrate that a person's dietary intake of dioxins depends on the relative
intake of foods with high or low levels of contamination, and human background levels will vary
accordingly. The same applies to experimental animals in bioassays, although in those cases the
background intake can in principle be controlled. Some of the effects of TCDD and other AhR
agonists in enhancing the early initiation stages of cancers are considered to occur as a result of
prenatal exposures that are not included in the standard National Toxicology Program (NTP)
bioassay protocol (Brown et al., 1998; Muto et al., 2001). Further, to enhance reproducibility
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
and keep statistical fluctuations to a minimum, the standard NTP assays are deliberately run on
groups of animals that are relatively uniform genetically, fed uniform diets, and have the
minimum possible exposures to toxicants other than the agent(s) being tested. This tends to
reduce the potential for observing the consequences of potential interactive effects that might
occur in the diverse human population with its variety of dietary and other exposures to a wide
range of potentially interacting substances and conditions.
A critical question is the extent to which the background exposure influences the
dose-response curve, and how this background should be taken into account.40 One idea,
articulated in the recent NRC (2009) report on science and decisions, involves an "interacting
background." This can be implemented by computing a virtual dose B which, according to the
selected dose-response model, would explain a chosen fraction of the background response. If
the chosen model for dose 8 is f(8), the model can be adapted to account for an interacting
background by writing P(8) = f(5 + B) - f(B). This can alter the model's behavior at zero dose.
For example, if f(8) = 8n/(8n + EC so'1), the derivative d(f)/d(S) is nSn~'EC5o7(Sn + ECso")2,
which goes to zero as 8—>0, if n > 1. However, replacing 8 with (8 + B) evidently changes the
derivative at zero to nBn 'ECso'VCB11 + ECso")- This model is not yet estimable from data, as we
have no way of choosing from the available animal data the fraction of background response to
be explained by the model when applied to humans (although judgments could be made if we
had better information about the details of the processes that are involved in causing various
human health effects). However, as a conceptual model, it serves to remind us that the manner
of accounting for background exposures can influence a model's behavior in the low-dose
region.
6.4.2.5. Feasibility of Quantifying the Uncertainties Encountered When Choosing Specific
Studies and Subsets of Data (e.g., Species and Gender)
Species, strain, gender, life stage, and other characteristics of experimental animals are
selected for a given study based on previous knowledge (e.g., of the species sensitivity,
40"Effects of exposures that add to background processes and background endogenous and exogenous exposures can
lack a threshold if a baseline level of dysfunction occurs without the toxicant and the toxicant adds to or augments
the background process. Thus, even small doses may have a relevant biologic effect. That may be difficult to
measure because of background noise in the system but may be addressed through dose-response modeling
procedures" (NRC, 2009).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
availability of strains having little genetic variation for the endpoints in question, relevance of
the MO A, and degree to which the endpoints are similar for humans). Many other decisions are
made in designing a bioassay study; will the animals be sacrificed at the termination of the study
(if not a lifetime study), or will they be allowed to live out their natural lives? What dosing
regimen should be applied? How will the animals be fed and handled? Although such questions
may engender uncertainty in the minds of the experimenters, and reviewers; such uncertainty is
not amenable for quantitative uncertainty analysis unless and until there are quantitative models,
with parameters estimable from data, that can predict the effect of these choices on the response
function.
6.4.2.6. Feasibility of Quantifying the Uncertainties Encountered when Choosing Specific
Endpoints for Dose-Response Modeling
Standard experimental protocols guide the selection of exposure/dosing conditions for a
given bioassay, including the amount, delivery vehicle, route, timing, dosing frequency and
duration, and dose spacing. The goal is to find the dose range where the experimental animals
begin to respond adversely, to help anchor the lower end of the dose-response relationship, and
to avoid multiple experiments in which all or none of the animals respond. A common
recommendation is that the dose levels be chosen such that the increments in probability of
response are roughly equal. Hence, the choice of endpoint, spacing, and number of animals
should be made with these factors in mind. Of particular importance is the number of animals at
each dose level in relation to the choice of endpoint and probability of response. Using more
animals at the lower dose levels increases the probability of seeing some animals respond; on the
other hand, this approach will give higher weight to the low-dose responses in model fitting and
uncertainty quantification. Including many experiments at low dose with no expected response
could provide deceptive graphs.
One of the most clearly nonlinear responses to high-dose dioxin exposure cited by the
committee is cholangiocarcinoma (bile duct cancer) in rats (see Figure 6-3). The model
Prob(dose) = dosen/(dosen + ECso") with Hill coefficient n was fit to these data, yielding a
maximum likelihood estimate of n = 2.79 (Hattis, 2009). For n > 1, this is a zero slope at zero
dose model (see Section 5.2.3.3). If we choose a linear extrapolation below the POD at dose
3.82 ng/kg-day with probability of response = 0.02, then at dose 1.7 ng/kg-day, the predicted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
probability of response is 0.009; with 46 rats exposed to this dose we expect to see no responses,
which is what we see. However, sublinear models will yield a higher probability of seeing
nothing. A graph like Figure 6-3 would suggest that the linear extrapolation was missing the
experiments at doses 1.7 and 0.52 ng/kg-day, indicating lack of fit, whereas the null response is
what we should expect on the linear model.
The conclusion with regard to the feasibility of this quantitative uncertainty analysis
echoes that of the previous paragraph: such uncertainty is not amenable for quantitative analysis
unless and until there are quantitative models, with parameters estimable from data, that predict
the effect of these choices on the response function, although sensitivity analyses can be done
showing the consequences of assuming different amounts of interacting background within the
context of a specific nonlinear model.
6.4.2.7. Feasibility of Quantifying the Uncertainties Encountered when Choosing a Specific
Dose Metric (Trade-Off Between Confidence in Estimated Dose and Relevance of
MOA)
The concept of dose is not straightforward. To review, U.S. EPA (2005) Cancer
Guidelines provide the following taxonomy:
• Exposure is contact of an agent with the outer boundary of an organism.
• Exposure concentration is the concentration of a chemical in its transport or
carrier medium at the point of contact.
• Dose is the amount of a substance available for interaction with metabolic
processes or biologically significant receptors after crossing the outer boundary of
an organism.
• Potential dose is the amount ingested, inhaled, or applied to the skin.
• Applied dose is the amount of a substance presented to an absorption barrier and
available for absorption (although not necessarily having yet crossed the outer
boundary of the organism).
• Absorbed dose is the amount crossing a specific absorption barrier (e.g., the
exchange boundaries of skin, lung, and digestive tract) through uptake processes.
• Internal dose is a more general term, used without respect to specific absorption
barriers or exchange boundaries. Delivered dose is the amount of the chemical
available for interaction by any particular organ or cell.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Using the applied dose or absorbed dose would yield statistically more powerful results
and enable more precise predictions. If it is not possible to measure these, then other available
dose metrics such as potential dose or exposure are used. Due to toxicokinetic variability,
different individuals receiving the same exposure may not have the same applied or absorbed
dose. Hence, use of an exposure metric would add variability to the predicted results. The dose
metric should be selected that (1) has the most proximate possible causal relation to the
production of an adverse health endpoint, and (2) is most easily related to the units of (external)
exposure that will be the basis for assessing and controlling human exposures.
6.4.2.8. Feasibility of Quantifying the Uncertainties Encountered When Choosing Model
Type and Form
U.S. EPA (2009d) draft white paper on probabilistic methods notes: "There is no
consensus on any one well-accepted general methodology for dealing with model uncertainty,
although there are various examples of efforts to do so." Model uncertainty was introduced in
Section 6.1.3.4. Many statistical techniques are available to evaluate model adequacy or to
choose a "best" model. Although it is tempting to qualify such deliberations as "uncertainty that
a model is true," one must remember that all models, being idealizations, are false. Ultimately,
one is interested in uncertainty with regard to observable phenomena, not with regard to models.
Models are merely tools for describing the phenomena. Nonetheless, the choice of a model
constrains the ways in which uncertainty can be represented, and the question addressed in this
paragraph is: How should one deal with these constraints? A recent study of uncertainty
modeling in dose response (Cooke, 2009) addresses precisely this issue and provides ample
technical detail to frame possible options.
Before elaborating on exotic approaches to model uncertainty, it is well advised to
appreciate a feature in the standard statistical treatment of uncertainty. Consider a model based
on experimental data, typically bioassay data, in which a certain number of study subjects are
exposed to varying doses of a test substance, and in which the numbers of subjects exhibiting a
response are tallied. Values for the parameters in the model are chosen by the principle of
maximal likelihood: those values are chosen which render the data as likely as possible.
According to standard practice, a model is chosen that best fits the data according to one of the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
accepted criteria, such as reduced R2, or the Akaike Information Criterion. There might be many
incompatible models that are nearly as good.
One can ask the following: If the experiments on which the model is based were repeated,
sampling the same number of experimental subjects from the distribution posited by the model,
how much could our parameter estimates change? This is described by a joint distribution over
the model's parameters, which captures sampling uncertainty under the assumption that the
model is true. Now, all models are false, and as our sample sizes grow the lack of fit in the
model becomes increasingly apparent. At the same time, the sample fluctuations in parameter
estimates—assuming the model is true—become ever smaller. In very large epidemiological
studies, standard statistical methods can produce razor-thin confidence bands in this way, which
fail to capture experts' uncertainty regarding observable phenomena.41
The exotic methods sketched in the beginning of Section 6.4.2 may be viewed as attempts
to deal with this feature. Probabilistic inversion methods were deployed on a large scale in the
joint U.S. NRC-EU uncertainty analyses noted in Section 6.1. Distributions over model
parameters are intended to capture an antecedently defined uncertainty over observable
phenomena predicted by the model. This method was applied in dispersion and deposition
modeling and further environmental transport models (including uptake) for radionuclides. In
most cases, the observable uncertainty was based on structured expert judgment, but it has also
been based on binomial uncertainty in bioassay studies. A potential drawback is that it may not
prove possible to capture the observable uncertainty in this way with a classically best-fitting
model, and new models may be required.
Nonparametric Bayesian methods arose in the biomedical and reliability fields. They
start with a prior distribution over all nondecreasing dose-response functions, and update these
with observations from a bioassay study. No further assumptions regarding parametric form are
introduced, but the prior distribution remains important for doses outside the range of
observations. Bayesian model averaging starts with a prior distribution over a set of candidate
models, and updates this distribution with bioassay data. The method is flexible and intuitive,
though attenuation of the effect of the prior on the posterior must be verified.
41 See for example Tuomisto et al. (2008, Table 6) for a comparison of experts' uncertainty in health effects of fine
particulates with uncertainties derived from sampling uncertainty from large epidemiological studies.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-32 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
All these approaches represent attempts to capture "extramodel uncertainty," that is,
uncertainty that is not conditional on the truth of the model. This is an active research area, and
the last word on capturing extramodel uncertainty in quantitative uncertainty analysis has not
been spoken. A major effort with regard to dioxin would be indicated when the strengths and
weakness of the exotic methods are well understood.
6.4.2.9. Threshold MOA for Cancer
The NAS committee avers that knowledge of the AhR binding MOA entails that there is
a response threshold for TCDD cancer induction. The differences between individual and
population thresholds are not discussed, but the following two possibilities are distinguishable:
1. The threshold is the same for each individual; since human variability in AhR binding
affinity is rather large (see Section 5.2.3.3), this entails that the threshold is not affected
by the binding affinity.
2. The threshold varies across individuals and is related to the individual AhR binding
affinity.
These two positions are different. As shown in Section 5.2.3.3, it is quite possible that
each individual in a population has a threshold whereas the population dose-response relation is
linear. Because the NAS committee does not distinguish which of these positions it holds, the
feasibility of quantitative uncertainty analysis is examined here for both.
i. Quantitative uncertainty analysis concerns a mathematical model. In case (i), this model
would show how the existence of the AhR binding would induce a threshold,
independently of the strength of the binding. Assessing the feasibility of quantitative
uncertainty analysis must await the elaboration of such a model.
ii. In this case, it must be shown that the distribution of thresholds, and the dose-response
function above the threshold, is such as to induce a population "zero slope at zero dose"
(ZS@Z) model. Recall, the burden of proof is on this (ZS@Z) model. Scoping the
population variability with regard to AhR-mediated mechanisms in general, and dioxin
sensitivity in particular, is an active area of research. It involves phenotyping human
AhR-mediated responsiveness and relating this to polymorphisms in the human
population. Harper et al. (2002) report that a 10-fold variation in binding affinity of AhR
for TCDD in human placental tissue did not reveal any polymorphisms, suggesting that
the relation between phenotypical and genotypical variation is tenuous at best. Tuomisto
et al. (1999) demonstrate large variations in efficacy in two rat strains whose binding
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-33 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
affinity is similar (Long-Evans, Kd = 3.4, Han/Wistar, Kd = 3.9 [Connor and Alyward,
2006]), and they also show that this variation is endpoint-specific. The responses in both
strains are similar for cytochrome P450 (CYP)l A1 induction, but very dissimilar for
thymus atrophy, serum bilirubin, and mortality. Toide et al. (2003) suggest that common
biochemical measures of EROD activity might be mediated by CYP1B1 and CYP1A2.
The differences in serum bilirubin at doses around 10 |ig/kg are about a factor of 30.
Han/Wistar rats seldom die at this dose, while mortality of Long Evans rats is about 50%.
The mechanisms are not understood.
Although the mass action dose-response model does not have a threshold, it is possible
that certain enzymes block the receptor binding, and until these are overwhelmed, no response
occurs. The availability of such enzymes may vary from individual to individual, and may or
may not covary with the dissociation constant Kd. Pursuing these lines of research may result in
a convincing demonstration of a population (ZS@Z) model. Such a model would express the
individual threshold in terms of parameters that could be estimated with uncertainty from the
data.
6.4.2.10. Feasibility of Quantifying the Uncertainties Encountered when Selecting the BMR
The NAS committee explicitly requested that the uncertainty attending the choice of a
BMR be quantified. First of all, simply plugging other values in for the BMR does not constitute
a quantitative analysis of uncertainty. The plugged-in values must be sampled from some
uncertainty distribution. Since this concerns volitional uncertainty, there is no underlying
distribution from which to sample, unless the choice of BMR is related to some claim about the
state of the world.
6.5. CONCLUSIONS
The main conclusions regarding the feasibility of quantitative uncertainty analysis are
first summarized in relation to specific suggestions made by the NAS committee. A suggested
research agenda follows.
6.5.1. Summary of NAS Suggestions and Responses
On page 130 of their report (NAS, 2006a), NAS makes specific suggestions regarding
uncertainty quantification. These are reformatted and presented in italics below. Following each
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-34 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
suggestion, a summary of the discussion in this section is given, with reference to the section in
which it is addressed.
EPA should have addressed quantitatively the following sources of uncertainty:
• Basis for risk quantification:
1. bioassay data,
2. occupational cohort data.
Response: (a) Classical statistical methods yield distributions on model parameters
which reflect sample fluctuations, assuming that the model is true. This type of
uncertainty is taken into account in the BMDL. Exotic methods can account for
uncertainty which is not conditional on the truth of a model, at least for bioassay data
(see Section 6.4.2). (b) For epidemiological data, the dose reconstruction often involves
assumptions which may support data driven uncertainty analysis, if sufficient data can
be retrieved. Examples discussed above include back-casted TCDD level, biological
half life, body fat and background (see Section 6.4.2.2). Uncertainty in the choice of
bioassay data sets or choice of occupational cohort data sets is volitional, and is not
quantified by sampling an input distribution. To be amenable for quantitative
uncertainty analysis, the choice must be linked to a statement about the state of the
world (see Section 6.1.1).
• Epidemiology data to use:
1. risk estimate developed with data aggregated from all suitable studies,
2. risk estimate or estimates developed using each study individually.
• Factors affecting extrapolation from occupational to general population cohorts,
including differences in baseline health status, age distribution, the healthy worker
survivor effect, and background exposures.
Response: (a) Quantitative uncertainty analysis based on meta-analysis data poses
challenges owing to differences in study protocols. Exotic methods might take us further,
the question is whether the restriction to data driven methods (as opposed to expert
judgment or Bayesian methods) could be maintained (see Sections 6.4.2.2 and 6.4.2.3).
(b) If the general population is characterized by distributions over known confounders
whose coefficients are estimated from the epidemiological studies, then uncertainty over
these coefficients can be extracted with the methods mentioned in Section 6.4.2.1.
Uncertainty due to missing covariates is intractable for data driven uncertainty analysis
(see Section 6.4.2.2).
• Bioassay data to use:
1. risk estimate developed with the single data set implying the greatest risk (that is,
single study, tumor site, gender),
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-35 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
2. risk estimate developed with multiple data sets satisfying an a priori set of
selection criteria.
Response: (a) Echoing response l.b, uncertainty in choice of data sets is volitional and is
not quantified by sampling an input distribution. To be amenable for quantitative
uncertainty analysis the choice must be linked to a statement about the state of the world
(see Section 6.1.1). (b) The issue here is similar to the meta-analysis addressed in (2.a).
• Dose-response model:
1. linear dose response,
2. nonlinear dose.
Response: (a) Presumably the low dose extrapolation is meant. When the linear model is
used by default, the uncertainty is volitional. To be amenable for quantitative uncertainty
analysis the choice must be linked to a statement about the state of the world (see
Section 6.1.1). The EDX as POD for the linear extrapolation can be subjected to
quantitative uncertainty analysis, if based on sufficient bioassay data, (b) presumably
'nonlinear dose response' is meant. It is possible that human thresholds exist, and that
the distribution of thresholds can be characterized in the human population. In as much
as the mechanisms for this are not yet understood, there is no quantitative model
expressing threshold as a function of parameters which could be estimated, with
uncertainty, from data. This currently limits the application of uncertainty quantification
(see Section 6.4.2.9).
• Dose metric:
1. average daily intake,
2. area under the blood concentration-time curve,
3. lifetime average body burden,
4. peak body burden,
5. other.
Response: The dose metric is chosen to maximize causal proximity to the endpoint,
while maintaining the link to measured exposure (see Section 6.4.2.7). There may be
uncertainty with regard to which metric is optimal. If an inappropriate metric is chosen
in a bioassay study, this would be expressed in noisier responses which would tend to
suppress the dependence of endpoint on dose. A data driven quantitative uncertainty
analysis of dose metric would require a mathematical model expressing endpoints as a
function, inter alia, of dose metric, with parameters estimated from data.
• Dose metric—biological measure:
1. free dioxin,
2. bound dioxin.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-36 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
Response: The issue is whether all TCDD available for AhR binding, or only the bound
TCDD, should be used as a dose metric. Binding affinity is determined by more factors
than genetic polymorphisms and these other factors are poorly understood (see
Section 6.4.2.9). A quantitative uncertainty analysis must await the formulation of a
quantitative model expressing binding affinity in terms of parameters which can be
estimated from data.
• POD:
a. ED10,
b. ED05,
c. EDoi
Response: Uncertainty in choosing a POD is volitional. Uncertainty in the value of an
EDX can be quantified in a data driven manner if sufficient bioassay data is at hand (see
Section 6.4.1.1).
• Value from ED distribution to use:
1. ED,
2. lower confidence bound value for the ED (LED),
3. upper confidence boundfor the ED (UED).
Response: Given that uncertainty on the POD is quantified, it is a trivial exercise to
derive a distribution of the slopes of a linear low dose extrapolation, and hence a
distribution of a risk specific dose. This would seem preferable to choosing between a
lower, upper, or nominal value.
• Where alternative assumptions or methodologies could not be ruled out as implausible or
unreasonable, EPA could have estimated the corresponding risks and reported the
impact of these alternatives on the risk assessment results. The potential impacts of four
sources of uncertainty are discussed below.
• The full range ofplausible parameter values for the dose-response functions used
to characterize the dose-response relationship for the three occupational cohort
studies selected by EPA (Ott and Zober, 1996; Becher et al., 1998; Steenland et
al., 2001).
• Use of other points of departure, not just the EDoi (or LEDoi), to develop a CSF.
• Alternative dose-response functional forms as well as goodness of fit of all
models, especially at low doses.
• Uncertainty introduced by estimation of historical occupational exposures.
Response: (a) The study of Steenland et al. (2001) was selected to illustrate the
possibilities and limitations of quantitative uncertainty analysis for this type of study (see
Section 6.4.2.2). (b) The possibilities for uncertainty quantification with regard to the
POD are discussed in Section 6.4.1.1 and in the POD bullet above, (c) Goodness of fit at
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-37 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
any measured dose is evaluated in standard packages. There may be different models
with comparable goodness of fit at observed doses which differ strongly at doses outside
the measured range. Extra model uncertainty, that is, uncertainty which is not conditional
on the truth of any given model, is addressed by the exotic methods (see Section 6.4.2).
(d) The feasibility of quantifying uncertainty in occupational exposure is study specific.
The example of Steenland et al. (2001) was discussed in some detail (see
Section 6.4.2.2). In general, the problem is not so much quantifying the exposure
uncertainty, but in quantifying the dependence between the endpoints and the exposure
uncertainty.
6.5.2. How Forward? Beyond RfDs and Cancer Slope Factors to Development of
Predictive Human Dose-Response Functions
Uncertainty quantification is an emerging area in science. There are many examples of
highly vetted and peer-reviewed uncertainty analyses based on structured expert judgment.
Under this process, experts in effect synthesize a wide diversity of information in generating
their subjective probability distributions. Where considerable data exist for an environmental
pollutant, such as for the well-studied TCDD, it is natural to ask whether these extensive data can
be leveraged more directly in uncertainty quantification. This is an area where research could be
focused. The requisite knowledge does not yet exist, but there are promising lines of attack. It is
therefore not a question of convening blue-ribbon panels to reveal the proper approach; instead
multiple approaches should be encouraged to try out new ideas and share experiences.
An important idea that has been pioneered in Europe is to organize bench-test exercises
where different approaches are applied to a common problem. This focuses the discussion on
real issues and builds a community of capable practitioners. Such initiatives have proven much
more productive than simply supporting individual researchers to explore their ideas.
Areas for which bench-test exercises might be appropriate include:
• Testing "exotic methods" for capturing model uncertainty
• Combining bioassay and epidemiological data for uncertainty quantification
• Applicability of structured expert judgment, e.g., for low-dose extrapolation
• Conducting dependence modeling, dependence inference, and dependence elicitation
(such as with regard to TEFs).
Looking beyond compounds for which considerable data exist, there will always be a
need to evaluate new substances. The target will be a simple method that:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-38 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
1. Can yield predictions of toxicological indicators with uncertainty via a valid probabilistic
mechanism.
2. Could evolve from approaches based on similarities (such as a random chemical model)
under which the new substance could be seen as a random sample from a reference
distribution of chemicals considered sufficiently similar, e.g., in terms of structure,
physicochemical properties, and biological activity (potency).
3. Is nondisruptive to the traditional process.
This last feature is very important because as a practical matter, any collective push
forward in this area must extend from (not abruptly depart from) current methodology. For
example, the discussion surrounding uncertainty factors suggests that a probabilistically valid
inference system could substantially differ from the current system. Nonetheless, to meld with
current practice, it must initialize on the current system and allow this system to evolve in a
measured fashion.
Additional research topics relevant to dioxin that could further inform health assessments
include population variability of biokinetic constants, threshold mechanisms for the mass action
model, and polymorphism at frequencies lower than 1%. Further data and improved
methodologies in these areas, combined with developments illustrated elsewhere in this report,
will help reduce uncertainties and strengthen our understanding of potential health implications
of environmental contaminants.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
Table 6-1. Key sources of uncertainty
Selection of endpoint and of species/strain, gender, life stage, other subject characteristics
- critical effect
- sensitivity (e.g., species, life stage)
- human relevance
Selection of key study(ies): human data and bioassays (strength, inclusion criteria)
- epidemiological studies, clinical/case reports (exposure estimate)
- adequacy of study design, statistical power (exposure term, histopathology)
- human relevance of bioassays (TK, MOA, endpoint)
- data uncertainty, confidence in data; database deficiencies
Use of TK, dosimetry; body burden; species differences, cross-species extrapolation
- bioavailability, dose dependence
- half life, life stage, body fat, other compartments, age, other factors
- body burden (peak, steady state, lifetime average)
- physiologically-based pharmacokinetic (PBPK) modeling
- scaling (human equivalents), adjustments (default and non-default; with TP)
Selection of dose metric
- intake (averaging time)
- background (what place on the dose-response curve)
- free vs. receptor-bound TCDD
- tissue-specific concentration
- lipid-normalized level
Selection of POD
- selection (e.g., NOAEL/LOAEL, BMDL, ED0i, 05,10)
- derivation method (e.g., BMD)
- choice of model form (e.g., Hill, Weibull)
- statistical uncertainty at/confidence in POD
Selection of dose-response model (e.g., biologically based, multi-stage) and of BMR
- biological plausibility, MOA
- model type and form, alternative functional forms
- range of plausible parameter values
- goodness of fit, especially at low doses
Selection of low-dose extrapolation approach
- linear/nonlinear
- threshold/nonthreshold
Human population variability
- subpopulations (e.g., occupational, general public, sensitive groups)
- polymorphisms
- life stage, other features
- individual vs. population threshold
Characterization of risk/effect
- adversity of effect (vs. in normal range of variation and adaptation)
- uncertainty factors (TK; TD; chemical-specific vs. default; justification)
- consistency of methods for endpoints with common MOA
- back-extrapolation from occupational data
- MOE, RfD; beyond a point estimate for SF
3 PBPK = physiologically-based pharmacokinetic; SF = slope factor; TD = toxicodynamic; TK =
4 toxicokinetic.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-40 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table 6-2. PODs and amenability for uncertainty quantification
2
POD
Data profile
Choice
Uncertainty quantification
LOAEL
Experimental dose
level from set of
exposure-response data
Choose set of
exposure-response
measurements
No
NOAEL
Experimental dose
level from set of
exposure-response data
Choose set of
exposure-response
measurements
No
BMDL
Estimate from
bioassay data
Choose BMR, choose
dose-response relation
No, the BMDL is a quantile of
an uncertainty distribution
assuming that the
dose-response model is true
EDX
Estimate from set of
exposure-response data
Choose bioassay
experiments to estimate EDX
Yes, if full bioassay data are
available
3
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-41 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
1QOOO.Q"
1000.0
i
Q
Q 30.0
O
I-
~u
Q.
1.0
' •&>•:
*&J7 •*
•'jEr?
m m
• • •
o.t
3
4
5
6
7
0.1 1,0 10,0 l§0,0 1000.0 10000,0 1000M.O
Back-Extrapolated TCDD
Figure 6-1. Back-casted vs. predicted TCDD serum levels for a worker
subset.
Source: Steenland et al. (2001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-42 DRAFT—DO NOT CITE OR QUOTE
-------
12378-PeGDD
123478-HxCDD
123789-HxCDD
1234678-HpCDD
OCDD
TCDF
12378-PeCDF
23478-PeCDF
123478-HxC DF
123678-HxCDF
234678-HxCDF
OCDF
LEGEND
10th 25th; 50th 75th 90th
Who19S8tef
-n_j
~ r1
0.00001
PCB-77
PC B-126
PC B-169
PCB-105
PCB-114
PCB-118
PCB-123
PC B-156
PCB-157
PC B-189
I—DO—N
D—I'
0 0001 0.001 0.01
Relative Potencies (REPs)
0.1
H3»
X
ffl •
CZLZZZZ3-
LEGEND
•HHW
mzm
10th 25th; 50th 75th 90th
H
WMO,mTEF
0.0000001 0.000001 0.00001 0.0001 0.001
Relative Potencies (REPs)
0.01
0.1
1
2
3
4
Figure 6-2. Distribution of in vivo unweighted REP values in the 2004 database.
Source: Van den Berg et al. (2006), reprinted with permission from Haws et al. (2006).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-43 DRAFT—DO NOT CITE OR QUOTE
-------
1
0.5
Obs Fract Bile Duct Ca in Rats
Hill Mod Fit—Central Est of N
(25/53)
"O
CD
CD
CL
(4/49)
"O
(0/49, 48 and 46)
(1/50)
LL
2
4
6
8
10
12
14
16
18
20
2 Human equivalent dose (ng/kg-day), from simple body weight-1/4 projection
3
4 Figure 6-3. Plot of observed rat cholagiocarcinoma incidence and central
5 estimate of Hill Model fit to the data vs. human equivalent dose.
6
7 Source: Hattis (2009).
8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 6-44 DRAFT—DO NOT CITE OR QUOTE
-------
1
REFERENCES
2
3 Abbott, BD; Birnbaum, LS; Diliberto, JJ. (1996). Rapid distribution of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)
4 to embryonic tissues in C57BL/6N mice and correlation with palatal uptake in vitro. Toxicol. Appl. Pharmacol
5 141(1): 256-263.
6 Abraham, K; Krowke, R; Neubert, D. (1988) Pharmacokinetics and biological activity of 2,3,7,8-tetrachlorodibenzo-
7 p-dioxin. 1. Dose-dependent tissue distribution and induction of hepatic ethoxyresorufin O-deethylase in rats
8 following a single injection. Arch Toxicol 62(5):359-368.
9 Abraham, K; Knoll, A; Ende, M; et al. (1996) Intake, fecal excretion, and body burden of polychlorinated dibenzo-
10 p-dioxins and dibenzofurans in breast-fed and formula-fed infants. PediatrRes 40(5):671-679.
11 Abraham, K; Geusau, A; Tosun, Y; et al. (2002) Severe 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) intoxication:
12 Insights into the measurement of hepatic cytochrome P450 1A2 induction. Clin Pharm Therapeut 72(2):163-174.
13 Aittomaki, A; Lahelma, E; Roos, E; et al. (2005) Gender differences in the association of age with physical
14 workload and functioning. Occup Environ Med 62(2):95-100.
15 Akhmedkhanov, A; Revich, B; Adibi, JJ; et al. (2002) Characterization of dioxin exposure in residents of
16 Chapaevsk, Russia. J Expo Anal Environ Epidemiol 12(6):409-417.
17 Akhtar, FZ; Garabrant, DH; Ketchum, NS; et al. (2004) Cancer in US Air Force veterans of the Vietnam War. J
18 Occup Environ Med 46(2): 123-136.
19 Alaluusua, S; Calderara, P; Gerthoux, PM; et al. (2004) Developmental dental aberrations after the dioxin accident
20 inSeveso. Environ Health Perspect 112(13): 1313-1318.
21 Alvarez-Pedrerol, M; Ribas-Fito, N; Torrent, M; et al. (2008) Thyroid disruption at birth due to prenatal exposure to
22 beta-hexachlorocyclohexane. Environ Intl 34:737-740.
23 Amin, S; Moore, RW; Peterson, RE; et al. (2000) Gestational and lactational exposure to TCDD or coplanar PCBs
24 alters adult expression of saccharin preference behavior in female rats. Neurotoxicol Teratol 22(5):675-682.
25 Andersen, ME; Mills, JJ; Gargas, ML; et al. (1993) Modeling receptor-mediated processes with dioxin: implications
26 for pharmacokinetics and risk assessment. Risk Anal 13(1):25—36.
27 Andersen, ME; Birnbaum, LS; Barton, HA; et al. (1997) Regional hepatic CYP1A1 and CYP1A2 induction with
28 2,3,7,8-tetrachlorodibenzo-p-dioxin evaluated with a multicompartment geometric model of hepatic zonation.
29 Toxicol Appl Pharmacol 144(1): 145—155.
30 Andersson, P; McGuire, J; Rubio, C; et al. (2002) A constitutively active dioxin/aryl hydrocarbon receptor induces
31 stomach tumors. Proc Natl Acad Sci 99(15): 9990-9995.
32 Ariens, EJ; Van Rossum, JM; Koopman, PC. (1960) Receptor reserve and threshold phenomena. I. Theory and
33 experiments with autonomic drugs tested on isolated organs. Arch Int Pharmacodyn 127:459-478.
34 Armstrong, BG. (1995) Comparing standardized mortality ratios. Ann Epidemiol 5(l):60-64.
35 ATSDR (Agency for Toxic Substances and Disease Registry), Public Health Service, U.S. Department of Health
36 and Human Services. (1998) Toxicological profile for chlorinated dibenzo-p-dioxins (CDDs). ATSDR, Atlanta, GA.
37 Available at: http://www.atsdr.cdc.gov/toxprofiles/tpl04.pdf.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-l DRAFT—DO NOT CITE OR QUOTE
-------
1 Aylward, LL; Hays, SM; Karch, NJ; et al. (1996) Relative susceptibility of animals and humans to the cancer hazard
2 posed by 2,3,7,8-tetrachlorodibenzo-p-dioxin using internal measures of dose. Environ Sci Technol
3 30(12):3534-3543.
4 Aylward, L; Kirman, C; Cher, D; et al. (2003) Reanalysis of dioxin cancer threshold. Environ Health Perspect
5 111(10):A510. Available at: http://www.pubmedcentral.nih.gov/picrender.fcgi?artid=1241631&blobtype=pdf.
6 Aylward, LL; Brunet, RC; Carrier, G; et al. (2005a) Concentration dependent TCDD elimination kinetics in humans:
7 toxicokinetic modeling for moderately to highly exposed adults from Seveso, Italy, and Vienna, Austria, and impact
8 on dose estimates for the NIOSH cohort J Expo Anal Environ Epidemiol 15:51-65.
9 Aylward, LL; Brunet, RC; Starr, TB; et al. (2005b) Exposure reconstruction for the TCDD-exposed NIOSH cohort
10 using a concentration- and age-dependent model of elimination. Risk Anal 25(4):945-956.
11 Aylward, LL; Bodner, KM; Collins, JJ. (2007) Exposure reconstruction for a dioxin-exposed cohort: integration of
12 serum sampling data and work histories. Organohal Comp 69:2063-2066.
13 Aylward, LL; Goodman, JE; Charnley, G; et al. (2008) A margin-of-error approach to assessment of noncancer risks
14 of dioxins based on human exposure and response data. Environ Health Perspect 116(10): 1344—1351.
15 Aylward, LL; Bodner, KM; Collins, JJ; et al. (2009) TCDD exposure estimation for workers at a New Zealand
16 2,4,5-T manufacturing facility based on serum sampling data. J Expo Sci Environ Epidemiol (epub doi:
17 10.1038/jes.2009.31). Available at http://www.nature.com/jes/journal/vaop/ncurrent/full/jes20093 la.html.
18 Baccarelli, A; Mocarelli, P; Patterson, DG, Jr; et al. (2002) Immunologic effects of dioxin: new results from Seveso
19 and comparison with other studies. Environ Health Perspect 110(12): 1169—1173.
20 Baccarelli, A; Pesatori, AC; Masten, SA; et al. (2004) Aryl-hydrocarbon receptor-dependent pathway and toxic
21 effects of TCDD in humans: a population-based study in Seveso, Italy. Toxicol Lett 149(l-3):287-293.
22 Baccarelli, A; Pesatori, AC; Consonni, D; et al. (2005) Health status and plasma dioxin levels in chloracne cases 20
23 years after the Seveso, Italy accident. Br J Dermatol 152(3):459-465.
24 Baccarelli, A; Hirt, C; Pesatori, AC; et al. (2006) T(14;18) translocations in lymphocytes of healthy dioxin-exposed
25 individuals from Seveso, Italy. Carcinogenesis. 27(10):2001-2007. Available at:
2 6 http: //www. ncbi. nlm. nih. gov/pubmed/16543249.
27 Baccarelli, A; Giacomini, SM; Corbetta, C; et al. (2008) Neonatal thyroid function in Seveso 25 years after maternal
28 exposure to dioxin. PLos Medicine 5(7): 1133-1142. Available at:
29 http://www.plosmedicine.Org/article/info:doi/10.137 l/journal.pmed.0050161.
30 Bang, KM; Kim, JH. (2001) Prevalence of cigarette smoking by occupation and industry in the United States. Am J
31 Ind Med 40(3):233-239.
32 Banks, YB; Birnbaum, LS. (1991) Absorption of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) after low dose dermal
33 exposure. Toxicol ApplPharmacol 107(2):302-310.
34 Banks, YB; Brewster, DW; Birnbaum, LS. (1990) Age-related changes in dermal absorption of 2,3,7,8-
35 tetrachlorodibenzo-p-dioxin and 2,3,4,7,8-pentachlorodibenzofuran. Fundam Appl Toxicol 15(1): 163-173.
36 Baron, JM; Zwadio-Klarwasser, G; Jugert, F; et al. (1998) Cytochrome P450 1B1: a major P450 isoenzyme in
37 human blood monocytes and macrophage subsets. Biochem Pharmacol 56:1105-1110.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-2 DRAFT—DO NOT CITE OR QUOTE
-------
1 Barouki, R; Coumoul, X; Fernandez-Salguero, PM. (2007) The aryl hydrocarbon receptor, more than a xenobiotic-
2 interacting protein. FEBS Lett. 581(19):3608-3615.
3 Bastomsky, CH. (1977) Enhanced thyroxine metabolism and high uptake goiters in rats after a single dose of
4 2,3,7,8-tetrachlorodibenzo-p-dioxin. Endocrinology 101:292-296.
5 Bates, MN; Buckland, SJ; Garrett, N; et al. (2004) Persistent organochlorines in the serum of the non-occupationally
6 exposed New Zealand population. Chemosphere 54(10): 1431—1443.
7 Becher, H; Flesch-Janys, D; Kauppinen, T; et al. (1996) Cancer mortality in German male workers exposed to
8 phenoxy herbicides and dioxins. Cancer Causes Control 7:312-321.
9 Becher, H; Steindorf, K; Flesch-Janys, D. (1998) Quantitative cancer risk assessment for dioxins using an
10 occupational cohort. Environ Health Perspect 106(Suppl 2):663-670.
11 Beebe, LE; Anver, MR; Riggs, CW; et al. (1995) Promotion of N-nitrosodimethylamine-initiated mouse lung
12 tumors following single or multiple low dose exposure to 2,3,7,8- tetrachlorodibenzo-p-dioxin. Carcinogenesis
13 16:1345-1349.
14 Bell, DR; Clode, S; Fan, MQ; et al. (2007a) Toxicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin in the developing male
15 Wistar(Han) rat. II: Chronic dosing causes developmental delay. Toxicol Sci 99(l):224-233.
16 Bell, DR; Clode, S; Fan, MQ; et al. (2007b) Relationships between tissue levels of 2,3,7,8-tetrachlorodibenzo-p-
17 dioxin (TCDD), mRNAs, and toxicity in the developing male Wistar (Han) rat. Toxicol Sci 99(2):591-604.
18 Bernert, JT; Turner, WE; Patterson, DG; et al. (2007) Calculation of serum total lipid concentrations for the
19 adjustment of persistent organohalogen toxicant measurements in human samples. Chemosphere 68:824-831.
20 Bertazzi, PA; Zocchetti, C; Pesatori, AC; et al. (1989) Ten-year mortality study of the population involved in the
21 Seveso incident in 1976. Am J Epidemiol 129(6): 1187-1200.
22 Bertazzi, A; Pesatori, AC; Consonni, D; et al. (1993) Cancer incidence in a population accidentally exposed to
23 2,3,7,8-tetrachlorodibenzo-para-dioxin. Epidemiology 4(5):398-406.
24 Bertazzi, PA; Zocchetti, C; Guercilena, S; et al. (1997) Dioxin exposure and cancer risk: a 15-year mortality study
25 after the "Seveso accident". Epidemiology 8(6):646-652.
26 Bertazzi, PA; Consonni, D; Bachetti, S; et al. (2001) Health effects of dioxin exposure: a 20-year mortality study.
27 Am J Epidemiol 153(11): 1031-1044.
28 Birnbaum, LS. (1986) Distribution and excretion of 2,3,7,8-tetrachlorodibenzo-p-dioxin in congenic strains of mice
29 which differ at the Ah locus. Drug Metab Dispos 14(l):34-40.
30 Black, JW: Leff, P. (1983) Operational models of pharmacological agonism. Proc. R. Soc. Lond. B 220:141-162.
31 Blankenship, A; Matsumura, F. (1997) 2,3,7,8-Tetrachlorodibenzo-p-dioxin-induced activation of a protein tyrosine
32 kinase pp60src in murine hepatic cytosol using a cell-free system. Mol Pharmacol 52: 667-675.
33 Bock, KW. (1994) Aryl hydrocarbon or dioxin receptor: biologic and toxic responses. Rev Physiol Biochem
34 Parmacol 125:1-42.
35 Bock, KW; Gschaidmeier, H; Heel, H; et al. (1998) AH receptor-controlled transcriptional regulation and function
36 of rat and human UDP-glucuronosyltransferase isoforms. Adv Enzyme Regul 38:207-222.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-3 DRAFT—DO NOT CITE OR QUOTE
-------
1 Bodner, KM; Collins, JJ; Bloemen, LJ; et al. (2003) Cancer risk for chemical workers exposed to 2,3,7,8-
2 tetrachlorodibenzo-p-dioxin. Occup Environ Med 60(9):672-675.
3 Bond, GG; Wetterstroem, NH; Roush, GJ; et al. (1988) Cause specific mortality among employees engaged in the
4 manufacture, formulation, or packaging of 2,4-dichlorophenoxyacetic acid and related salts. Br J Ind Med
5 45(2):98-105.
6 Bond, GG; McLaren, EA; Brenner, FE; et al. (1989) Incidence of chloracne among chemical workers potentially
7 exposed to chlorinated dioxins. J Occup Med 31(9):771-774.
8 Boverhoff, DR; Burgoon, LD; Tashiro, C; et al. (2005) Temporal and dose-dependent hepatic gene expression
9 patterns in mice provide new insights into TCDD-mediated hepatotoxicity. Toxicol Sci 85:1048-1063.
10 Bowman, RE; Schantz, SL; Weerasinghe, NCA; et al. (1989a) Chronic dietary intake of 2,3,7,8-tetrachlorobibenzo-
11 p-dioxin (TCDD) at 5 or 25 parts per trillion in the monkey: TCDD kinetics and dose-effect estimate of reproductive
12 toxicity. Chemosphere 18:243-252.
13 Bowman, RE; Schantz, SL; Gross, ML; et al. (1989b) Behavioral effects in monkeys exposed to 2,3,7,8-TCDD
14 transmitted maternally during gestation and for four months of nursing. Chemosphere 18:235-242.
15 Brand, KP; Rhomberg, L; Evans, JS. (1999) Estimating noncancer uncertainty factors: are ratios NOAELS
16 informative? Risk Anal 19(2):295-308. Available at:
17 http://www.springerlink.com/content/1376403247ul4513/fulltext.pdf.
18 Brand, KP; Catalano, PJ; Hammitt, JK; et al. (2001) Limitations to empirical extrapolation studies: the case of BMD
19 ratios. Risk Anal 21(4):625-640.
20 Bretagnolle, J; Huber-Carol, C. (1988) Effects of omitting covariates in Cox's model of survival data. Scand J Stat
21 15:125-138.
22 Brouwer, A; Morse, DC; Lans, MC; et al. (1998) Interactions of persistent environmental organohalogens with the
23 thyroid hormone system: mechanisms and possible consequences for animal and human health. Toxicol Indust
24 Health 14(l-2):59-84.
25 Brown, J; Goossens, LH; Kraan, BCP. (1997) Probabilistic accident consequence uncertainty study: food chain
26 uncertainty assessment. Report Number: NUREG/CR-6523, EUR 16771, SAND97-0335. Nuclear Regulatory
27 Commission, Washington, DC; Commission of the European Communities, Brussels; 74 pp.
28 Brown, NM; Manzolillo, PA; Zhang, JX; et al. (1998) Prenatal TCDD and predisposition to mammary cancer in the
29 rat. Carcinogenesis 19(9): 1623-1629.
30 Budinsky, RA; Paustenbach, D; Fontaine, D; et al. (2006) Recommended relative potency factors for 2,3,4,7,8
31 pentachlorodibenzofuran: the impact of different dose metrics. Toxicol Sci 91(l):275-285.
32 Buelke-Sam, J; Holson, JF; Nelson, CJ. (1982a) Blood flow during pregnancy in the rat: II. Dynamics of and litter
33 variability in uterine flow. Teratology 26: 279-288.
34 Buelke-Sam, J; Nelson, CJ; Byrd, RA; et al. (1982b) Blood flow during pregnancy in the rat: I. Flow patterns to
35 maternal organs. Teratology 26:269-277.
36 Bueno de Mesquita, HB; Doornbos, G; van der Kuip, DAM; et al. (1993) Occupational exposure to phenoxy
37 herbicides and chlorophenols and cancer mortality in The Netherlands. Am J Ind Med 23:289-300.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-4 DRAFT—DO NOT CITE OR QUOTE
-------
1 Burleson, GR; Lebrec, H; Yang, YG; et al. (1996) Effect of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) on
2 influenza vims host resistance in mice. Fund Appl Toxicol 29:40-47.
3 Bussard, D; Preuss, P; White, P. (2009) Conclusions. In: Cooke, RM; ed. Uncertainty modeling in dose response:
4 bench testing environmental toxicity. Statistics in Practice. New York, NY: Wiley.
5 Cantoni, L; Salmona, M; Rizzardini, M. (1981) Porphyrogenic effect of chronic treatment with 2,3,7,8-
6 tetrachlorodibenzo-p-dioxin in female rates. Dose-effect relationship following urinary excretion of porphyrins.
7 Toxicol Appl Pharmacol 57:156-157.
8 Carrier, G; Brunet, RC; Brodeur, J. (1995a) Modeling of the toxicokinetics of polychlorinated dibenzo-p-dioxins
9 and dibenzofurans in mammalians, including humans. I. Nonlinear distribution of PCDD/PCDF body burden
10 between liver and adipose tissues. Toxicol Appl Pharmacol 131(2):253-266.
11 Carrier, G; Brunet, RC; Brodeur, J. (1995b) Modeling of the toxicokinetics of polychlorinated dibenzo-p-dioxins
12 and dibenzofurans in mammalians, including humans. II. Kinetics of absorption and disposition of PCDDs/PCDFs.
13 Toxicol Appl Pharmacol 131(2):267-276.
14 Cesana, GC; de Vito, G; Ferrario, M; et al. (1995) Trends of smoking habits in northern Italy (1986-1990). Euro J
15 Epidemiol 11(3):251-258.
16 Checkoway, H; Pearce, N; Crawford-Brown, DJ. (1989) Research methods in occupational epidemiology. New
17 York, NY: Oxford University Press; 366 pp.
18 Cheng, H; Aylward, L; Beall, C; et al. (2006) TCDD exposure-response analysis and risk assessment. Risk Anal
19 26(4):1059-1071.
20 Chevrier, J; Eskenazi, B; Bradman, A; et al. (2007) Associations between prenatal exposure to polychlorinated
21 biphenyls and neonatal thyroid-stimulating hormone levels in a Mexican-American population, Salinas Valley,
22 California. Environ Health Perspect 115:1490-1496.
23 Chiaro, C; Morales, JL; Prabhu, KS; et al. (2008) Leukotriene A4 metabolites are endogenous ligands for the AH
24 receptor. Biochemistry 47(32):8445-8455.
25 Choi, BC. (1992) Definition, sources, magnitude, effect modifiers, and strategies of reduction of the healthy worker
26 effect. JOccupMed34(10):979-988.
27 Chu, I; Lecavalier, P; Hakansson, H; et al. (2001) Mixture effects of 2,3,7,8-tetrachlorodibenzo-p-dioxin and
28 polychlorinated biphenyl congeners in rats. Chemosphere 43:807-814.
29 Clark, GC; Tritscher, A; Maronpot, R; et al. (1991) Tumor promotion by TCDD infemale rats. In: Gallo, M;
30 Scheuplein, R; van der Heijden, K; eds. Banbury Report 35: biological basis for risk assessment of dioxin and
31 related compounds. Cold Spring Harbor, NY: Cold Spring Harbor Laboratory; pp. 389-404.
32 Clegg, LX; Li, FP; Hankey, BF; et al. (2002) Cancer survival among US whites and minorities: a SEER
33 (Surveillance, Epidemiology, and End Results) program population-based study. Arch Intern Med 162:1985-1993.
34 Clewell, HJ; Gentry, PR; Covington, TR; et al. (2004) Evaluation of the potential impact of age- and gender-specific
35 pharmacokinetic differences on tissue dosimetry. Toxicol Sci 79(2):381-93.
36 Cole, P; Trichopoulos, D; Pastides, H; et al. (2003) Dioxin and cancer: a critical review. Regul Toxicol Pharmacol
37 38(3):378-388.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-5 DRAFT—DO NOT CITE OR QUOTE
-------
1 Collins, JJ; Bodner, K; Aylward, LL; et al. (2009) Mortality rates among trichlorophenol workers with exposure to
2 2,3,7,8-tetrachlorodibenzo-p-dioxin. Am JEpid 170(4):501-506.
3 Connor, KT; Aylward, L. (2006) Human response to dioxin: aryl hydrocarbon receptor (AhR) molecular structure,
4 function, and dose-response data for enzyme induction indicate an impaired human AhR. J Toxicol Environ Health
5 B 9(2):147-171.
6 Consonni, D; Pesatori, AC; Zocchetti, C; et al. (2008) Mortality in a population exposed to dioxin after the Seveso,
7 Italy, accident in 1976: 25 years of follow-up. Am J Epidemiol 167(7):847-858.
8 Cooke, RM; ed. (2009) Uncertainty modeling in dose response: bench testing environmental toxicity. Statistics in
9 Practice. New York, NY: Wiley, John & Sons, Inc; 248 pp.
10 Cooper, GS; Klebanoff, MA; Promislow, J; et al. (2005) Polychlorinated biphenyls and menstrual cycle
11 characteristics. Epidemiology 16(2): 191-200.
12 Cox, DR. (2006) Combination of data. In: Encyclopedia of Statistical Sciences, Kotz, S; Read, CB; Balakrishnan, N;
13 et al., editors. Hoboken: Wiley, pp. 1074-1081.
14 Crofton, KM; Craft, ES; Hedge, JM; et al. (2005) Thyroid-hormone-disrupting chemicals: evidence for dose-
15 dependent additivity or synergism. Environ Health Perspect 113(11): 1549-1554.
16 Croutch, CR; Lebofsky, M; Schramm, K-W; et al. (2005) 2,3,7,8-tetrachlorordibenzo-p-dioxin (TCDD) and
17 1,2,3,4,7,8-hexachlorodibenzo-p-dioxin (HxCDD) alters body weight by decreasing insulin-like growth factor I
18 (IGF-I) signaling. Toxicol Sci 85:560-571.
19 Crump, K. 2002. Critical issues in benchmark calculations from continuous data. Crit Rev. Toxicol. 21:133-153.
20 Crump, KS; Hoel, DG; Langley, CH; et al. (1976) Fundamental carcinogenic processes and their implications for
21 low dose risk assessment. Cancer Research 36:2973-2979.
22 Crump, KS; Canady, R; Kogevinas, M. (2003) Meta-analysis of dioxin cancer dose response for three occupational
23 cohorts. Environ Health Perspect 111(5):681—687.
24 D'Amico, M; Agozzino, E; Biagino, A; et al. (1999) Ill-defined and multiple causes on death certificates~a study of
25 misclassification in mortality statistics. Eur J Epidemiol 15(2): 141—148.
26 DeCaprio, AP; McMartin, DN; O'Keefe, PE; et al. (1986) Subchronic oral toxicity of 2,3,7,8-tetrachlorodibenzo-p-
27 dioxin in the guinea pig: comparisons with a PCB-containing transformer fluid pyrolysate. Fund Appl Toxicol
28 6:454-463.
29 DeKoning, EP; Karmaus, W. (2000) PCB exposure in utero and via breast milk. A review. J Expo Anal Environ
30 Epidemiol 10:285-293.
31 Devito, MJ; Ma, X; Babish, JG; et al. (1994) Dose-response relationships in mice following subchronic exposure to
32 2,3,7,8-tetrachlorodibenzo-p-dioxin: CYP1A1, CYP1A2, estrogen receptor, and protein tyrosine phosphoylation.
33 Toxicol Appl Pharmacol 124:82-90.
34 Diliberto, JJ; Akubue, PI; Luebke, RW; et al. (1995) Dose-response relationships of tissue distribution and induction
35 of CYP1A1 and CYP1A2 enzymatic activities following acute exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin
36 (TCDD) in mice. Toxicol Appl Pharmacol 130(2): 197-208.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-6 DRAFT—DO NOT CITE OR QUOTE
-------
1 Diliberto, JJ; Jackson, JA; Birnbaum, LS. (1996) Comparison of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)
2 disposition following pulmonary, oral, dermal and parenteral exposures to rats. Toxicol Appl Pharmacol 138:158-
3 168.
4 Diliberto, JJ; Burgin, DE; Birnbaum, LS. (1997) Role of CYP1A2 in hepatic sequestration of dioxin: studies using
5 CYP1A2 knock-out mice. Biochem Biophys Res Commun 236:431-433.
6 Diliberto, JJ; Burgin, DE; Birnbaum, LS. (1999) Effects of CYP1A2 on disposition of 2,3,7,8-tetrachlorodibenzo-p-
7 dioxin, 2,3,4,7,8-pentachlorodibenzofuran, and 2,2',4,4',5,5'-hexachlorobiphenyl in CYP1A2 knockout and parental
8 (C57BL/6N and 129/Sv) strains of mice. Toxicol Appl Pharmacol 159(l):52-64.
9 Diliberto, JJ; DeVito, MJ; Ross, DG; et al. (2001) Subchronic exposure of [3H]-2,3,7,8-tetrachlorodibenzo-p-dioxin
10 (TCDD) in female B6C3F1 mice: relationship of steady-state levels to disposition and metabolism. Toxicol Sci
11 61:241-255.
12 Dolwick, KM; Schmidt, JV; Carver, LA; et al. (1993) Cloning and expression of a human Ah receptor cDNA. Mol
13 Pharmacol 44(5):911-917.
14 Dunson, DB; Baird, DD. (2001) A flexible parametric model for combining current status and age at first diagnosis
15 data. Biometrics 57(2):396-403.
16 EC (European Commission). (2009) CORDIS, Community Research and Development Information Service.
17 Nuclear Energy Library, Archives (archived July 17). Available at: http://cordis.europa.eu/fp5-
18 euratom/src/lib_docs.htm.
19 Ema, M; Ohe, N; Suzuki, M; et al. (1994) Dioxin binding activities of polymorphic forms of mouse and human
20 arylhydrocarbon receptors. J Biol Chem269(44):27337-27343.
21 Emond, C; Birnbaum, LS; DeVito, M. (2004) Physiologically based pharmacokinetic model for developmental
22 exposures to TCDD in the rat. Toxicol Sci 80(1): 115—133.
23 Emond, C; Michalek, JE; Birnbaum, L; et al. (2005) Comparison of the use of a physiologically-based
24 pharmacokinetic model and a classical pharmacokinetic models for dioxin exposure assessments. Environ. Health
25 Perspect 113(12): 1666-1668.
26 Emond, C; Birnbaum, LS; Devito, MJ. (2006) Use of a physiologically based pharmacokinetic model for rats to
27 study the influence of body fat mass and induction of CYP1A2 on the pharmacokinetics of TCDD. Environ. Health
28 Perspect 114(9): 1394-1400.
29 Eskenazi, B; Mocarelli, P; Warner, M; et al. (2000) Seveso women's health study: a study of the effects of 2,3,7,8-
30 tetrachlorodebenzo-p-dioxin on reproductive health. Chemosphere 40:1247-1253.
31 Eskenazi, B; Warner, M; Mocarelli, P; et al. (2002a) Serum dioxin concentrations and menstrual cycle
32 characteristics. Am J Epidemiol 156(4):383-392. Available at: http://aje.oxfordjournals.Org/cgi/reprint/156/4/383.
33 Eskenazi, B; Mocarelli, P; Warner, M; et al. (2002b) Serum dioxin concentrations and endometriosis: a cohort study
34 in Seveso, Italy. Environ Health Perspect 110(7):629-634.
3 5 Eskenazi, B; Mocarelli, P; Warner, M; et al. (2003) Maternal serum dioxin levels and birth outcomes in women of
36 Seveso, Italy. Environ Health Perspect 111(7):947-953.
37 Eskenazi, B; Mocarelli, P; Warner, M; et al. (2004) Relationship of serum TCDD concentrations and age at
3 8 exposure of female residents of Seveso, Italy. Environ Health Perspect 112(l):22-27.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-7 DRAFT—DO NOT CITE OR QUOTE
-------
1 Eskenazi, B; Warner, M; Marks, AR; et al. (2005) Serum dioxin concentrations and age at menopause. Environ
2 Health Perspect 113(7):858-862.
3 Eskenazi, B; Warner, M; Samuels, S; et al. (2007) Serum dioxin concentrations and risk of uterine leiomyoma in the
4 Seveso Women's Health Study. Am J Epidemiol 166(l):79-87.
5 Fattore, E; Trossvik, C; Hakansson, H. (2000) Relative potency values derived from hepatic vitamin A reduction in
6 male and female Sprague-Dawley rats following subchronic dietary exposure to individual polychlorinated dibenzo-
7 p-dioxin and dibenzofuran congeners and a mixture thereof. Toxicol and Appl Pharmacol 165:184-194.
8 Fernandez-Salguero, PM; Hilbert, DM; Rudikoff, S; et al. (1996) Aryl-hydrocarbon receptor-deficient mice are
9 resistant to 2,3,7,8-tetrachlorodibenzo-p-dioxin-induced toxicity. Toxicol Appl Pharmacol 140:173-179.
10 Ferriby, LL; Knutsen, JS; Harris, M; et al. (2007) Evaluation of PCDD/F and dioxin-like PCB serum concentration
11 data from the 2001-2002 National Health and Nutrition Examination Survey of the United States population. J Expo
12 Sci Environ Epidemiol 17(4):358-371.
13 Fingerhut, MA; Halperin, WE; Marlow, DA; et al. (1991) Cancer mortality in workers exposed to 2,3,7,8-
14 tetrachlorodibenzo-p-dioxin. N Engl J Med 324(4):212-218.
15 Fisher, J. W., Whittaker, T. A., Taylor, D. H., Clewell, H. J., 3rd, and Andersen, M. E. (1989). Physiologically based
16 pharmacokinetic modeling of the pregnant rat: a multiroute exposure model for trichloroethylene and its metabolite,
17 trichloroacetic acid. Toxicol Appl Pharmacol 99:395-414.
18 Flesch-Janys, D. (1997) Analyses of exposure to polychlorinated dibenzo-p-dioxins, furans, and
19 hexachlorocyclohexane and different health outcomes in a cohort of former herbicide-producing workers in
20 Hamburg, Germany. Teratog Carcinog Mutagen 17(4-5):257-264.
21 Flesch-Janys, D; Gurn, P; Jung, D. (1994) First results of an investigation of the elimination of polychlorinated
22 dibenzo-p-dioxins and dibenzofurans (PCDD/F) in occupationally exposed persons. Organohal Comp (21):93-99.
23 Flesch-Janys, D; Berger, J; Gurn, P; et al. (1995) Exposure to polychlorinated dioxins and furans (PCDD/F) and
24 mortality in a cohort of workers from a herbicide-producing plant in Hamburg, Federal Republic of Germany. Am J
25 Epidemiol 142(11): 1165-1175.
26 Flesch-Janys, D; Becher, H; Gurn, P; et al. (1996) Elimination of polychlorinated dibenzo-p-dioxins and
27 dibenzofurans in occupationally exposed persons. J Tox Environ Health 47(4):363-378.
28 Flesch-Janys, D; Steindorf, K; Gurn, P; et al. (1998) Estimation of the cumulated exposure to polychlorinated
29 dibenzo-p-dioxins/furans and standardized mortality ratio analysis of cancer mortality by dose in an occupationally
30 exposed cohort. Environ Health Perspect 106(Suppl 2):655-662.
31 Flodstrom, S; Ahlborg, UG. (1991) Promotion of hepatocarcinogenesis in rats by PCDDs and PCDFs. In: Gallo,
32 MA; Scheuplein, RJ; van der Heijden, KA; eds. Banbury Report 35: biological basis for risk assessment of dioxin
3 3 and related compounds. Cold Spring Harbor, NY: Cold Spring Harbor Laboratory; pp. 405-414.
34 Fox, TR; Best, LL; Goldsworthy, SM; et al. (1993) gene expression and cell proliferation in rate liver after 2,3,7,8-
35 tetrachlorodibenzo-p-dioxin exposure. Cancer Research 53:2265-2271.
36 Franczak, A; Nynca, A; Valdez, KE; et al. (2006) Effects of acute and chronic exposure to the aryl hydrocarbon
37 receptor agonist 2,3,7,8-tetrachlorodibenzo-p-dioxin on the transition to reproductive senescence in female Sprague-
38 Dawley rats. Biol Reprod 74:125-130.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-8 DRAFT—DO NOT CITE OR QUOTE
-------
1 Fretland, AJ; Safe, S; Hankinson, O. (2004) Lack of antagonism of 2,3,7,8-tetrachlorodibenzo-p-dioxin's (TCDDs)
2 induction of cytochrome P4501A1 (CYP1A1) by the putative selective aryl hydrocarbon receptor modulator 6-alkyl-
3 1,3,8-trichlorodibenzofuran (6-MCDF) in the mouse hepatoma cell line Hepa-lclc7. Chemico-Biol Interac
4 150:161-170.
5 Fritz, W; Lin, TM; Safe, S; et al. (2009) The selective aryl hydrocarbon receptor modulator 6-methyl-l,3,8-
6 trichlorodibenxofuran inhibits prostate tumor metastasis in TRMP mice. Biochem Pharmacol 77(7): 1151-1160.
7 Fujii-Kuriyama, Y; Ema, M; Mimura, J; et al. (1995) Polymorphic forms of the Ah receptor and induction of the
8 CYP1A1 gene. Pharmacogenetics 5(S): 149-153.
9 Gasiewicz, TA; Henry, EC; Collins, LL. (2008) Expression and activity of aryl hydrocarbon receptors in
10 development and cancer. Crit Rev Eukaryot Gene Expr 18:279-321.
11 Gaylor, DW; Kodell, RL. (2000) Percentiles of the product of uncertainty factors for establishing probabilistic risk
12 doses. Risk Anal 20(2):245-250.
13 Ge, N-L; Elferink, CJ. (1998) A direct interation between the aryl hydrocarbon receptor and retinoblastoma protein.
14 J Biol Chem 273(35):22708-22713.
15 Geusau, A; Schmaldienst, S; Derfler, K; et al. (2002) Severe 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)
16 intoxication: kinetics and trials to enhance elimination in two patients. Arch Toxicol 76:316-325.
17 Geyer, H; Scheunert, I; Korte, F. (1986) Bioconcentration potential of organic environmental chemicals in humans.
18 Regul Toxicol Pharmacol 6(4):313-347.
19 Geyer, HJ; Scheuntert, I; Rapp, K; et al. (1990) Correlation between acute toxicity of 2,3,7,8-tetrachlorodibenzo-p-
20 dioxin (TCDD) and total body fat content in mammals. Toxicology 65(l-2):97-107.
21 Geyer, HJ; Schramm, K-W; Scheunert, I; et al. (1997) Considerations on genetic and environmental factors that
22 contribute to resistance or sensitivity of mammals including humans to toxicity of 2,3,7,8-tetrachlorodibenzo-p-
23 dioxin (TCDD) and related compounds. Ecotoxicol Environ Safety 36:213-230.
24 Gielen, JE; Nebert, DW. (1971) Aryl hydrocarbon hydroxylase induction in mammalian liver cell culture. I.
25 Stimulation of enzyme activity in nonhepatic cells and in hepatic cells by phenobarbital, polycyclic hydrocarbons,
26 and 2,2-bis(p-chlorophenyl)-l,l,l-trichloroethane. J Biol Chem 246:5189-5198.
27 Goodman, DG; Sauer, RM. (1992) Hepatotoxicity and carcinogenicity in female Sprague-Dawley rats treated with
28 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD): a pathology working group reevaluation. Regulat Toxicol Pharmacol
29 15:245-252.
30 Goossens, LH; Cooke, RM; Kraan, BCP. (1996) Evaluation of weighting schemes for expert judgment studies.
31 Commission of European Communities, Directorate-General for Science, Research and Development, XII-F-6, Delft
32 University of Technology, Delft, The Netherlands.
33 Goossens, LH; Boardman, J; Harper, FT; et al. (1997) Probabilistic accident consequence uncertainty study:
34 uncertainty assessment for deposited material and external doses. Washington, DC, Brussels-Luxembourg;
35 NUREG/CR-6526, EUR 16772, SAND97-2323.
36 Goossens, LH; Harrison, JD; Harper, FT; et al. (1998) Probabilistic accident consequence uncertainty study:
37 uncertainty assessment for internal dosimetry. Washington, DC, Brussels-Luxembourg; NUREG/CR-6571, EUR
38 16773, SAND98-0119.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-9 DRAFT—DO NOT CITE OR QUOTE
-------
1 Goossens, LH; Kraan, BCP; Cooke, RM; et al. (2001a) Methodology and processing techniques. EUR 18827EN,
2 European Commission, Luxembourg.
3 Goossens, LH; Kraan, BCP; Cooke, RM; et al. (2001b) Overall uncertainty analysis. European Commission,
4 Luxembourg; Project Report, EUR 18826EN.
5 Goossens, LH; Kraan, BCP; Cooke, RM; et al. (2001c) Nuclear science and technology countermeasures uncertainty
6 assessment. European Commission, Luxembourg; EUR 18821EN.
7 Goossens, LH; Kraan, BCP; Cooke, RM; et al. (200 Id) Uncertainty from the atmospheric dispersion and deposition
8 module. European Commission, Luxembourg; EUR 18822EN.
9 Goossens, LH; Kraan, BCP; Cooke, RM; et al. (200le) Uncertainty from the early and late health effects module.
10 European Commission, Luxembourg; EUR 18824EN.
11 Goossens, LH; Kraan, BCP; Cooke, RM; et al. (200 If) Uncertainty from the food chain module. European
12 Commission, Luxembourg; EUR 18823EN.
13 Goossens, LH; Kraan, BCP; Cooke, RM; et al. (200 lg) Uncertainty from the dose module. European Commission,
14 Luxembourg; Project Report, EUR 18825EN.
15 Graham, MJ; Lucier, GW; Linko, P; et al. (1988) Increases in cytochrome P-450 mediated 17B-estradiol 2-
16 hydroxylase activity in rat liver microsomes after both acute administration and subchronic administration of
17 2,3,7,8-tetrachlorodibenzo-p-dioxin in a two-stage hepatocarcinogenises model. Carcinogenesis 9(11): 1935-1941.
18 Grassman, JA; Needham, LL; Masten, S A; et al. (2000) Evidence of hepatic sequestration of dioxin in humans? An
19 examination of tissue levels and CYP1A2 expression. Organohal Comp 48:87-90.
20 Greenlee, WF; Hushka, LJ; Hushka, DR. (2001) Molecular basis of dioxin actions: evidence supporting
21 chemoprotection. Toxicol Path 29(l):6-7. Available at: http://tpx.sagepub.eom/cgi/reprint/29/l/6.
22 Guess, HA; Hoel, DG. (1977) The effect of dose on cancer latency period. J Environ Pathol Tox l(2):279-286.
23 Haarmann-Stemmann, T; Bothe, H; Abel, J. (2009) Growth factors, cytokines and their receptors as downstream
24 targets of arylhydrocarbon receptor (AhR) signaling pathways. Biochem Pharmacol 77:508-520.
25 Hahn, ME; Allan, LL; Sherr, DH. (2009) Regulation of constitutive and inducible AHR signaling: complex
26 interactions involving the AHR repressor. Biochem Pharmacol 77(4):485-497.
27 Hakk, H; Diliberto, JJ; Birnbaum, LS. (2009) The effect of dose on 2,3,7,8-TCDD tissue distribution, metabolism
28 and elimination in CYP1A2 (-/-) knockout and C57BL/6N parental strains of mice. Toxicol Appl Pharmacol Nov
29 15; 241(1): 119—126. Epub 2009 Aug 18.
30 Harper, FT; Goossens, LH; Cooke, RM; et al. (1995) Probabilistic accident consequence uncertainty study:
31 dispersion and deposition uncertainty assessment. Washington, DC, Brussels-Luxembourg; NUREG/CR-6244,
32 EUR 15855 EN, SAND94-1453.
33 Harper, PA; Wong, JMY; Lam, MSM; et al. (2002) Polymorphisms in the human AH receptor. Chemico-Biol
34 Interac 141(1-2): 161-187.
35 Harrad, S; Wang, Y; Sandaradura, S; et al. (2003) Human dietary intake and excretion of dioxin-like compounds. J
3 6 Environ Monit 5:224-228.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-10 DRAFT—DO NOT CITE OR QUOTE
-------
1 Hassoun, EA; Wilt, SC; DeVito, MJ; et al. (1998) Induction of oxidative stress in brain tissues of mice after
2 subchronic exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol Sci 42:23-27.
3 Hassoun, EA; Li, F; Abushaban, A; et al. (2000) The relative abilities of TCDD and its congeners to induce
4 oxidative stress in the hepatic and brain tissues of rats after subchronic exposure. Toxicology 145:103-113.
5 Hassoun, EA; Wang, H; Abushaban, A. (2002) Induction of oxidative stress following chronic exposure to TCDD,
6 2,3,4,7,8-pentachlorodibenzofuran, and 2,3 ',4,4',5-pentachlorobiphenyl. J Toxicol Environ Health A 65:825-842.
7 Hassoun, EA; Al-Ghafri, M; Abushaban, A. (2003) The role of antioxidant enzymes in TCDD-induced oxidative
8 stress in various brain regions of rats after subchronic exposure. Free Rad Biol Medicine 35(9): 1028-1036.
9 Hattis, D. (1996) Human interindividual variability in susceptibility to toxic effects-from annoying detail to a
10 central determinant of risk. Toxicology 111:5-14.
11 Hattis, D. (2009) High-throughput testing - the NRC vision, the challenge of modeling dynamic changes in
12 biological systems, and the reality of low-throughput environmental health decision making. Risk Anal
13 29(4):483-484.
14 Hattis, D; Burmaster, DE. (1994) Assessment of variability and uncertainty distributions for practical risk analyses.
15 Risk Anal 14:713-730.
16 Hattis, D; Banati, P; Goble, R. (1999) Distributions of individual susceptibility among humans for toxic effects~for
17 what fraction of which kinds of chemicals and effects does the traditional 10-fold factor provide how much
18 protection?" Ann NY Acad Sci 895:286-316.
19 Hattis, D; Baird, S; Goble, R. (2002) A straw man proposal for a quantitative definition of the RfD. Drug Chem
20 Toxicol 25(4):403-436.
21 Hattis, D; Ginsberg, G; Sonawane, B; et al. (2003) Differences in pharmacokinetics between children and adults-II.
22 Childrens variability in drug elimination half-lives and in some parameters needed for physiologically-based
23 pharmacokinetic modeling. Risk Anal 23:117-142.
24 Haws, LC; Su, SH; Harris, M; et al. (2006) Development of a refined database of mammalian relative potency
25 estimates for dioxin-like compounds. Toxicol Sci 89(l):4-30.
26 Henck, JM; New, MA; Kociba, RJ; et al. (1981) 2,3,7,8-Tetrachlorodibenzo-p-dioxin: acute oral toxicity in
27 hamsters. Toxicol Appl Pharmacol 59:405-407.
28 Henrikson, GL; Ketchum, NS; Michalek, JE; et al. (1997) Serum dioxin and diabetes mellitus in veterans of
29 operation ranch hand. Epidemiology 8(3):252-258.
30 Hertz-Picciotto, I. (1995) Epidemiology and quantitative risk assessment: a bridge from science to policy. Am J
31 Public Health 85(4):484-491.
32 Higgins, JPT; Thompson, SG; Spiegelhalter, DJ. (2009) Re-evaluation of random-effects meta analysis. J R Stat
33 Soc Series A. 172(1): 137-159.
34 Hochstein, JR; Render, JA; Bursian, SJ; Aulerich, RJ. (2001) Chronic toxicity of dietary 2,3,7,8-tetrachlorodibenzo-
35 p-dioxin to mink. Vet Human Toxicol 43(3): 134-139.
36 Hoel, DG; Portier, CJ. (1994) Nonlinearity of dose-response functions for carcinogenicity. Environ Health Perspec
37 102(Suppl 1): 109—113. Available at:
38 http://www.pubmedcentral.nih.gov/picrender.fcgi?artid=1566901&blobtype=pdf.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
1 Hoglund, M; Sehn, L; Connors, JM; et al. (2004) Identification of cytogenetic subgroups and karyotypic pathways
2 of clonal evolution in follicular lymphomas. Genes Chromos Cancer 39(3): 195-204.
3 Hojo, R; Stern, S; Zareba, G; et al. (2002) Sexually dimorphic behavioral responses to prenatal dioxin exposure.
4 Environ Health Perspect 110(3):247-254.
5 Hooiveld, M; Heederik, DJ; Kogevinas, M; et al. (1998) Second follow-up of a Dutch cohort occupationally
6 exposed to phenoxy herbicides, chlorophenols, and contaminants. Am J Epidemiol 147(9):891-901.
7 Huff, JE. (1992) 2,3,7,8-TCDD: a potent and complete carcinogen in experimental animals. Chemosphere
8 25:173-176.
9 Huff, JE; Salmon, AG; Hooper, NK; et al. (1991) Long-term carcinogenesis studies on 2,3,7,8-tetrachlorodibenzo-p-
10 dioxin and hexachlorodibenzo-p-dioxins. Cell Biol Toxicol 7(l):67-93.
11 Hurst, CH; Abbott, BD; DeVito, MJ; et al. (1998) 2,3,7,8-tetrachlorodibenzo-p-dioxin in pregnant Long Evans rats:
12 disposition to maternal and embryo/fetal tissues. Toxicol Sci 45:129-136.
13 Hurst, CH; DeVito, MJ; Birnbaum, LS. (2000a) Tissue disposition of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)
14 in maternal and developing Long-Evans rats following subchronic exposure. Toxicol Sci 57(2):275-283.
15 Hurst, CH; DeVito, MJ; Setzer, RW; et al. (2000b) Acute administration of 2,3,7,8-tetrachlorodibenzo-p-dioxin
16 (TCDD) in pregnant Long Evans rats: association of measured tissue concentrations with developmental effects.
17 Toxicol Sci 3(2):411-420.
18 IARC (International Agency for Research on Cancer). (1997) IARC Monographs on the Evaluation of Carcinogenic
19 Risks to Humans. Vol. 69. Polychlorinated dibenzo-para-dioxins and polychlorinated dibenzofurans. Lyon, France:
20 International Agency for Research on Cancer; 666 pp.
21 Ikeda, M; Tamura, M; Yamashita, J; et al. (2005) Repeated in utero and lactational 2,3,7,8-tetrachlorodibenzo-/?-
22 dioxin exposure affects male gonads in offspring, leading to sex ratio changes in F2 progeny. Toxicol Appl
23 Pharmacol 206:351-355.
24 ILSI (International Life Sciences Institute). (1994) Physiological parameter values for PBPK models. Washington,
25 DC: Risk Science Institute.
26 Institute of Medicine. (1994) Veterans and Agent Orange. Washington, DC: National Academies Press; 832 pp.
27 Institute of Medicine. (2006) Veterans and Agent Orange: update 2000. 7th edition. Washington, DC: National
28 Academies Press; 896 pp.
29 Ishihara, K; Warita, K; Tanida, T; et al. (2007) Does paternal exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin
30 (TCDD) affect sex ratio of offspring? J Vet Med Sci 69(4):347-352.
31 James, WH. (1995) What stabilizes the sex ratio? Ann Hum Genet 59(2):243-249.
32 Jorgensen, N; Andersen, AG; Eustache, F; et al. (2001) Regional differences in semen quality in Europe. Hum
33 Reprod 16:1012-1019.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-12 DRAFT—DO NOT CITE OR QUOTE
-------
1 Kang, S-H; Kodell, RL; Chen, JJ. (2000) Incorporating model uncertainties along with data uncertainties in
2 microbial risk assessment. Regul Toxicol Pharmacol 32:68-72. Available at:
3 http://www.sciencedirect.com/science ?_ob=ArticleURL&_udi=B6WPT-45BCPMP-
4 14&_user=1722207&_rdoc=l&_fmt=&_orig=search&_sort=d&_docanchor=&view=c&_searchStrId=1043774060
5 &_rerunOrigin=google&_acct=C000053990&_version=l&_urlVersion=0&_userid=1722207&md5=29e6d3709cdb
6 a2152a90d24cea739a68.
7 Kattainen, H; Tuukanan, J; Simanainen, U; et al. (2001) In utero/lactational 2,3,7,8-tetrachlorodibenzo-p-dioxin
8 exposure impairs molar tooth development in rats. Toxicol Appl Pharmacol 17:216-224.
9 Kauppinen, TP; Partanen, TJ; Hernberg, SG; et al. (1993) Chemical exposures in manufacture of phenoxy herbicides
10 and chlorophenols and in spraying of phenoxy herbicides. Amer J Indust Med 23(6):903-920.
11 Keller, JM; Huet-Hudson, YM; Leamy, LJ. (2007) Qualitative effects of dioxin on molars vary among inbred mouse
12 strains. Arch Oral Biol 52:450-454.
13 Keller, JM; Zelditch, ML; Huet, YM; et al. (2008a) Genetic differences in sensitivity to alterations of mandible
14 structure caused by the teratogen 2.3,7,8-tctrachlorodibenzo-/?-dioxin. Toxicol Pathol 36:1006-1013.
15 Keller, JM; Huet-Hudson, Y; Leamy, LJ. (2008b) Effects of 2,3,7,8-tetrachlorodibenzo-p-dioxin on molar
16 development among non-resistant inbred strains of mice: a geometric morphometric analysis. Growth Devel Aging
17 71:3-16.
18 Kerger, BD; Leung, HW; Scott, P; et al. (2006) Age- and concentration-dependent elimination half-life of 2,3,7,8-
19 tetrachlorodibenzo-p-dioxin in Seveso children. Environ Health Perspect 114(10): 1596-1602.
20 Kerger, BD; Leung, HW; Scott, PK; et al. (2007) Refinements on the age-dependent half-life model for estimating
21 child body burdens of polychlorodibenzodioxins and dibenzofurans. Chemosphere 67(9):S272-278.
22 Ketchum, NS; Michalek, JE; Burton JE. (1999) Serum dioxin and cancer in veterans of Operation Ranch Hand. Am
23 J Epidemiol 149(7):630-639.
24 Kim, AH; Kohn, MC; Nyska, A; et al. (2003) Area under the curve as a dose metric for promotional responses
25 following 2,3,7,8-tetrachlorodibenzo-p-dioxin exposure. Toxicol Appl Pharmacol 191:12-21.
26 Kitchin, KT; Woods, JS. (1979) 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) effects on hepatic microsomal
27 cytochrome P-448-mediated enzyme activities. Toxicol Appl Pharmacol 47:537-546.
28 Kociba, RJ; Keeler, PA; Park, GN; et al. (1976) 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD): results of a 13 week
29 oral toxicity study in rats. Toxicol Appl Pharmacol 35:553-574.
30 Kociba, RJ; Keyes, DG; Beyer, JE; et al. (1978) Results of a two-year chronic toxicity and oncogenicity study of
31 2,3,7,8-tetrachlorodibenzo-p-dioxin in rats. Toxicol Appl Pharmacol 46(2):279-303.
32 Kogevinas, M; Becher, H; Benn, T; et al. (1997) Cancer mortality in workers exposed to phenoxy herbicides,
33 chlorophenols, and dioxins. An expanded and updated international cohort study. Am J Epidemiol 145:1061-1075.
34 Kohn, MC; Melnick, RL. (2002) Biochemical origins of the non-monotonic receptor-mediated dose-response. J
35 Mol Endocrinol 29:113-123. Available at: http://jme.endocrinology-journals.Org/cgi/reprint/29/l/113.
36 Kohn, MC; Lucier, GW; Clark, GC; et al. (1993) A mechanistic model of effects of dioxin on gene expression in the
37 rat liver. Toxicol Appl Pharmacol 120(1): 138-154.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-13 DRAFT—DO NOT CITE OR QUOTE
-------
1 Kohn, MC; Sewall, CH; Lucier, GW; et al. (1996) A mechanistic model of effects of dioxin on thyroid hormones in
2 the rat. Toxicol Appl Pharmacol 136(l):29-48.
3 Kohn, MC; Walker, NJ; Kim, AH; et al. (2001) Physiological modeling of a proposed mechanism of enzyme
4 induction by TCDD. Toxicology 162:193-208.
5 Kolluri, SK; Weiss, C; Koff, A; et al. (1999) p27(Kipl) induction and inhibition of proliferation by the intracellular
6 Ah receptor in developing thymus and hepatoma cells. Genes Dev. 13:1742-1753.
7 Kopylev L; Chen C; White P. (2007) Towards quantitative uncertainty assessment for cancer risks: central
8 estimates and probability distributions of risk in dose-response modeling. Regul Toxicol Pharmacol 49(3):203-207.
9 Kopylev, L; Fox, J; Chen, C. (2009) Combining risks from several tumors using Markov chain Monte Carlo. In:
10 Cooke, R; ed. Uncertainty modeling in environmental toxicity dose response: bench testing the models. New York:
11 Wiley; 230 pp.
12 Kreuzer, PE; Csanady, GA; Baur, C; et al. (1997) 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) and congeners in
13 infants. A toxicokinetic model of human lifetime body burden by TCDD with special emphasis on its uptake by
14 nutrition. Arch Toxicol 71(6):383-400.
15 Krishnan, K; Andersen, ME. (1991) Interspecies scaling in pharmacokinetics. In: Rescingo, A; Thakkur, A; eds.
16 New trends in pharmacokinetics. New York, NY: Plenum Press; pp. 203-226.
17 Krowke, R; Chahoud, I; Baumann-Wilschke, I.; et al. (1989) Pharmacokinetics and biological activity of 2,3,7,8-
18 tetrachlorodibenzo-p-dioxin. 2. Pharmacokinetics in rats using a loading-dose/maintenance-dose regime with high
19 doses. Arch Toxicol 63:356-360.
20 Kuchiiwa, S; Cheng, S-B; Nagatomo, I; et al. (2002) In utero and lactational exposure to 2,3,7,8-tetrachlorodibenso-
21 /j-dioxin decreases serotonin-immunoreactive neurons in raphe nuclei of male mouse offspring. Neurosci Lett
22 317:73-76.
23 Kurowicka, D; Cooke, RM. (2006) Uncertainty analysis with high dimensional dependence modelling. West
24 Sussex, England: John Wiley & Sons; 302 pp.
25 LaKind, JS; Berlin, CM; Park, CN; et al. (2000) Methodology for characterizing distributions of incremental body
26 burdens of 2,3,7,8-TCDD and DDE from breast milk in North American nursing infants. J Toxicol Environ Health
27 A 59(8):605-639.
28 Lakshmanan, MR; Campbell, BS; Chirtel, SJ; et al. (1986) Studies on the mechanism of absorption and distribution
29 of 2,3,7,8-tetrachlorodibenzo-p-dioxin in the rat. J Pharmacol Exp Ther 239:673-677.
30 Landi, MT; Consonni, D; Patterson, DG, Jr; et al. (1998) 2,3,7,8-Tetrachlorodibenzo-p-dioxin plasma levels in
31 Seveso 20 years after the accident. Environ Health Perspect 106(5):273-277.
32 Larsen, JC. (2006) Risk assessments of polychlorinated dibenzo-p-dioxins, polychloriniated dibenzofurans, and
33 dioxin-like polychlorinated biphenyls in food. Mol Nutr Food Res 50(10):885-896.
34 Latchoumycandane, C; Mathur, PP. (2002) Effects of vitamin E on reactive oxygen species-mediated 2,3,7,8-
35 tetrachlorodi-benzo-p-dioxin toxicity in rat testis. J Appl Toxicol 22(5):345-351.
36 Latchoumycandane, C; Chitra, KC; Mathur, PP. (2002a) The effect of 2,3,7,8-tetrachlorodibenzo-p-dioxin on the
37 antioxidant system in mitochondrial and microsomal fractions in rat testis. Toxicology 171:127-135.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-14 DRAFT—DO NOT CITE OR QUOTE
-------
1 Latchoumycandane, C; Chitra, KC; Mathur, PP. (2002b) Induction of oxidative stress in rat epididymal sperm after
2 exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Arch Toxicol 76:113-118.
3 Latchoumycandane, C; Chitra, KC; Mathur, PP. (2003) 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) induces
4 oxidative stress in the epididymis and epididymal sperm of adult rats. Arch Toxicol 77:280-284.
5 Lawrence, GS; Gobas, FA. (1997) A pharmacokinetic analysis of interspecies extrapolation in dioxin risk
6 assessment. Chemosphere 35(3):427-452.
7 Lean, MEJ; Han, TS; Deurenberg, P. (1996) Predicting body composition by densitometry from simple
8 anthropometric measurements. Am J Clin Nut 63:4-14.
9 Lee, DJ; Fleming, LE; Arheart, KL; et al. (2007) Smoking rate trends in U.S. occupational groups: the 1987 to 2004
10 National Health Interview Survey. J Occup Environ Med 49(1):75-81.
11 Lehman, AJ; Fitzhugh, OG. (1954) 100-Fold margin of safety. Assoc Food Drug Off USQ Bull 18:33-35.
12 Leo, A; Hansch, C; Elkins, D. (1971) Partition coefficients and their uses. Chem Rev 71:525-616.
13 Leung, HW; Ku, RH; Paustenbach, DJ; et al. (1988) A physiologically based pharmacokinetic model for 2,3,7,8-
14 tetrachlorodibenzo-p-dioxin in C57BL/6J and DBA/2J mice. Toxicol Lett 42(l):15-28.
15 Leung, HW; Poland, A; Paustenbach, D; et al. (1990) Pharmacokinetics of (125-I)-2-lodo-3,7,8-trichlorodibenzo-p-
16 dioxininmice: analysis with a physiological modeling approach. Toxicol Appl Pharmacol 103:411-419.
17 Leung, HW; Kerger, BD; Paustenbach, DJ. (2006) Elimination half-lives of selected poly chlorinated dibenzodioxins
18 and dibenzofurans in breast-fed human infants. J Toxicol Environ Health A 69(6):437-443.
19 Li, B; Liu, H-Y; Dai, L-J; et al. (2006) The early embryo loss caused by 2,3,7,8-tetrachlorodibenzo-p-dioxin may be
20 related to the accumulation of this compound in the uterus. Reprod Toxicol 21:301-306.
21 Li, CY; Sung, FC. (1999) A review of the healthy worker effect in occupational epidemiology. Occup Med
22 49(4):225-229.
23 Li, X; Johnson, DC; Rozman, KK. (1997) 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) increases release of
24 luteinizing hormone and follicle-stimulating hormone from the pituitary of immature female rats in vivo and in vitro.
25 Toxicol Appl Pharmacol 142(2): 264--269.
26 Limbird, LE. (1996) Cell surface receptors: a short course on theory and method. USA: Springer; 258 pp.
27 Longnecker, MP; Gladen, BC; Patterson, DG, Jr; et al. (2000) Polychlorinated biphenyl (PCB) exposure in relation
28 to thyroid hormone levels in neonates. Epidemiology 11:249-254.
29 Lorber, M; Patterson, D; Huwe, J; et al. (2009) Evaluation of background exposures of Americans to dioxin-like
30 compounds in the 1990s and the 2000s. Chemosphere 77:640-651.
31 Lorenzen, A; Okey, AB. (1991) Detection and characterization of Ah receptor in tissue and cells from human
32 tonsils. Toxicol Appl Pharmacol 107(2):203-214.
33 Lucier, GW; Rumbaugh, RC; McCoy, Z; et al. (1986) Ingestion of soil contaminated with 2,3,7,8-tetrachloro-
34 dibenzo-p-dioxin (TCDD) alters hepatic enzyme activities in rats. Fund Appl Toxicol 6:364-371.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-15 DRAFT—DO NOT CITE OR QUOTE
-------
1 Lucier, GW; Tritscher, A; Goldsworthy, T; et al. (1991) Ovarian hormones enhance 2,3,7,8-tetrachlorodibenzo-p-
2 dioxin-mediated increases in cell proliferation and preneoplastic foci in a two-stage model for rat
3 hepatocarcinogenesis. Cancer Res 51(5): 1391—1397.
4 Lutz, WK. (1990) Dose-response relationship and low dose extrapolation in chemical carcinogenesis.
5 Carcinogenesis 11:1243-1247.
6 Lutz, WK. (1999) Dose-response relationships in chemical carcinogenesis reflect diffeences in individual
7 susceptibility. Consequences for canceer risk assessment, extrapolation, and prevention. Human Exp Toxicol
8 18:707-712.
9 Lutz, WK. (2001) Susceptibility differences in chemical carcinogenesis linearize the dose-response relationship:
10 threshold doses can be defined only for individuals. DNA Repair 482:71-76.
11 Lutz, WK; Gaylor, DW. (2008) Letter to the editor. Dose-response relationships for cancer incidence reflect
12 susceptibility distributions. Chem. Res. Toxicol. 21:971-973.
13 Mackie, D; Liu, J; Loh, YS; et al. (2002) No evidence of dioxin cancer threshold. Environ Health Perspect
14 111(9): 1145—1147. Available at: http://www.ehponline.org/members/2003/5730/5730.pdf.
15 Mally, A; Chipman, JK. (2002) Non-genotoxic carcinogens: early effects on gap junctions, cell proliferation and
16 apoptosis in the rat. Toxicology 180:233-248.
17 Manchester, DK; Gordon, SK; Golas, C; et al. (1987) Ah receptor in human placenta: stabilization by molybdate and
18 characterization of binding of 2,3,7,8-tetrachlorodibenzo-p-dioxin, 3-methylcholanthrene, and benzo(a)pyrene.
19 Cancer Res 47(18):4861^1868.
20 Manz, A; Berger, J; Dwyer, JH; et al. (1991) Cancer mortality among workers in chemical plant contaminated with
21 dioxin. Lancet 338(8773):959-964.
22 Markowski, VP; Zareba, G; Stern, S; et al. (2001) Altered operant responding for motor reinforcement and the
23 determination of benchmark doses following perinatal exposure to low-level 2,3,7,8-tetrachlorodibenzo-p-dioxin.
24 Environ Health Perspect 109(6):621-627.
25 Maronpot, RR; Montgomery, CA; Boorman, GA; et al. (1986) National Toxicology Program nomenclature for
26 hepatoproliferative lesions of rats. Toxicol Pathol 14(2):263-273.
27 Maronpot, RR; Pitot, HC; Peraino, C. (1989) Use of rat liver altered focus models for testing chemicals that have
28 completed two-year carcinogenicity studies. Toxicol Pathol 17(4):651-662.
29 Maronpot, RR; Foley, JF; Takahashi, K; et al. (1993) Dose-response for TCDD promotion of hepatocarcinogenesis
30 in rats initiated with DEN: histologic, biochemical, and cell proliferation endpoints. Environ Health Perspect
31 101:634-642.
32 Maruyama, W; Yshida, K; Tanaka, T; et al. (2002) Determination of tissue-blood partition coefficients for a
33 physiological model for humans, and estimation of dioxin concentration in tissues. Chemosphere 46:975-985.
34 Matsumoto, Y; Ide, F; Kishi, R; et al. (2007) Aryl hydrocarbon receptor plays a significant role in mediating
3 5 airborne particulate-induced carcinogenesis in mice. Environ Sci Technol 41(10):3775-3780.
36 McBride, DI; Collins, JJ; Humphry, NF; et al. (2009a) Mortality in workers exposed to 2,3,7,8-tetrachlorodibenzo-
37 p-dioxin at a trichlorophenol plant in New Zealand. J Occup Environ Med 51(9):1049-1056.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-16 DRAFT—DO NOT CITE OR QUOTE
-------
1 McBride, DI; Burns, CJ; Herbison, GP; et al. (2009b) Mortality in employees at a New Zealand agrochemical
2 manufacturing site. Occup Med 59(4):255-263.
3 McEwen, LN; Kim, C; Haan, M; et al. (2006) Diabetes reporting as a cause of death. Diabetes Care 29(2):247-253.
4 McMichael, AJ. (1976) Standardized mortality ratios and the "healthy worker effect": scratching beneath the
5 surface. J Occup Med 18:128-131.
6 McMillan, BJ; Bradfield, CA. (2007) The aryl hydrocarbon receptor sans xenobiotics: endogenous function in
7 genetic model systems. Mol Pharmacol 72:487-498.
8 McNulty, WP; Nielsen-Smith, KA; Lay, JO, Jr; et al. (1982) Persistence of TCDD in monkey adipose tissue. Food
9 Chem Toxicol. 20(6):985-986.
10 Michalek, JE; Pavuk, M. (2008) Diabetes and cancer in veterans of Operation Ranch Hand after adjustment for
11 calendar period, days of spraying, and time spent in Southeast Asia. J Occup Environ Med 50(3):330-340.
12 Michalek, JE; Pirkle, JL; Caudill, SP; et al. (1996) Pharmacokinetics of TCDD in veterans of Operation Ranch
13 Hand: 10-year follow-up. J Tox Environ Health 47(3):209-220.
14 Michalek, JE; Pirkle, JL; Needham, LL; et al. (2002) Pharmacokinetics of 2,3,7,8-tetrachlorodibenzo-p-dioxin in
15 Seveso adults and veterans of operation Ranch Hand. J Expo Anal Environ Epidemiol 12:44-53.
16 Micka, J; Milatovich, A; Menon, A; et al. (1997) Human Ah receptor (AHR) gene: localization to 7pl5 and
17 suggestive correlation of polymorphism with CYP1A1 inducibility. Pharmacogenetics 7:95-101.
18 Miettinen, HM; Sorvari, R; Alaluusua, S; et al. (2006) The Effect of perinatal TCDD exposure on caries
19 susceptibility in rats. Toxicol Sci 91(2):568-575.
20 Milbrath, MO; Wenger, Y; Chang, CW; et al. (2009) Apparent half-lives of dioxins, furans, and poly chlorinated
21 biphenyls as a function of age, body fat, smoking status, and breast-feeding. Environ Health Perspect
22 117(3):417-425.
23 Mocarelli P; Brambilla P; Gerthoux PM; et al. (1996) Change in sex ratio with exposure to dioxin [letter]. Lancet
24 348:409.
25 Mocarelli, P; Gerthoux, PM; Ferrari, E; et al. (2000) Paternal concentrations of dioxin and sex ratio of offspring.
26 Lancet 355:1858-1863.
27 Mocarelli, P; Gerthoux, PM; Patterson, DG, Jr; et al. (2008) Dioxin exposure, from infancy through puberty,
28 produces endocrine disruption and affects human semen quality. Environ Health Perspect 116(l):70-77.
29 Monson, RR. (1986) Observations on the healthy worker effect. J Occup Med 28(6):425-433.
30 Moser, GA; McLaghlan, MS. (2001) The influence of dietary concentration on the absorption and excretion of
31 persistent lipophilic organic pollutants in the human intestinal tract. Chemosphere 45:201-211.
32 Motulsky, HJ; Christopoulos, A. (2003) Fitting models to biological data using linear and nonlinear regression. A
33 practical guide to curve fitting. San Diego: GraphPad Software, Inc., www.graphpad.com
34 Muller, A; De La Rochebrochard, E; Labbe-Decleves, C; et al. (2004) Selection bias in semen studies due to self-
35 selection of volunteers. Human Reprod 19:2838-2844.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-17 DRAFT—DO NOT CITE OR QUOTE
-------
1 Murdoch, DJ; Krewski, D. (1988) Carcinogenic risk assessment with time-dependent exposure patterns. Risk Anal
2 Dec;8(4):521-530.
3 Murdoch, DJ; Krewski, D; Wargo, J. (1992) Cancer risk assessment with intermittent exposure. Risk Anal
4 Dec;12(4):569-577.
5 Murphy, JM; Sexton, DMH; Barnett, DN; et al. (2004) Quantification of modeling uncertainties in a large ensemble
6 of climate change simulations. Nature 430:768-772. Available at:
7 http://www.seas.harvard.edu/climate/pdf/Uncertainty 1 .pdf.
8 Murray, FJ; Smith, FA; Nitschke, KD; et al.(1979) Three-generation reproduction study of rats given 2,3,7,8-
9 tetrachlorodibenzo-p-dioxin (TCDD) in the diet. Toxicol Appl Pharmacol 50:241-252.
10 Muto, T; Wakui, S; Imano, N; et al. (2001) In-utero and lactational exposure of 3,3',4,4',5-pentachlorobiphenyl
11 modulate dimethlbenz[a]anthracene-induced rat mammary carcinogenesis. J Toxicol Pathol 14(3):213-224.
12 Available at: http://www.jstage.jst.go.jp/article/tox/14/3/213/_pdf.
13 Myers, HE; Thompson, ML. (1998) Meta-analysis and occupational epidemiology. Occup Medic 48:99-101.
14 Nagel, S; Berger, J; Flesch-Janys, D; et al. (1994) Mortality and cancer mortality in a cohort of female workers of a
15 herbicide producing plant exposed to polychlorinated dibenzo-p-dioxins and furans. Inform Biomet Epidemiol Med
16 Biol 25:32-38.
17 NAS (National Academy of Sciences). (2006a) Health risks from dioxin and related compounds: evaluation of the
18 EPA reassessment. Washington, DC: National Academies Press. Available at
19 http://www.nap.edu/catalog.php?record_id=l 1688.
20 NAS (National Academy of Sciences). (2006b) Health risks from dioxin and related compounds: listen to the
21 briefing (audio recording of public briefing led by D. Eaton, chair, NAS Dioxin Committee), Washington, DC.
22 Available at: http://www.nap.edu/webcast/webcast_detail.php?webcast_id=328 (July 11).
23 NAS (National Academy of Sciences). (2009) Toward a unified approach to dose-response assessment: the need for
24 an improved dose-response framework. Science and Decisions: Advancing Risk Assessment (pp. 127-187).
25 National Resource Council. Washington: National Academies Press.
26 NASA (National Aeronautics and Space Administration). (2002) Probabilistic risk assessment procedures guide for
27 NASA managers and practitioners. Version 1.1. Office of Safety and Mission Assurance, NASA Headquarters
28 Washington, DC (Aug.). Available at: http://www.hq.nasa.gov/office/codeq/doctree/praguide.pdf.
29 Nebert, DW; Petersen, DD; Fornace, AJ, Jr. (1990) Cellular responses to oxidative stress: the [Ah] gene battery as a
3 0 paradigm. Environ Health Perspect 88:13-25.
31 Nebert, DW; Peterson, DD; Puga, A. (1991) Human Ah locus polymorphism and cancer: Inducibility of CYPIA1
32 and other genes by combustion products and dioxin. Pharmacogenetics 1:68-78.
33 Needham, LL; Gerthoux, PM; Patterson, DG; et al. (1994) Half-life of 2,3,7,8-tetrachlorodibenzo-p-dioxin in serum
34 of Seveso adults: Interim report. Organohal Comp 21:81—85.
35 Needham, LL; Gerthoux, PM; Patterson, DG, Jr; et al. (1997) Serum dioxin levels in Seveso, Italy, population in
36 1976. Teratog Carcinog Mutagen 17:225-240.
37 Needham, LL; Barr, DB; Caudill, SP; et al. (2005) Concentrations of environmental chemicals associated with
3 8 neurodevelopmental effects in the US population. NeuroToxicology 26(4):531-545.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-18 DRAFT—DO NOT CITE OR QUOTE
-------
1 Nessel, CS; Amoroso, MA; Umbreit, TH; et al. (1992) Transpulmonary uptake and bioavailability of 2,3,7,8-TCDD
2 from respirable soil particles. Chemosphere 25(l-2):29-32.
3 Nilsson, CB; Hakansson, H. (2002) The retinoid signaling system - a target in dioxin toxicity. Crit Rev Toxicol
4 32(3):211-232.
5 Nishimura, N; Yonemoto, J; Miyabara, Y; et al. (2005) Altered thyroxin and retinoid metabolic response to 2,3,7,8-
6 tetrachlorodibenzo-p-dioxin in aryl hydrocarbon receptor-null mice. Arch Toxicol 79:260-267.
7 Niskar, A; Needham, LL; Rubin, C; et al. (2009) Serum dioxin, polychlorinated biphenyls, and endometriosis: a
8 case-control study in Atlanta. Chemosphere 74(7):944-949.
9 Nohara, K; Fujimaki, N; Tsukumo, S; et al. (2000) The effects of perinatal exposure to low doses of 2,3,7,8-
10 tetrachlorodibenzo-p-dioxin on immune organs in rats. Toxicology 154:123-133.
11 Nohara, K; Izumi, H; Tamura, S; et al. (2002) Effect of low-dose 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) on
12 influenza A virus-induced mortality in mice. Toxicology 170:131-138.
13 Nolan, KJ; Smith, FA; Hefner, JG. (1979) Elimination and tissue distribution of 2,3,7,8-tetrachlorodibenzo-p-dioxin
14 (TCDD) in female guinea pigs following a single oral dose. Toxicol Appl Pharmacol 48:162.
15 NRC (National Research Council). (1983) Risk assessment in the federal government: managing the process.
16 Washington, DC: National Academies Press. Available at: http://www.nap.edu/openbook.php?isbn=0309033497.
17 NRC (National Research Council). (1989) Improving risk communication. Washington, DC: National Academies
18 Press. Available at: http://www.nap.edu/openbook.php?isbn=0309039436.
19 NRC (National Research Council). (1991) Human exposure assessment for airborne pollutants: advances and
20 opportunities. Washington, DC: National Academies Press. Available at:
21 http://books.nap.edu/openbook.php?record_id=1544.
22 NRC (National Research Council). (1993) Issues in risk assessment. Washington, DC: National Academies Press.
23 Available at: http://www.nap.edu/catalog.php?record_id=2078.
24 NRC (National Research Council). (1994) Science and judgment in risk assessment. Washington, DC: National
25 Academies Press. Available at: http://www.nap.edu/openbook.php?record_id=2125&page=629.
26 NRC (National Research Council). (2002) Estimating the public health benefits of proposed air pollution
27 regulations. Washington, DC: National Academies Press. Available at:
28 http://www.nap.edu/catalog.php?record_id=10511. )
29 NRC (National Research Council). (2006) Health risks from dioxin and related compounds: evaluation of the EPA
30 reassessment. Washington, DC: National Academies Press. Available at:
31 http://www.nap.edu/catalog.php?record_id=l 1688.
32 NRC (National Research Council). (2007) Scientific review of the proposed risk assessment bulletin from the Office
33 of Management and Budget. Washington, DC: National Academies Press. Available at:
34 http://www.nap.edu/catalog.php?record_id=l 1811.
35 NRC (National Research Council). (2009) Science and decisions: advancing risk assessment. Committee on
36 Improving Risk Analysis Approaches Used by the U.S. EPA. Washington, DC: National Academies Press.
37 Available at: http://www.nap.edu/catalog.php?record_id=12209.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-19 DRAFT—DO NOT CITE OR QUOTE
-------
1 NTP (National Toxicology Program). (1982) NTP Technical Report on carcinogenesis bioassay of 2,3,7,8-
2 tetrachlorodibenzo-p-dioxin in Osborne-Mendel rats and B6C3F1 mice (gavage study). Public Health Service, U.S.
3 Department of Health and Human Services; NTP TR 209. Available from the National Institute of Environmental
4 Health Sciences, Research Triangle Park, NC.
5 NTP (National Toxicology Program). (2006) Toxicology and carcinogenesis studies of a mixture of
6 2.3,7.8-tetrachlorodiben/o-/)-dioxin (TCDD) (CAS No. 1746-01-6), 2,3,4,7,8-pentachlorodibenzofuran (PeCDF)
7 (CAS No. 57117-31-4), and 3,3',4,4',5-pentachlorobiphenyl (PCB 126) (CAS No. 57465-28-8) in female Harlan
8 Sprague-Dawley Rats (gavage studies). Public Health Service, U.S. Department of Health and Human Services;
9 NTP TR 526. Available from the National Institute of Environmental Health Sciences, Research Triangle Park, NC.
10 Available at: http://ntp.mehs.nih.gov/index.cfm?objectid=070B7300-0E62-BF12-F4C3E3B5B645A92B.
11 Oehlert, GW. (1992) A note on the delta method. Am Stat 46(l):27-29.
12 Ohsako, S; Miyabara, Y; Nishimura, N; et al. (2001) Maternal exposure to a low dose of 2,3,7,8-
13 tetrachlorodibenzo-p-dioxin (TCDD) suppressed the development of reproductive organs of male rats: dose-
14 dependent increase of mRNA levels of 5a-reductase type 2 in contrast to decrease of androgren receptor in the
15 pubertal ventral prostate. Toxicol Sci 60:132-143.
16 Okey, AB; Riddick, DS; Harper; PA. (1994) The Ah receptor: mediator of the toxicity of 2,3,7,8-tetrachlorodibenzo-
17 /j-dioxin (TCDD) and related compounds. Toxicol Lett 70:1-22.
18 Olson, JR; Holscher, MA; Neal, RA. (1980) Toxicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin in the Golden Syrian
19 Hamster. Toxicol Appl Pharmacol 55:67-78.
20 Olson, JR; McGarrigle, BP; Gigliotti, PJ; et al. (1994) Hepatic uptake and metabolism of 2,3,7,8-tetrachlorodibenzo-
21 /j-dioxin and 2,3,7,8-tetrachlorodibenzofuran. Fundam Appl Toxicol 22:631-640.
22 OMB (Office of Management and Budget). (2006) Proposed risk assessment bulletin (Draft for public review).
23 Washington, DC (Jan.). Available at:
24 http://www.whitehouse.gOv/omb/inforeg/proposed_risk_assessment_bulletin_010906.pdf.
25 Ott, MG; Zober A. (1996) Cause specific mortality and cancer incidence among employees exposed to 2,3,7,8-
26 TCDD after a 1953 reactor accident. Occup Environ Med 53:606-612.
27 Ott, MG; Olson, RA; Cook, RR; et al. (1987) Cohort mortality study of chemical workers with potential exposure to
28 the higher chlorinated dioxins. J Occup Med 29(5):422-429.
29 Ott, MG; Messerer, P; Zober, A. (1993) Assessment of past occupational exposure to 2,3,7,8-tetrachlorodibenzo-p-
3 0 dioxin using blood lipid analyses. Int Arch Occup Environ Health 65:1-8.
31 Papke, O; Ball, M; Lis, A. (1994) PCDD/PCDF in humans - an update of background data. Chemosphere
32 29:2355-2360.
33 Pekelis, M; Nicolich, MJ; Gauthier, JS. (2003) Probabilistic framework for the estimation of the adult and child
34 toxicokinetic intraspecies uncertainty factors. Risk Anal 23(6): 1239-1255. Available at:
35 http://www3.interscience.wiley.com/cgi-bin/fulltext/118881007/PDFSTART.
36 Percy, C; Stanek, E, III; Gloeckler, L. (1981) Accuracy of cancer death certificates and its effect on cancer mortality
37 statistics. Am J Public Health 71(3):242-250.
3 8 Pereg, D; Dewailly, E; Poirier, GG; et al. (2002) Environmental exposure to polychlorinated biphenyls and placental
39 CYP1A1 activity in Inuit women from northern Quebec. Environ Health Perspect 110(6):607-613 (June).
40 Available at: http://www.pubmedcentral.nih.gov/picrender.fcgi?artid=1240878&blobtype=pdf.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-20 DRAFT—DO NOT CITE OR QUOTE
-------
1 Pesatori, AC; Zocchetti, C; Guercilena, S; et al. (1998) Dioxin exposure and non-malignant health effects: A
2 mortality study. Occup Environ Med 55:126-131.
3 Pesatori, AC; Consonni, D; Bachetti, S; et al. (2003) Short- and long-term morbidity and mortality in the population
4 exposed to dioxin after the "Seveso Accident." Industrial Health41:127-138.
5 Piacitelli, L; Sweeney, MH; Fingerhut, M. (1992) Serum levels of 2,3,7,8-substituted PCDDs among workers
6 exposed to 2,3,7,8-TCDD contaminated chemicals. Chemosphere 25:251-254.
7 Pipe, NG; Smith, T; Halliday, D; et al. (1979) Changes in fat, fat-free mass and body water in human normal
8 pregnancy. Br J Obstet Gynaecol 86(12):929-940.
9 Pirkle, JL; Wolfe, WH; Patterson, DG; et al. (1989) Estimates of the half-life of 2,3,7,8-tetrachlorodibenzo-p-dioxin
10 in Vietnam Veterans of Operation Ranch Hand. J Tox Environ Health 27(2): 165-171.
11 Pitot, HC; Goldsworthy, T; Campbell, HA; et al. (1980) Quantitative evaluation of the promotion by 2,3,7,8-
12 tetrachlorodibenzo-p-dioxin of hepatocarcinogenesis from diethylnitrosamine. Cancer Res 40:3616-3620.
13 Pohjanvirta, R; Tuomisto, J. (1994) Short-term toxicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin in laboratory
14 animals: Effects, mechanisms, and animal models. Pharmacol Rev 46:483-549.
15 Pohjanvirta, R; Tuomisto, L; Tuomisto, J. (1989) The central nervous system may be involved in TCDD toxicity.
16 Toxicology 58:167-174.
17 Poiger, M; Schlatter, C. (1986) Pharmacokinetics of 2,3,7,8-TCDD in man. Chemosphere 15:9-12.
18 Poland, A;Glover, E. (1980) 2,3,7,8-Tetrachlorodibenzo-p-dioxin: Segregation of toxicity with the Ah locus. Mol.
19 Pharmacol. 17:86-94.
20 Poland, A; Glover, E. (1990) Characterization and strain distribution pattern of the murine Ah receptor specified by
21 the Ahd and Ahb-3 alleles. Mol Pharmacol 38(3):306-312.
22 Poland, A; Palen, D; Glover, E. (1982) Tumor promotion by TCDD in skin of HRS/J mice. Nature
23 300(5889):271-273.
24 Poland, A; Palen, D; Glover, E. (1994) Analysis of the four alleles of the murine aryl hydrocarbon receptor. Mol
25 Pharmacol 46:915-921.
26 Popp, JA; Crouch, E; McConnell, EE. (2006) A weight-of-evidence analysis of the cancer dose-response
27 characteristics of 2,3,7,8-tetrachlorodibenzodioxin (TCDD). Toxicol Sci 89(2): 361-369.
28 Potter, CL; Sipes, IG; Russell, DH. (1983) Hypothyroxinemia and hypothermia in rats in response to 2,3,7,8-
29 tetrachlorodibenzo-p-dioxin administration. Toxicol Appl Pharmacol 69(l):89-95.
30 Potter, CL; Moore, RW; Inhorn, SL; et al. (1986) Thyroid status and thermogenesis in rats treated with 2,3,7,8-
31 tetrachlorodibenzo-p-dioxin. Toxicol Appl Pharmacol 84(l):45-55.
32 PoulinP; Theil FP. (2001) Prediction of pharmicokinetics prior to in vivo studies. 1. mechanism-based prediction of
33 volume of distribution. J Pharm Sci 91(1): 129—156.
34 Puga, A; Nebert, DW; Carier, F. (1992) Dioxin induces expression of c-fos and c-jun proto-oncogenes and a large
3 5 increases in transcription factor AP-1. DNA Cell Biol 11:269-281.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-21 DRAFT—DO NOT CITE OR QUOTE
-------
1 Ramadoss, P; Perdew, GH. (2004) Use of 2-Azido-3-[125I]iodo-7,8-dibromodibenzo-/?-dioxin as a probe to
2 determine the relative ligand affinity of human versus mouse aryl hydrocarbon receptor in cultured cells. Molec
3 Pharmacol 66:129-136.
4 Ramsey, JC; Hefner, JG; Karbowski, RJ; et al. (1982) The in vivo biotransformation of 2,3,7,8-tetrachlorodibenzo-
5 p-dioxin (TCDD) in the rat. Toxicol Appl Pharmacol 65(1): 180-184.
6 Rao, MS; Subbarao, V; Prasad, JD; et al. (1988) Carcinogenicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin in the
7 Syrian golden hamster. Carcinogenesis 9(9): 1677-1679.
8 Reddy, M; Yang, R; Clewell, HJ; et al. (2005) Physiologically based pharmacokinetic modeling: science and
9 applications. Hoboken, New Jersey: John Wiley & Sons.
10 Revich, B; Aksel, E; Ushakova, T; et al. (2001) Dioxin exposure and public health in Chapaevsk, Russia.
11 Chemosphere 43(4-7):951-966.
12 Revich, B; Sergeyev, O; Zeilert, V; et al. (2005) Chapaevsk, Russia: 40 years of dioxins exposure on the human
13 health and 10 years of Russian - USA epidemiological studies.
14 Rier, SE; Martin, DC; Bowman, RE; et al. (1993) Endometriosis in rhesus monkeys (Macaca mulata) following
15 chronic exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Fundam Appl Toxicol 21:433-441.
16 Rier, SE; Martin, DC; Bowman, RE; et al. (1995) Immunoresponsiveness in endometriosis: implications of
17 estrogenic toxicants. Environ Health Perspect 103 (Suppl 7): 151-156.
18 Rier, SE; Turner, WE; Martin, DC; et al. (2001a) Serum levels of TCDD and dioxin-like chemicals in rhesus
19 monkeys chronically exposed to dioxin: correlation of increased serum PCB levels with endometriosis. Toxicol Sci
20 59:147-159.
21 Rier, SE; Coe, CL; Lemieux, AM; et al. (200 lb) Increased tumor necrosis factor-a production by peripheral blood
22 leukocytes from TCDD-exposed rhesus monkeys. Toxicol Sci 60:327-337.
23 Roberts, EA; Shear, NH; Okey, AB; et al. (1985) The Ah receptor and dioxin toxicity: from rodent to human tissues.
24 Chemosphere 14:661-674.
25 Roberts, EA; Golas, CL; Okey, AB. (1986) Ah receptor mediating induction of aryl hydrocarbon hydroxylase:
26 detection in human lung by binding of 2,3,7,8-[3H]tetrachlorodibenzo-p-dioxin. Cancer Res 46:3739-3743.
27 Rohde, S; Moser, GA; Papke,0; et al. (1999) Clearance of PCDD/Fs via the gastrointestinal tract in occupationally
28 exposed persons. Chemosphere 38(14):3397-3410.
29 Roth, WL; Ernst, S; Weber, LW; et al. (1994) A pharmacodynamically responsive model of 2,3,7,8-
3 0 tetrachlorodibenzo-p-dioxin (TCDD) transfer between liver and fat at low and high doses. Toxicol Appl Pharmacol
31 127:151-162.
32 Rothman, KJ. (1986) Modern epidemiology. Toronto: Little, Brown, and Company.
33 Roy, T; Hammerstrom, K; Schaum, J. (2008) Percutaneous absorption of 2,3,7,8-tetrachlorodibenzo-p-dioxin
34 (TCDD) from soil. J Toxicol Environ Health A 71(23): 1509—1515.
3 5 Rozman, KK. (2000) The role of time in toxicology or Haber's c x t product. Toxicol 149:35-42.
36 Ryan, JJ; Schecter, A. (2000) Exposure of Russian phenoxy herbicide producers to dioxin. J Occup Environ Med
37 42:861-870.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-22 DRAFT—DO NOT CITE OR QUOTE
-------
1 Ryan, JJ; Amirova, Z; Carrier, G. (2002) Sex ratios of children of Russian pesticide producers exposed to dioxin.
2 Environ Health Perspect 110(11):A699-701.
3 Saltelli, A; Chan, K; Scott, EM; eds. (2000) Sensitivity analysis. Wiley Series in Probability and Statistics, John
4 Wiley & Sons Ltd., England (Oct.) (reprinted Nov. 2001, John Wiley & Sons, Inc., New York, NY).
5 Sandau, CD; Ayotte, P; Dewailly, E; et al. (2002) Pentachlorophenol and hydroxylated polychlorinated biphenyl
6 metabolites in umbilical cord plasma of neonates from coastal populations in Quebec. Environ Health Perspect
7 110(4):411-417.
8 Santostefano, MJ; Johnson, KL; Whisnant, NA; et al. (1996) Subcellular localization of TCDD differs between the
9 liver, lungs, and kidneys after acute and subchronic exposure: species/dose comparison and possible mechanism.
10 Fundam Appl Toxicol 34(2):265-275.
11 Santostefano, MJ; Wang, X; Richardson, VM; et al. (1998) A pharmacodynamic analysis of TCDD-induced
12 cytochrome P450 gene expression in multiple tissues: dose- and time-dependent effects. Toxicol Appl Pharmacol
13 151:294-310.
14 Saracci, R; Kogevinas, M; Bertazzi, PA; et al. (1991) Cancer mortality in workers exposed to chlorophenoxy
15 herbicides and chlorophenols. Lancet 338(8774): 1027-1032.
16 Sauer, RM. (1990) 2,3,7,8-tetrachlorodibenzo-p-dioxin in Sprague-Dawley rats. Submitted to the Maine Scientific
17 Advisory Panel by Pathco, Inc., Ijamsville, MD. March 13, 1990.
18 Schantz, SL; Bowman, RE. (1989) Learning in monkeys exposed perinatally to 2,3,7,8-tetrachloridibenzo-p-dioxin
19 (TCDD). Neurotoxicol Teratol 11:13-19.
20 Schantz, SL; Laughlin, NK; Van Valkenberg, HC; et al. (1986) Maternal care by rhesus monkeys exposed to either
21 leador2,3,7,8-tetrachlorodibenzo-p-dioxin. Neurotoxicology 7(2):637-650.
22 Schantz, SL; Seo, BW; Moshtaghian, J; et al. (1996) Effects of gestational and lactational exposure to TCDD or
23 coplanar PCBs on spatial learning. Neurotoxicol Teratol 18(3):305—313.
24 Schecter, A; Cramer P; Boggess K; et al. (1997) Levels of dioxins, dibenzofurans, PCB and DDE congeners in
25 pooled food samples collected in 1995 at supermarkets across the United States. Chemosphere 34(5-7): 1437-1447.
26 Schwartz, M; Appel, KE. (2005) Carcinogenic risks of dioxin: mechanistic considerations. Regul Toxicol
27 Pharmacol 43(1): 19-34.
28 Seidel, SD; Winters, GM; Rogers, WJ; et al. (2001) Activation of the Ah receptor signaling pathway by
29 prostaglandins. J BiochemMol Toxicol 15:187-196.
30 Self, SG; Liang, KY. (1987) Asymptotic properties of maximum likelihood estimators and likelihood ratio tests
31 under nonstandard conditions. J Am Stat Assoc 82: 605-610.
32 Seo, B-W; Li, M-H; Hansen, LG; et al. (1995) Effects of gestational and lactational exposure to coplanar
33 polychlorinated biphenyl (PCB) congeners or 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) on thyroid hormone
34 concentrations in weanling rats. Toxicology Letters 78(3):253--262.
3 5 Sewall, CH; Lucier, GW; Tritscher, AM; et al. (1993) TCDD-mediated changes in hepatic epidermal growth factor
36 receptor may be a critical event in the hepatocarcinogenic action of TCDD. Carcinogenesis 14:1885-1893.
37 Sewall, CH; Flagler, N; Vanden Heuvel, JP; et al. (1995) Alterations in thyroid function in female Sprague-Dawley
3 8 rats following chronic treatment with 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol Appl Pharmacol 132:237-244.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-23 DRAFT—DO NOT CITE OR QUOTE
-------
1 Shi, Z; Valdez, KE; Ying, AY; et al. (2007) Ovarian endocrine disruption underlies premature reproduction
2 senescence following environmentally relevant chronic exposure to aryl hydrocarbon receptor agonist 2,3,7,8-
3 tetrachlorodibenzo-p-dioxin. Biol Reprod 30(4):293-342.
4 Shu, H; Teitelbaum, P; Ebb, AS; et al. (1988) Bioavailability of soil-bound TCDD: dermal bioavailability in the rat.
5 Fundam Appl Toxicol (2):335-343.
6 Siemiatycki, J; Wacholder, S; Dewar, R; et al. (1988) Degree of confounding bias related to smoking, ethnic group,
7 and socioeconomic status in estimates of the associations between occupation and cancer. J Occup Med
8 30(8):617-625.
9 Sikov, M. (1970) Radiation biology of the fetal and juvenile mammal. Science 167(3925):1640-1641.
10 Simanainen, U; Tuomisto, JT; Tuomisto, J; Viluksela, M. (2002) Structure-activity relationships and dose responses
11 of polychlorinated dibenzo-p-dioxins for short-term effects in 2,3,7,8-tetrachlorodibenzo-p-dioxin-resistant and -
12 sensitve rat strains. Toxicol Appl Pharmacol 181:38-47.
13 Simanainen, U; Haavisto, T; Tuomisto, J.T; et al. (2004) Pattern of male reproductive system effects after in utero
14 and lactational 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) exposure in three differentially TCDD-sensitive rat
15 lines. Toxicol Sci 80(1): 101-108.
16 Simon, T; Aylward, LL; Kirman, CR; et al. (2009) Estimates of cancer potency of 2,3,7,8-
17 tetrachlorodibenzo(p)dioxin using linear and non-linear dose-response modeling and toxicokinetics. Toxicol Sci
18 Advance access published online on September 23, 2009. doi: 10.1093/toxsci/kfp232.
19 Slezak, BP; Hatch, GE; DeVito, MJ; et al. (2000) Oxidative stress in female B6C3F1 mice following acute and
20 subchronic exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD). Toxicol Sci 54:390-398.
21 Smart, J; Daly, A. (2000) Variation in induced CYP1A1 levels: relationship to CYP1A1, Ah receptor, and GSTM1
22 polymorphisms. Pharmacogenetics 10(1): 11-24.
23 Smialowicz, RJ; Burgin, DE; Williams, WC; et al. (2004) CYP1A2 is not required for 2,3,7,8-tetrachlorodibenzo-p-
24 dioxin-induced immunosuppression. Toxicology 197(1): 15-22.
25 Smialowicz, RJ; DeVito, MJ; Williams, WC; et al. (2008) Relative potency based on hepatic enzyme induction
26 predicts immunosuppressive effects of a mixture of PCDDS/PCDFS and PCBS. Toxicol Appl Pharmacol 227:477-
27 484.
28 Smith, AH; Lopipero, P. (2001) Invited commentary: How do the Seveso findings affect conclusions concerning
29 TCDD as a human carcinogen? Am J Epidemiol 153(11): 1045-1047.
30 Smith, AH; Fisher, DO; Pearce, N; et al. (1982) Congenital defects and miscarriages among New Zealand 2, 4, 5-T
31 sprayers. Arch Environ Health 37(4): 197-200.
32 Spiegelhalter, D; Thomas, A; Best, N; et al. (2003) BUGS 0.5 Bayesian inference using Gibbs sampling manual,
33 version ii. MRC Biostatistics Units, Institute of Public Health, Cambridge. Available at:
34 http://193.60.86.19/bugs/documentation/Download/manual05.pdf
35 Squire, RA. (1980) Pathologic evaluations of selected tissues from the Dow Chemical TCDD and 2,4,5-T rat
36 studies. Submitted to Carcinogen Assessment Group, U.S. Environmental Protection Agency on August 15 under
37 Contract No. 68-01-5092.
3 8 Starr, TB. (2003) Significant issues raised by meta-analyses of cancer mortality and dioxin exposure. Environ
39 Health Perspect 111(12): 1443-1147.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-24 DRAFT—DO NOT CITE OR QUOTE
-------
1 Staskal, DF; Diliberto, JJ; Devito, MJ; et al. (2005) Inhibition of human and rat CYP1A2 by TCDD and dioxin-like
2 chemicals. Toxicol Sci 84:225-231.
3 Stayner, L; Bailer, AJ; Smith, R; et al. (1999) Sources of uncertainty in dose-response modeling of epidemiological
4 data for cancer risk assessment. Ann NY Acad Sci 895:212-222.
5 Stayner, L; Steenland, K; Dosemeci, M; et al. (2003) Attenuation of exposure-response curves in occupational
6 cohort studies at high exposure levels. Scand J Work Environ Health 29(4):317-324.
7 Steenland, K; Deddens, J. (2003) Dioxin: exposure-response analyses and risk assessment. Ind Health 41:175-180.
8 Steenland, K; Piacitelli, L; Deddens, J; et al. (1999) Cancer, heart disease, and diabetes in workers exposed to
9 2,3,7,8-tetrachlorodibenzo-p-dioxin. J Natl Cancer Inst 91(9):779-786.
10 Steenland, K; Deddens, J; Piacitelli, L. (2001) Risk assessment for 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)
11 based on an epidemiologic study. Am J Epidemiol 154(5):451-458.
12 Stellman, SD; Stellman, JM. (1986) Estimation of exposure to Agent Orange and other defoliants among American
13 troops in Vietnam: a methodological approach. Am J Ind Med 9:305-321.
14 Stephenson, RP. (1956) A modification of receptor theory. BrJPharmacol 11:379.
15 Sugita-Konishi, Y; Kobayashi, K; Naito, H; et al. (2003) Effect of lactational exposure to 2,3,7,8-
16 tetrachlorodibenzo-p-dioxin on the susceptibility to Listeria infection. Biosci Biotechnol Biochem 67(l):89-93.
17 Swartout, JC; Price, PS; Dourson. ML; et al. (1998) A probabilistic framework for the reference dose (probabilistic
18 RfD). Risk Anal 18(3):271-282.
19 Takemoto, K; Nakajima, M; Fujiki, Y; et al. (2004) Role of the aryl hydrocarbon receptor and Cyplbl in the anti-
20 estrogenic activity of 2,3,7,8-tetrachlorodibenzo-p-dioxin. Arch Toxicol 78:309-315.
21 Thiess, AM; Frentzel-Beyme, R. (1977) Mortality study of persons exposed to dioxin following an accident which
22 occurred in the BASF on 17 November 1953. In: Proceedings of the 5th International Conference Medichem, 1977.
23 San Francisco, CA. pp. 228-236.
24 Thiess, AM; Frentzel-Beyme, R; Link, R. (1982) Mortality study of persons exposed to dioxin in a trichlorophenol-
25 process accident that occurred in the BASF AG on November 17, 1953. Am J Ind Med 3(2): 179-189.
26 Tian, Y; Ke, S; Denison, MS; et al. (1999) Ah receptor and NF-kB interactions, a potential mechanism for dioxin
27 toxicity. J Biol Chem 274: 510-515.
28 t'Mannetje, A; McLean, D; Cheng, S; et al. (2005) Mortality in New Zealand workers exposed to phenoxy
29 herbicides and dioxins. Occup Environ Med 62(l):34-40.
30 Toide, K; Yamazaki, JH; Nagashima, R; et al. (2003) Aryl hydrocarbon hydroxylase represents CYP1B1 and not
31 CYP1 Al, in human freshly isolated white cells: Trimodal distribution of Japanese population according to induction
32 ofCYPlBl mRNA by environmental dioxins. Cancer Epidemiol Biomark Prev 12:219-222.
33 Toth, L; Somfai-Relle, S; Sugar, J; et al. (1979) Carcinogenicity testing of herbicide 2,4,5-trichlorophenoxyethanol
34 containing dioxin and of pure dioxin in Swiss mice. Nature 278:548-549.
35 Tritscher, AM; Mahler, J; Portier, CJ; et al. (2000) Induction of lung lesions in female rats following chronic
36 exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol Pathol 28:761-769.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-25 DRAFT—DO NOT CITE OR QUOTE
-------
1 Tuomisto, JT; Viluksela, M; Pohjanvirta, R; et al. (1999) The AH receptor and a novel gene determine acute toxic
2 responses to TCDD: segregation of the resistant alleles to different rat lines. Toxicol Appl Pharm 155(1):71—81.
3 Tuomisto, JT; Wilson, AM; Evans, JS; et al. (2008) Uncertainty in mortality response to airborne fine particulate
4 matter: combining European air pollution experts. Reliab Eng Syst Saf 93(5):775-777.
5 Umemura, T; Kai, S; Hasegawal, R; et al. (1999) Pentachlorophenol (PCP) produces liver oxidative stress and
6 promotes but does not initiate hepatocarcinogenesis in B6C3F1 mice. Carcinogenesis 20(6): 1115-1120.
7 U.S. DOE (Department of Energy). (1992) DOE standard, hazard categorization, and accident analysis techniques
8 for compliance with DOE Order 5480.23, nuclear safety analysis reports. DOE-STD-1027-92 (Change Notice No. 1,
9 Sept. 1997), Washington, DC. Available at:
10 http://www.hss.doe.gOv/nuclearsafety/ns/techstds/standard/stdl027/sl027cnl.pdf.
11 U.S. EPA. (1996a) Proposed guidelines for carcinogen risk assessment. Risk Assessment Forum, Washington, DC.
12 Federal Register 61:17960-18011.
13 U.S. EPA. (1998b) Guidelines for neurotoxicity risk assessment. Federal Register 63(93):26926-26954. Available
14 from: http://cfpub.epa.gov/ncea/raf/recordisplay.cfm?deid= 12479.
15 U.S. EPA. (2000) Benchmark dose technical guidance document. October. External review draft. Washington, DC.
16 EPA/630/R-00/001.
17 U.S. EPA. (2003) Exposure and human health reassessment of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and
18 related compounds [NAS review draft]. Volumes 1-3. National Center for Environmental Assessment, Washington,
19 DC; EPA/600/P-00/001 Cb. Available at: http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
20 U.S. EPA. (2005) Guidelines for carcinogen risk assessment. Risk Assessment Forum, Washington, DC.
21 EPA/630/P-03/001F. Available at http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=116283.
22 U.S. EPA. (2008a) Framework for application of the toxicity equivalence methodology for polychlorinated dioxins,
23 furans, and biphenyls in ecological risk assessment. Risk Assessment Forum, Washington, DC; EPA/100/R-08/004
24 U.S. EPA. (2008b) 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) dose-response studies: preliminary literature
25 search results and request for additional studies. National Center for Environmental Assessment, Cincinnati, OH;
26 EPA/600/R-08/119. Available at: http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=199923.
27 U.S. EPA. (2008c) Benchmark dose technical guidance document. Risk Assessment Forum, Washington, DC;
28 EPA/630/R-00/0001F.
29 U.S. EPA. (2009a) Integrated Risk Information System (IRIS). National Center for Environmental Assessment,
30 Washington, DC. Available at: http://www.epa.gov/ncea/iris.
31 U.S. EPA. (2009b) Summary of U.S. EPA dioxin workshop: February 18-20, 2009. National Center for
32 Environmental Assessment, Cincinnati, OH; EPA/600/R-09/027.
33 U.S. EPA. (2009c) Using probabilistic methods to enhance the role of risk analysis in decision-making with case
34 study examples. Risk Assessment Forum, Washington, DC. EPA/100/R-09/001.
35 U.S. EPA. (2010) Recommended toxicity equivalency factors (TEFs) for human health risk assessments of dioxin
36 and dioxin-like compounds [external review draft]. Risk Assessment Forum, Washington, DC; NCEA-C-2100.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-26 DRAFT—DO NOT CITE OR QUOTE
-------
1 U.S. NRC (Nuclear Regulatory Commission). (1975) Reactor safety study - an assessment of accident risks in U.S.
2 commercial nuclear power plants. NUREG-75/014 (WASH-1400). Prepared by N. Rasmussen (MIT) et al., for the
3 U.S. Nuclear Regulatory Commission, Rockville, MD.
4 U.S. NRC (Nuclear Regulatory Commission). (1981) Fault tree handbook. NUREG-0492, Systems and Reliability
5 Research, Office of Nuclear Regulatory Research, Rockville, MD. Available at: http://www.nrc.gov/reading-
6 rm/doc-collections/nuregs/staff/sr0492/sr0492.pdf.
7 U.S. NRC (Nuclear Regulatory Commission). (1983) A guide to the performance of probabilistic risk assessments
8 for nuclear power plants. Final Report. NUREG/CR-2300. Rockville, MD. Available at:
9 http://www.nrc.gov/reading-rm/doc-collections/nuregs/contract/cr2300/vol2/cr2300v2-a.pdf.
10 U.S. NRC (Nuclear Regulatory Commission). (1991) Severe accident risks: an assessment for five U.S. nuclear
11 power plants. NUREG-1150. Final Summary Report, Final Report. Rockville, MD. Selected portions available at:
12 http://www.nrc.gov/reading-rm/doc-collections/nuregs/staff/srll50/; http://www.nrc.gov/reading-rm/doc-
13 collections/nuregs/staff/srl 150/v3/srl 150v3 .pdf.
14 Van Birgelen, AP; Van den Berg, M. (2000) Toxicokinetics. Food Addit Contam 17:267-273.
15 Van Birgelen, AP; Van der Kolk, J; Fase, KM; et al. (1995) Subchronic dose-response study of 2,3,7,8-
16 tetrachlorodibenzo-p-dioxin in female Sprague-Dawley rats. Toxicol Appl Pharmacol 132:1-13.
17 Van den Berg, M; de Vroom, E; Olie, K; et al. (1986) Bioavailability of PCDDs and PCDFs of fly ash after semi-
18 chronic oral ingestion by guinea pig and Syrian golden hamster. Chemosphere 15(4):519—53 3.
19 Van den Berg, M; Birnbaum, L; Bosveld, AT; et al. (1998) Toxic equivalency factors (TEFs) for PCBs, PCDDs,
20 PCDFs for humans and wildlife. Environ Health Perspect 106(12):775-792.
21 Van den Berg, M; Birnbaum, LS; Denison, M; et al. (2006) The 2005 World Health Organization reevaluation of
22 human and mammalian toxic equivalency factors for dioxins and dioxin-like compounds. Tox Sci 93(2):223-241.
23 Available at: http://toxsci.oxfordjournals.Org/cgi/reprint/93/2/223.
24 Van der Molen, G; Kooijman, A; Slob, W. (1996) A generic toxicokinetic model for persistent lipophilic
25 compounds in humans: an application to TCDD. Fund Appl Toxicol 31:83-94.
26 Van der Molen, GW; Kooijman, SALM; Michalek, JE; et al. (1998) The estimation of elimination rates of persistent
27 compounds: a re-analysis of 2,3,7,8-tetrachlorodibenzo-p-dioxin levels in Vietnam veterans. Chemosphere 37(9-
28 12): 1833—1844.
29 Van der Molen, GW; Kooijman, BA; Wittsiepe, J; et al. (2000) Estimation of dioxin and furan elimination rates with
30 a pharmacokinetic model. J Expo Anal Environ Epidemiol 10:579-585.
31 Vanden Heuvel, JP; Clark, GC; Tritscher, A; et al. (1994) Accumulation of polychlorinated dibenzo-p-dioxins and
32 dibenzofurans in liver of control laboratory rats. Fundam Appl Toxicol 23:465-469.
33 Vanni, H; Kazeros, A; Wang, R; et al. (2009) Cigarette smoking induces overexpression of a fat-depleting gene
34 AZGP1 in the human airway epithelium. Chest 135(5): 1197-1208. Available at:
3 5 http://chestjournal.chestpubs.Org/content/135/5/l 197.full.pdf+html.
36 Viluksela, M; Bager, Y; Tuomisto, JT. (2000) Liver tumor-promoting activity of 2,3,7,8-tetrachlorodibenzo-p-
37 dioxin (TCDD) in TCDD-sensitive and TCDD-resistant rat strains. Cancer Res 60:6911-6920.
38 Vos, JG; Moore, JA; Zinkl, JG. (1973) Effect of 2,3,7,8-tetrachlorodibenzo-p-dioxin on the immune system of
3 9 laboratory animals. Environ Health Perspect 5:149-162.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-27 DRAFT—DO NOT CITE OR QUOTE
-------
1 Walker, NJ; Portier, CJ; Lax, SF; et al. (1999) Characterization of the dose-response of CYP1B1, CYP1A1, and
2 CYP1A2 in the liver of female Sprague-Dawley rats following chronic exposure to 2,3,7,8-tetrachlorodibenzo-p-
3 dioxin. Toxicol Appl Pharmacol l;154(3):279-286.
4 Wang, X; Santostefano, MJ; Evans, MV; et al. (1997) Determination of parameters responsible for pharmacokinetic
5 behavior of TCDD infemale sprague-dawley rats. Toxicol Appl Pharmacoll47:151-168.
6 Wang, X; Santostefano, MJ; Devito, MJ; et al. (2000). Extrapolation of a PBPK model for dioxins across dosage
7 regimen, gender, strain, and species. Toxicol Sci 56(l):49-60.
8 Wang, SL; Su, PH; Jong, SB; et al. (2005) In utero exposure to dioxins and polychlorinated biphenyls and its
9 relations to thyroid function and growth hormone in newborns. Environ Health Perspect 113:1645-1650.
10 Warner, M; Eskenazi, B; Mocarelli, P; et al. (2002) Serum dioxin concentrations and breast cancer risk in the
11 Seveso Women's Health Study. Environ Health Perspect 110(7):625-628.
12 Warner, M; Samuels, S; Mocarelli, P; et al. (2004) Serum dioxin concentrations and age at menarche. Environ
13 Health Perspect 112(13): 1289-1292.
14 Warner, M; Eskenazi, B; Olive, DL; et al. (2007) Serum dioxin concentrations and quality of ovarian function in
15 women of Seveso. Environ Health Perspect 115(3):336—340.
16 Weber, R; Schmitz, H-J; Schrenk, D; et al. (1997) Metabolic degradation, inducing potency, and metabolites of
17 fluorinated and chlorinated-fluorinated dibenzodioxins and dibenzofurans. Chemosphere 34(l):29-40.
18 Wendling, JM; Orth, RG; Poiger, H. (1990) Determination of [3H]-2,3,7,8-tetrachlorodibenzo-p-dioxin in human
19 feces to ascertain its relative metabolism in man. Anal Chem 62(8):796-800.
20 White, KL, Jr; Lysy, HH; McCay, JA; et al. (1986) Modulation of serum complement levels following exposure to
21 polychlorinated dibenzo-p-dioxins. Toxicol Appl Pharmacol 84:209-219.
22 WHO (World Health Organization). (1978) International Classification of Diseases: Ninth Revision. ISBN-13
23 9789241541336.
24 WHO (World Health Organization). (1988) Assessment of the health risk of dioxins: re-evaluation of the tolerable
25 daily intake (TDI). WHO Consultation May 25-29, 1998. Geneva, Switzerland: WHO European Centre for
26 Environmental Health and International Programme on Chemical Safety.
27 WHO (World Health Organization). (1998) Executive summary, assessment of health risk of dioxins: re-evaluation
28 of the tolerable daily intake (TDI), WHO Consultation, May 25-29.
29 WHO (World Health Organization). (2005) Chemical-specific adjustment factors for interspecies differences and
30 human variability: guidance document for use of data in dose/concentration-response assessment. Harmonization
31 Project Document 2, IPCS project on Harmonization of Approaches to the Assessment of Risk from Exposure to
32 Chemicals. Geneva.
33 Whysner, J; Williams, GM. (1996) 2,3,7,&tetrachlorodibenzo+hdioxin mechanistic data and risk assessment: gene
34 regulation, cytotoxicity, enhanced cell proliferation, and tumor promotion. Pharmacol Ther 71(1/2): 193-223.
3 5 Wittsiepe, J; Erlenkamper, B; Welge, P; et al. (2007) Bioavailability of PCDD/F from contaminated soil in young
36 Goettingen minipigs. Chemosphere 67(9):S355-S364.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-28 DRAFT—DO NOT CITE OR QUOTE
-------
1 Wong, T K; Domin, BA; Bent, PE; et al. (1986) Correlation of placental microsomal activities with protein detected
2 by antibodies to rabbit cytochrome P-450 isozyme 6 in preparations from humans exposed to polychlorinated
3 biphenyls, quaterphenyls, and dibenzofurans. Cancer Res 46:999-1004
4 Woods, CG; Burns, A; Bradford, BU; et al. (2007) WY-14,643-induced cell proliferation and oxidative stress in
5 mouse liver are independent of NADPH oxidase. Toxicol Sci 98(2):366-374.
6 Wyde, M; Wong, VA; Kim, AH; et al. (2001) induction of hepatic 8-oxo-deoxyguanosine adducts by 2,3,7,8-
7 tetrachlorodibenzo-p-dioxin in Sprague-Dawley rats is female-specific and estrogen-dependent. Chem Res Toxicol
8 14:849-855.
9 Wyde, M; Cambre, T; Lebetkin, M; et al. (2002) Promotion of altered hepatic foci by 2,3,7,8-tetrachlorodibenzo-p-
10 dioxin and 17b-estradiol in male Sprague-Dawley rats. Toxicol Sci 68:295-303.
11 Yang, JZ; Agarwal, SK; Foster, WG. (2000) Subchronic exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin modulates
12 the pathophysiology of endometriosis in the cynomolgus monkey. Toxicol Sci 56:374-381.
13 Youakim, S. (2006) Risk of cancer among firefighters: a quantitative review of selected malignancies. Int Arch
14 Occup Environ Health 61(5):223-231.
15 Zack, JA; Gaffey, WR. (1983) A mortality study of workers employed at the Monsanto Company plant in Nitro,
16 West Virginia. Environ Sci Res 26:575-591.
17 Zack, JA; Suskind, R. (1980) The mortality experience of workers exposed to tetrachlorodibenzodioxin in a
18 trichlorophenol process accident. J Occup Med 22:11-14.
19 Zareba, G; Hojo,R; Zareba, GM; et al. (2002) Sexually dimorphic alterations of brain cortical dominance in rats
20 prenatally exposed to TCDD. J Appl Toxicol 22:129-137.
21 Zober, A; Papke, O. (1993) Concentrations of PCDDs and PCDFs in human tissue 36 years after accidental dioxin
22 exposure. Chemosphere 27:413-418.
23 Zober, A; Messerer, P; Huber, P. (1990) Thirty-four-year mortality follow-up of BASF employees exposed to
24 2,3,7,8-TCDD after the 1953 accident. Int Arch Occup Environ Health 62(2): 139-157.
25 Zober, A; Ott, MG; Messerer, P. (1994) Morbidity follow-up study of BASF employees exposed to 2,3,7,8-
26 tetrachlorodibenzo-p-dioxin (TCDD) after a 1953 chemical reactor incident. J Occup Environ Med 51:479-486.
27 Zober, A; Schilling, D; Ott, M; et al. (1998) Helicobacter pylori infection: prevalence and clinical relevance in a
28 large company. Occup Environ Med 40(7):586-594.
29
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/2010 R-29 DRAFT—DO NOT CITE OR QUOTE
-------
DRAFT
DO NOT CITE OR QUOTE
January 2010
Agency/Interagency Review Draft
APPENDIX A
Dioxin Workshop Report
NOTICE
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
EPA/600/R-09/027
May 2009
Summary of U.S. EPA
Dioxin Workshop
February 18-20, 2009
Cincinnati, Ohio
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH 45268
-------
DISCLAIMER
This document summarizes the discussions presented at the Dioxin Workshop in
February 2009, in Cincinnati, OH, as documented by the Session Co-Chairs. This document is
not all inclusive or binding. Conclusions and recommendations to the U.S. EPA may not
represent full consensus. The views expressed in this document are those of the Dioxin
Workshop Panelists and do not necessarily reflect the views and policies of the U.S.
Environmental Protection Agency. Mention of trade names or commercial products does not
constitute endorsement or recommendation for use.
Preferred Citation:
U.S. Environmental Protection Agency (U.S. EPA). (2009) Summary of U.S. EPA Dioxin Workshop:
February 18-20, 2009. U.S. Environmental Protection Agency, National Center for Environmental Assessment,
Cincinnati, OH. EPA/600/R-09/027.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-ii DRAFT—DO NOT CITE OR QUOTE
-------
TABLE OF CONTENTS
DIOXIN WORKSHOP TEAM A-iv
ACKNOWLEDGMENTS A-iv
INTRODUCTION A-l
REFERENCES A-2
SCIENTIFIC WORKSHOP TO INFORM THE TECHNICAL WORK PLAN FOR U.S.
EPA'S RESPONSE TO NAS COMMENTS ON THE HEALTH EFFECTS OF
DIOXIN PRESENTED IN U.S. EPA'S DIOXIN REASSESSMENT A-3
SESSION 1: QUANTITATIVE DOSE-RESPONSE MODELING ISSUES A-3
SESSION 2: IMMUNOTOXICITY A-6
SESSION 3 A: DOSE-RESPONSE FOR NEUROTOXICITY AND
NONREPRODUCTIVE ENDOCRINE EFFECTS A-8
SESSION 3B: DOSE-RESPONSE FOR CARDIOVASCULAR TOXICITY
AND HEPATOTOXICITY A-l 1
SESSION 4A: DOSE-RESPONSE FOR CANCER A-13
SESSION 4B: DOSE-RESPONSE FOR
REPRODUCTIVE/DEVELOPMENTAL TOXICITY A-16
SESSION 5: QUANTITATIVE UNCERTAINTY ANALYSIS OF DOSE-
RESPONSE A-20
APPENDIX A: 2009 U.S. EPA DIOXIN WORKSHOP AGENDA A-24
APPENDIX B: 2009 U.S. EPA DIOXIN WORKSHOP QUESTIONS TO GUIDE
PANEL DISCUSSIONS A-31
APPENDIX C: 2009 U.S. EPA DIOXIN WORKSHOP DRAFT SELECTION
CRITERIA TO IDENTIFY KEY IN VIVO MAMMALIAN STUDIES THAT
INFORM DOSE-RESPONSE MODELING FOR
2,3,7,8-TETRACHLORODIBENZO-p-DIOXIN (TCDD) A-34
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-iii DRAFT—DO NOT CITE OR QUOTE
-------
DIOXIN WORKSHOP TEAM
The Dioxin Workshop Team, under the leadership of Peter W. Preuss, Director, NCEA,
comprised the following members:
National Center for Environmental Assessment, Office of Research and Development,
U.S. Environmental Protection Agency, Cincinnati, OH 45268
Belinda S. Hawkins
Janet Hess-Wilson
Glenn Rice
Jeff Swartout
Linda K. Teuschler
Bette Zwayer
Argonne National Laboratory, Argonne, IL 60439
Maryka H. Bhattacharyya
Andrew Davidson
Mary E. Finster
Margaret M. MacDonell
David P. Peterson
ACKNOWLEDGMENTS
The Track Group, Alexandria, VA 22312
Kara Hennigan
Alan Minton
Brandy Quinn
ECFlex, Inc., Fairborn, OH 45324
Dan Heing
Heidi Glick
Amy Prues
Lana Wood
IntelliTech Systems, Inc., Fairborn, OH 45324
Cris Broyles
Luella Kessler
Stacey Lewis
Linda Tackett
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-iv DRAFT—DO NOT CITE OR QUOTE
-------
INTRODUCTION
This document provides a summary of the Scientific Workshop to Inform EPA's
Response to National Academy of Science Comments on the Health Effects of Dioxin in EPA's
2003 Dioxin Reassessment. The U.S. Environmental Protection Agency (U.S. EPA) and
Argonne National Laboratories (ANL), through an inter-Agency agreement with the U.S.
Department of Energy, convened this scientific workshop ("Dioxin Workshop") on February
18-20, 2009, in Cincinnati, Ohio. The goals of the Dioxin Workshop were to identify and
address issues related to the dose-response assessment of 2,3,7,8-tetrachl orodibenzo-/>dioxin
(TCDD). This report summarizes the discussions and conclusions from this workshop.
Previously, at the request of the U.S. EPA, the National Academy of Sciences (NAS) prepared a
report, Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment
(NAS, 2006), which made a number of recommendations to improve the U.S. EPA's risk
assessment for TCDD (U.S. EPA, 2003). The 3-day Dioxin Workshop was convened
specifically to ensure that the U.S. EPA's response to the NAS recommendations focuses on the
key issues and reflects the most meaningful science.
The Dioxin Workshop included seven scientific sessions:
(1) Session 1:
(2) Session 2:
(3) Session 3A
(4) Session 3B
(5) Session 4A
(6) Session 4B
(7) Session 5:
Quantitative Dose-Response Modeling Issues
Immunotoxicity
Dose-Response for Neurotoxicity and Nonreproductive Endocrine Effects
Dose-Response for Cardiovascular Toxicity and Hepatotoxicity
Dose-Response for Cancer
Dose-Response for Reproductive/Developmental Toxicity
Quantitative Uncertainty Analysis of Dose-Response
During each session, the U.S. EPA asked a panel of expert scientists to:
• identify and discuss the technical challenges involved in addressing the key NAS
comments on the TCDD dose-response assessment in the U.S. EPA Reassessment
(U.S. EPA, 2003);
• discuss approaches for addressing the key NAS comments; and
• identify important published, independently peer-reviewed literature, particularly studies
describing epidemiologic and in vivo mammalian bioassays, which are expected to be
most useful for informing the U.S. EPA's response.
The sessions were followed by open comment periods during which members of the
audience were invited to address the Panels. At the conclusion of the open comment periods, the
Panel Co-Chairs were asked to summarize and present the results of the panel discussions. The
summaries could include minority opinions stated by panelists. The main points derived from
the session summaries were used to prepare this document. Additionally, this document includes
a list of the session panelists and their affiliations and three appendices. Appendix A presents
the Dioxin Workshop Agenda. Appendix B identifies the charge questions presented to the
Panel. Appendix C describes draft study selection criteria proposed by the Dioxin Workshop
Team for consideration by the workshop panelists.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-l DRAFT—DO NOT CITE OR QUOTE
-------
REFERENCES
NAS (National Academy of Sciences). 2006. Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment. National Academies Press, Washington, DC
(July). Available at http://www.nap.edu/catalog.php7record id=l 1688.
U.S. EPA (U.S. Environmental Protection Agency). 2003. Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds. NAS
review draft, Volumes 1-3 (EPA/600/P-00/001Cb, Volume 1). U.S. Environmental Protection
Agency, National Center for Environmental Assessment, Washington, DC (December).
Available at http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-2 DRAFT—DO NOT CITE OR QUOTE
-------
SCIENTIFIC WORKSHOP TO INFORM I II I TECHNICAL WORK PLAN FOR U.S.
EPA'S RESPONSE TO NAS COMMENTS ON THE HEALTH EFFECTS OF DIOXIN
PRESENTED IN U.S. EPA'S DIOXIN REASSESSMENT
Dioxin Workshop Co-Chairs: Peter W. Preuss and Glenn Rice
The Dioxin Workshop session summaries were prepared by the session panel Co-Chairs
with input from the panelists, as requested by the U.S. EPA prior to the workshop. The Co-
Chairs subsequently presented these summaries to all of the workshop participants during
designated periods at the workshop. In these summaries, the U.S. EPA asked that the Co-Chairs
summarize the key issues from the panel discussions. Because the sessions were not designed to
achieve consensus among the panelists, the summaries do not necessarily represent consensus
opinions; rather, they reflect the essence of the panel discussions. Some of the specific points
may represent the views of multiple panelists, while others only the views of a single panelist.
Prior to the summarizations, there were opportunities for public comments on the discussion
topics. Some Co-Chairs met with their sessions' panelists after their sessions ended to develop
these summaries, while others developed reports based on their personal notes. Because Session
5 was the last session of the workshop—with little time provided to develop the summary—the
Co-Chairs circulated a draft for comment by the Session 5 panelists after the workshop, prior to
finalizing the session summary. The U.S. EPA collected the session summaries and then
prepared this document. A draft of this document was distributed to all of the session Co-Chairs
to provide them with a final opportunity to comment and make revisions. Finally, it should be
noted that U.S. EPA was not prescriptive to the session Co-Chairs with respect to the format of
the presentation materials and provided no specific instructions, resulting in unique formats
among the session summaries.
SESSION 1: QUANTITATIVE DOSE-RESPONSE MODELING ISSUES
This session discussed the general dose-response modeling issues related to TCDD.
Many of these issues were highlighted by NAS (2006). There was a general introductory
presentation on TCDD kinetics, including information and uncertainties pertaining to the
conversion of administered doses in animals to human body burden (BB) and additivity to
background issues. This presentation was followed by a Panel discussion on the state of the
science regarding dioxin dose-response modeling issues.
Session 1 Panelists (Session Co-Chairs are identified by asterisk)
• Bruce Allen, Bruce Allen Consulting
• Lesa Aylward, Summit Toxicology
• Roger Cooke, Resources for the Future
• Kenny Crump, Louisiana Tech University
• MikeDeVito, U.S. EPA
• Dale Hattis, Clark University
• Rick Hertzberg, Biomath Consulting
• Rob McDowell, U.S. Department of Agriculture
• Jim Olson, State University of New York, University at Buffalo
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-3 DRAFT—DO NOT CITE OR QUOTE
-------
• *Lorenz Rhomberg, Gradient
• Woody Setzer, U.S. EPA
• *Jeff Swartout, U.S. EPA
Please note that the use of the term "concluded" or "recommended" in this summary does not mean that a consensus
was reached. Session Summaries were written from the material prepared by the non-EPA/ANL Co-Chair and
represent a synopsis of the panel discussions.
Key Study Selection Criteria
The Panel discussed the advantages and disadvantages of using key study criteria
(Appendix C). They concluded that a priori criteria foster transparency and consistency, and
could deflect a posteriori criticism. However, the Panel also acknowledged that having a priori
criteria could introduce the potential for excluding useful data. Although the key study criteria
provided by the U.S. EPA listed studies using TCDD only as a criterion, the Panel posed the
possibility of using closely related dioxin-like compounds (DLCs) as surrogates for TCDD. The
criterion for use of data from mammalian studies only was one criterion that received generalized
support due to the lack of extrapolation protocols for nonmammalian species. The Panel also
discussed the specific exposure-duration criterion and asked if there should be a preference for
longer-term rather than acute studies. The Panel made three suggestions to modify U.S. EPA's
key study selection criteria:
(1) Define more relevant exposure-level (i.e., dose) cut points using tissue concentrations.
(2) Reword statistical criteria to include do-it-yourself analysis.
(3) Reword the response criteria to clarify "outside of normal range."
Dose Metrics
The Panel discussed the relative merits of various measures of dose for modeling TCDD
dose response. One general conclusion was that tissue concentration (TC) is the preferred
metric, especially lipid-adjusted TC, because this measure more closely approximates exposures
close to the target tissue when compared to administered doses. However, the Panel
acknowledged that these data are often unavailable. They further noted that BB, which is
defined as the concentration of TCDD in the body (ng/kg body weight) (U.S. EPA, 2003), might
be useful as a surrogate for TC provided the two measures were proportional.
The Panel suggested that a linear approach to BB estimation, which was utilized by
U.S. EPA (2003), is too simplistic because this approach does not take into account toxicokinetic
issues related to TCDD—e.g., sequestration in the liver and fat, age-dependent elimination, and
changing elimination rates over time. The Panel recommended the use of kinetic/mechanistic
modeling to the extent possible to quantify tissue-based metrics.
The Panel raised the issue of whether the preferred dose metric would be different for
different endpoints and exposure durations. This led to the Panel's comment that the peak
exposure might be a more important metric than average BB for variable exposure scenarios.
Given this discussion about different exposure durations being relevant to a specific endpoint,
the Panel suggested that the U.S. EPA also consider peak measures in dose-response modeling.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-4 DRAFT—DO NOT CITE OR QUOTE
-------
The last point raised in this part of the discussion centered on the possibility of dose
errors in experimental studies. The Panel highlighted the need for the U.S. EPA to consider dose
error (i.e., uncertainty in the x-axis of the dose-response curve) when using dose surrogates.
Dose-Response Modeling of Mammalian Bioassays
The Panel considered several issues related to dose-response modeling of mammalian
bioassay data for TCDD: supralinearity and incomplete response data ("anchoring"), defining the
benchmark response (BMR) level with respect to establishing the point of departure (POD), and
the use of threshold modeling—as further explained below.
The Panel discussed the specific issues of supralinearity and anchoring raised by the
U.S. EPA with respect to modeling noncancer endpoints. The panel recognized that, for many of
the most sensitive endpoints, the response at the lowest dose is high (e.g., quantal responses
above 25% and continuous endpoints differ substantially from the mean, often implying 100%
incidence in the treated animals). This lack of response anchoring at the low end of the dose-
response curve (near the BMR) results in the higher responses determining the shape of the
curve.
The Panel asked whether new tools might be needed or whether the current tools could be
applied differently. In the context of developing new tools, the Panel emphasized the need for
collaboration between biologists and mathematicians. When discussing application, the Panel
suggested that the problem with supralinearity might be overcome by simply dropping the
requirement for using the lower bound on the Benchmark Dose. In addition, the Panel posed
several more approaches for further consideration in dose-response modeling by the U.S. EPA:
(1) Combine similar data sets to fill in data gaps.
(2) Use mechanistic approaches to model the data gaps.
(3) Dichotomize continuous data.
Finally, the Panel acknowledged that, in certain situations, there simply may not be enough
information to provide meaningful answers.
The Panel discussed the BMR level for establishing a POD in the context of deriving a
Reference Dose (RfD). The Panel generally agreed that, while the effective dose level (ED0i)
used in the 2003 Reassessment may be useful for comparative analysis across endpoints, the
EDoi estimates developed for all endpoints considered in the Reassessment were not appropriate
for deriving an RfD because they were not based on the effect's adversity. The panel noted that
EDoi also is much lower than typical EPA BMR levels. The Panel recommended that the U.S.
EPA work to define endpoint-specific BMRs based on the consideration of adversity. Given that
the same uncertainty factor framework is applied to all PODs, the Panel emphasized the need for
consistency in BMRs; numerical consistency is needed for quantal BMRs and consistency in the
choice of biological relevance should be applied for continuous BMRs.
The Panel generally discouraged threshold modeling by stating that thresholds are very
difficult to pin down and suggested that the lower bound may always be zero.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-5 DRAFT—DO NOT CITE OR QUOTE
-------
Dose-Response Modeling of Epidemiological Studies
The Panel noted that many studies have been published with measured concentrations of
TCDD that could be used for dose reconstruction. In this discussion, the Panel acknowledged
that use of these data would entail dealing with toxicity equivalence (TEQ) issues and
pharmacokinetic (PK) modeling. Pertaining to the use of these data for quantitative risk
assessment by the U.S. EPA, the Panel posed the question, "At what point does indirect or
confounded human data supersede controlled animal bioassay data?", or alternatively, "How
much human data uncertainty can we tolerate?" The Panel suggested, at the least, that the
epidemiologic data could be used to "ground-truth" the animal bioassay modeling results.
Supporting Information
The Panel acknowledged that Ah receptor (AhR) binding affinities are not necessarily
tied to endpoint sensitivity, but they reiterated the need to consider mechanistic modeling to aid
in developing appropriate dose metrics or filling in data gaps in the existing dose-response data.
References
NAS (National Academy of Sciences). 2006. Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment. National Academies Press, Washington, DC
(July). Available at http://www.nap.edu/catalog.php?record id=l 1688.
U.S. EPA (U.S. Environmental Protection Agency). 2003. Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds. NAS
Review Draft (EPA/600/P-00/001Cb). U.S. Environmental Protection Agency, National Center
for Environmental Assessment, Washington, DC. Available at
http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
SESSION 2: IMMUNOTOXICITY
The U.S. EPA plans to consider development of a quantitative dose-response assessment
for the immunologic effects associated with TCDD exposure. Such an assessment would be
based on information in U.S. EPA (2003), NAS (2006) and key studies identified in this
workshop. The purpose of this session was to identify and discuss key issues pertaining to dose-
response assessment for dioxin-induced immunologic effects.
Session 2 Panelists (Session Co-Chairs are identified by asterisk)
• Roger Cooke, Resources for the Future
• Rob Goble, Clark University
• *Belinda Hawkins, U.S. EPA
• Nancy Kerkvliet, Oregon State University
• Manolis Kogevinas, Centre for Research in Environmental Epidemiology
• Robert Luebke, U.S. EPA
• Paolo Mocarelli, University of Milan
• * Allen Silverstone, State University of New York, Upstate Medical University
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-6 DRAFT—DO NOT CITE OR QUOTE
-------
• Courtney Sulentic, Wright State University
• Nigel Walker, National Institute of Environmental Health Sciences
Please note that the use of the term "concluded" or "recommended" in this summary does not mean that a consensus
was reached. Session Summaries were written from the material prepared by the non-EPA/ANL Co-Chair and
represent a synopsis of the panel discussions.
Key Study Selection Criteria
The Panel first addressed the Key Study Selection Criteria proposed by the U.S. EPA
(Appendix C). The Panel raised the issue that the key study criteria do not apply to most studies
designed to investigate immunotoxicity, including those used to calculate ED0iS (U.S. EPA,
2003). The Panel observed that most dioxin immunotoxicity studies are relatively high dose
(>200 ng/kg-d) acute studies and/or use parenteral rather than oral administration.
The Panel discussed several studies often considered important for assessing the
immunotoxic effects of TCDD exposure. The Oughton et al. (1995) mouse bioassay was
discussed and, although the study does meet the proposed criteria, it could not be considered a
key study; specifically, the Panel contended that since there were no functional alterations
observed or measured in this bioassay, the changes in cellular phenotypes are only "suggestive"
of immune alterations and cannot be regarded as having immunopathologic significance.
The Panel discussed two additional studies for further consideration by the U.S. EPA:
• Baccarelli et al. (2002). The Panel discussed this as a potentially key human
epidemiological study that should be reviewed and considered further by the U.S. EPA.
It measured the level of IgG, demonstrating a significant decline relative to dioxin body
burdens.
• Smialowicz et al. (2008). The Panel noted that this study identified the antibody response
to sheep red blood cells (SRBCs) as the critical effect, labeling this protocol as a
functional assay. The Panel stated that if modeled, the U.S. EPA could calculate the
BMR for this endpoint as 1 standard deviation from the control mean.
References
Baccarelli, A., P. Mocarelli, D.G. Patterson et al. 2002. Immunologic effects of dioxin: New
results from Seveso and comparison with other studies. Environ. Health Perspect.
110(12): 1169-1173.
NAS (National Academy of Sciences). 2006. Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment. National Academies Press, Washington, DC
(July). Available at http://www.nap.edu/catalog.php7record id=l 1688.
Oughton, J.A., C.B. Pereira, G.K. Dekrey, J.M. Collier, A.A. Frank and N.I. Kerkvliet. 1995.
Phenotypic analysis of spleen, thymus, and peripheral blood cells in aged C57BI/6 mice
following long-term exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol. Sci. 25(l):60-69.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-7 DRAFT—DO NOT CITE OR QUOTE
-------
Smialowicz, R.J., M.J. DeVito, W.C. Williams and L.S. Birnbaum. 2008. Relative potency
based on hepatic enzyme induction predicts immunosuppressive effects of a mixture of
PCDDS/PCDFS and PCBS. Toxicol. Appl. Pharmacol. 227(3):477-484.
U.S. EPA (U.S. Environmental Protection Agency). 2003. Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds. NAS
Review Draft (EPA/600/P-00/001Cb). U.S. Environmental Protection Agency, National Center
for Environmental Assessment, Washington, DC. Available at
http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
SESSION 3A: DOSE-RESPONSE FOR NEUROTOXICITY AND NONREPRODUCTIVE
ENDOCRINE EFFECTS
The U.S. EPA plans to consider development of a quantitative dose-response assessment
for neurological and/or nonreproductive endocrine effects associated with TCDD exposure.
Such an assessment would be based on information in U.S. EPA (2003), NAS (2006) and key
studies identified in this workshop. The purpose of this session was to identify and discuss key
issues pertaining to dose-response assessment for dioxin-induced neurological and/or
nonreproductive endocrine effects.
Session 3A Panelists (Session Co-Chairs are identified by asterisk)
• *Maryka Bhattacharyya, Argonne National Laboratory
• Mike DeVito, U.S. EPA
• Mary Gilbert, U.S. EPA
• Rob Goble, Clark University
• Nancy Kerkvliet, Oregon State University
• Fumio Matsumura, University of California-Davis
• Paolo Mocarelli, University of Milan
• Chris Portier, National Institute of Environmental Health Sciences
• Lorenz Rhomberg, Gradient
• Allen Silverstone, State University of New York, Upstate Medical University
• Marie Sweeney, National Institute of Occupational Safety and Health
• *Bernie Weiss, University of Rochester
Please note that the use of the term "concluded" or "recommended" in this summary does not mean that a consensus
was reached. Session Summaries were written from the material prepared by the non-EPA/ANL Co-Chair and
represent a synopsis of the panel discussions.
What Are the Key Questions Regarding These Endpoints?
The Panel used the following question to initiate discussion: "Are there identifiable
indices of neurotoxicity and nonreproductive endocrine effects in animal studies and human
populations? " Under this discussion topic, the Panel discussed three endpoints: neurotoxicity
(with focus on developmental exposures), thyroid dysfunction (e.g., thyroid hormone deficits),
and diabetes. The Panel also addressed the relevance of windows of vulnerability to each
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-8 DRAFT—DO NOT CITE OR QUOTE
-------
endpoint. The Panel acknowledged that, in some cases, the window of exposure may precede the
window of expression of toxicity.
Epidemiological Study Selection
Developmental Neurotoxicity
The Panel recognized that an unusual feature for this endpoint is that there are sufficient
human data for dose-response modeling (e.g., Dutch children [Huisman et al., 1995; Patandin et
al., 1999] and U.S. children [Jacobson and Jacobson, 1996]) and there is an internal dose metric
(serum concentrations). Additionally, the Panel discussed recent studies that address this
endpoint in humans (from Japan [reference not provided] and Holland [e.g., Koopman-Esseboom
et al., 1996; Vreugdenhil et al., 2002]). For continued investigation into this endpoint, the Panel
raised two issues to the U.S. EPA:
• Conduct an evaluation of whether a modeled effect can be attributed to TCDD and not
some other persistent organic pollutant (POP), although the Panel recognized that it is
unlikely U.S. EPA will be able to distinguish among these exposures because other POPs
are intrinsic confounders in the Dutch study.
• Allow animal data to inform the dose-response modeling of epidemiological data.
Thyroid Dysfunction
The Panel identified the availability of human data for this endpoint (e.g., Calvert et al.,
1999; Koopman-Esseboom et al., 1994). Much of the thyroid dysfunction literature has been
published since the 2003 Reassessment (e.g., Wang et al., 2005; Baccarelli et al., 2008). The
Panel also noted the availability of an internal dose metric (serum concentrations). Additionally,
the Panel discussed the mechanistic studies in animals that link TCDD to thyroid dysfunction.
For continued investigation into this endpoint, the Panel raised three issues for the U.S. EPA to
consider:
• Consider the newly available human data since the Reassessment.
• Investigate and clarify of the role of TCDD-induced thyroid dysfunction in
developmental neurotoxicity.
• Evaluate and determine whether an effect can be attributed to TCDD or other
contaminants.
Diabetes
The Panel discussed that data suggest that diabetes incidence in those under 55 years old
may be associated with exposure to PCBs. They acknowledged that whether this is a dioxin-like
compound (DLC) mediated effect or whether other POPs are responsible is still undetermined.
The Panel also acknowledged that no animal model exists for the investigation of xenobiotic-
induced diabetes, and that separating the injury dose level from the current body burdens would
depend on good pharmacokinetics in humans. For continued investigation into this endpoint, the
Panel listed two issues for the U.S. EPA to consider:
• Results from the Anniston study and the Great Lakes Fishermen study (references not
provided) should be examined for dose metrics (both studies examine human PCB
exposures).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-9 DRAFT—DO NOT CITE OR QUOTE
-------
• Changes of adipose tissue status need to be considered, given that dieting can cause
release of lipid-soluble contaminants.
References
Baccarelli, A., S.M. Giacomini, C. Corbetta et al. 2008. Neonatal thyroid function in Seveso 25
years after maternal exposure to dDioxin. PLoS Med. 5(7):el61.
doi: 10.1371/journal.pmed. 0050161.
Calvert, G.M., M.H. Sweeney, J. Deddens and D.K. Wall. 1999. Evaluation of diabetes
mellitus, serum glucose, and thyroid function among United States workers exposed to
2,3,7,8-tetrachlorodibenzo-p-dioxin. Occ. Env. Med. 56:270-276.
Huisman, M., C. Koopman-Esseboom, V. Fidler et al. 1995. Perinatal exposure to
polychlorinated biphenyls and dioxins and its effect on neonatal neurological development.
Early Hum. Devel. 41(2): 111-127.
Jacobson, J.L. and S.W. Jacobson. 1996. Intellectual impairment in children exposed to
polychlorinated biphenyls in utero. N. Engl. J. Med. 335:783-789.
Koopman-Esseboom, C., N. Weisglas-Kuperus, M.A.J. deRidder, C.G. Van derPaauw,
L.G.M.Th. Tuinstra and P.J.J. Sauer. 1996. Effects of polychlorinated biphenyl/dioxin exposure
and feeding type on infants' mental and psychomotor development. J. Pediatr. 97(5):700-706.
Koopman-Esseboom, C., D.-C. Morse, N. Weisglas-Kuperus et al. 1994. Effects of dioxins and
polychlorinated biphenyls on thyroid hormone status of pregnant women and their infants.
Pediatr. Res. 36:468-473.
Patandin, S., C.I. Lanting, P.G.H. Mulder, E.R. Boersma, P.J.J. Sauer and N. Weisglas-Kuperus.
1999. Effects of environmental exposure to polychlorinated biphenyls and dioxins on cognitive
abilities in Dutch children at 42 months of age. J. Pediatr. 134:33-41.
NAS (National Academy of Sciences). 2006. Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment. National Academies Press, Washington, DC
(July). Available at http://www.nap.edu/catalog.php7record id=l 1688.
U.S. EPA (U.S. Environmental Protection Agency). 2003. Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds. NAS
Review Draft (EPA/600/P-00/001Cb). U.S. Environmental Protection Agency, National Center
for Environmental Assessment, Washington, DC. Available at
http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
Vreugdenhil, H.J., C.I. Lanting, P.G. Mulder, E.R. Boersma and N. Weisglas-Kuperus. 2002.
Effects of prenatal PCB and dioxin background exposure on cognitive and motor abilities in
Dutch children at school age. J. Pediatr. 140:48-56.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-10 DRAFT—DO NOT CITE OR QUOTE
-------
Wang S.L., P.H. Su, S.B. Jong, Y.L. Guo, W.L. Chou and O. Papke. 2005. In utero exposure to
dioxins and polychlorinated biphenyls and its relations to thyroid function and growth hormone
in newborns. Environ. Health Perspect. 113:1645-1650.
SESSION 3B: DOSE-RESPONSE FOR CARDIOVASCULAR TOXICITY AND
HEPATOTOXICITY
The U.S. EPA plans to consider development of a quantitative dose-response assessment
for cardiovascular and/or hepatic effects associated with TCDD exposure. Such an assessment
would be based on information in U.S. EPA (2003), NAS (2006) and key studies identified in
this workshop. The purpose of this session was to identify and discuss key issues pertaining to
dose-response assessment for dioxin-induced cardiovascular and/or hepatic effects.
Session 3B Panelists (Session Co-Chairs are identified by asterisk)
• Bob Budinksy, Dow Chemical
• Manolis Kogevinas, Centre for Research in Environmental Epidemiology
• Rob McDowell, U.S. Department of Agriculture
• Jim Olson, State University of New York, University at Buffalo
• Marian Pavuk, Agency for Toxic Substances and Disease Registry
• *Jeff Swartout, U.S. EPA
• *Mary Walker, University of New Mexico
• Nigel Walker, National Institute of Environmental Health Sciences
Please note that the use of the term "concluded" or "recommended" in this summary does not mean that a consensus
was reached. Session Summaries were written from the material prepared by the non-EPA/ANL Co-chair and
represents a synopsis of the panel discussions.
Key Study Selection Criteria
The Panel initially focused on the draft key study selection criteria offered by the
U.S. EPA (Appendix C). The panel recommended that for cardiovascular effects, which are not
usually observed in rodents, the use of knockout mouse models (ApoE KO and LDLR KO) be
moved to the "primary" column because only these studies establish the cardiovascular toxicity
model in mice.
The panel also was concerned that the gavage procedure can increase mouse blood
pressure. Consequently, the panel recommended that gavage studies not be used for the blood
pressure endpoint (i.e., only dietary dosing studies should be considered).
Human Health Endpoints
In relation to the hepatic endpoint, the Panel acknowledged the large body of dose
response information on hepatic effects in rodents and that enzyme (mostly CYP1 Al) induction
was a sensitive effect. However, the Panel cited the lack of linkage of CYP1A1 to downstream
events, which complicates the toxicological interpretation of this endpoint, and concluded that
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
the more important liver effects in rodents are probably on the "road to cancer." The Panel noted
that hepatic effects were not seen in the epidemiological studies, but acknowledged that these
studies were not designed to detect them.
In relation to the cardiovascular endpoint, the Panel identified hypertension and ischemic
heart disease (IHD) as two key endpoints from the epidemiological studies. The Panel
recommended that the U.S. EPA perform a meta-analysis of these data. The Panel also
commented that recent animal studies support the observations linking TCDD exposure to IHD
and hypertension. In particular, the National Toxicology Program (NTP) study shows
inflammatory and structural effects on resistant vascular arterioles (NTP, 2006). Additional
evidence from the study suggests that the vascular effects may be CYP1A1-dependent. The
Panel suggested that the NTP study data might be used as a surrogate for dose-response
modeling of hypertension and that such an approach would be supported by data on the role of
AhR in vascular function and remodeling.
POD Issues
The Panel was not supportive of 1% of maximal response (ED0i), which was utilized in
the 2003 Reassessment. The Panel concluded that the POD should depend on the specific
endpoint and recommended the following to the U.S. EPA:
• For continuous measures, base the BMR on difference from control. Consider the
adversity level—at what point does the endpoint become adverse?
• For incidence data, set the BMR to a fixed-risk level.
Supporting Information
The Panel posed several suggestions to the U.S. EPA for reducing uncertainty and
improving the knowledge base for TCDD toxicity.
• Use in vitro data to define uncertainties, such as the relative sensitivity between rodents
and humans and around the definition of a POD.
• Consider studies on dioxin-like compounds (DLCs).
• Use PK modeling to define the dose metric for hepatic effects.
• Use body burden or serum concentrations for cardiovascular endpoints.
Finally, the Panel recommended that U.S. EPA finish the reassessment quickly and establish a
definitive plan to review and incorporate new data as they become available.
References
NAS (National Academy of Sciences). 2006. Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment. National Academies Press, Washington, DC
(July). Available at http://www.nap.edu/catalog.php7record id=l 1688.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-12 DRAFT—DO NOT CITE OR QUOTE
-------
NTP (National Toxicology Program). 2006. Toxicology and Carcinogenesis Studies of
2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) (CAS No. 1746-01-6) in Female Harlan Sprague-
Dawley Rats (Gavage Studies). U.S. Department of Health and Human Services. NTP TR 521.
Research Triangle Park, NC (April).
U.S. EPA (U.S. Environmental Protection Agency). 2003. Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds. NAS
Review Draft (EPA/600/P-00/001Cb). U.S. Environmental Protection Agency, National Center
for Environmental Assessment, Washington, DC. Available at
http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
SESSION 4A: DOSE-RESPONSE FOR CANCER
The U.S. EPA plans to consider development of a quantitative dose-response assessment
for cancer associated with TCDD exposure. Such an assessment would be based on information
in U.S. EPA (2003), NAS (2006) and key studies identified in this workshop. The purpose of
this session was to identify and discuss key issues pertaining to dose-response assessment for
dioxin-induced cancer.
Session 4A Panelists (Session Co-Chairs are identified by asterisk)
• Lesa Aylward, Summit Toxicology
• Kenny Crump, Louisiana Tech University
• Dale Hattis, Clark University
• * Janet Hess-Wilson, U.S. EPA
• Karen Hogan, U.S. EPA
• Manolis Kogevinas, Centre for Research in Environmental Epidemiology
• Marian Pavuk, Agency for Toxic Substances and Disease Registry
• Chris Portier, National Institute of Environmental Health Sciences
• Lorenz Rhomberg, Gradient
• Jay Silkworth, General Electric
• *Nigel Walker, National Institute of Environmental Health Sciences
Please note that the use of the term "concluded" or "recommended" in this summary does not mean that a consensus
was reached. Session Summaries were written from the material prepared by the non-EPA/ANL Co-chair and
represent a synopsis of the panel discussions.
Key Study Selection
The Panel discussed both human and rodent studies. In reviewing the epidemiological
data, the Panel agreed the EPA should focus on four cohort studies (Dutch cohort, NIOSH
cohort, BASF accident cohort, and Hamburg cohort) and pointed out that there are numerous
updates and reevaluations of data now in the literature and others will be published soon. The
Panel stated that it is appropriate for the U.S. EPA to consider the increase in total cancers for
modeling human cancer data, however, Non-Hodgkin's lymphoma, and lung tumors are the main
TCDD-related cancer types seen in humans exposed to TCDD. The Panel suggested the U.S.
EPA focus the quantitative dose-response modeling on the human data.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-13 DRAFT—DO NOT CITE OR QUOTE
-------
In reviewing the rat data, the Panel identified four new NTP rodent cancer bioassays with
liver and lungs as the main target organs. However, they suggested that dose-response modeling
efforts should model "all cancers" from these NTP data sets as well and use tumor incidence—
not individual rats as measures.
Key Study Selection Criteria
The Panel discussed whether data for TCDD only should be used or if PCB126 could be
used to develop a dose-response curve. From this discussion, the Panel reached a general
agreement that limiting the dose-response modeling and cancer assessment to TCDD only would
be the best approach.
Regarding the oral dosing regimens, the Panel discussed the differences in results from
different bioassays. They concluded that there were insufficient data to pick between oral feed
(Kociba et al., 1978) and oral gavage (NTP, 2006) studies, but stated "If all aspects of studies
were equal, an oral feed study is preferred." However, given that current data sets are not equal,
they agreed that U.S. EPA should consider both feed and gavage studies.
The Panel put forth the recommendation that studies that include initiation-promotion
model data and TgAC transgenic model data from oral exposure studies should be excluded from
the primary category in the key study selection criteria (Appendix C lists the draft study selection
criteria distributed prior to the meeting). Studies from both classifications should be moved to
the second tier.
The Panel was also unsupportive of the "response magnitude outside the range of normal
variability" criterion, as they did not believe it was applicable to a cancer endpoint.
Critical Endpoints to Consider
The Panel recognized that the MOA for TCDD includes cell growth/differentiation
dysregulation, that different endpoints (tumor types) across species may be expected, and that
there are differences in tumor sites across species. The Panel further acknowledged that there is
insufficient information to determine if rodent tumor types observed are relevant to humans.
Thus, the Panel suggests the following:
• U.S. EPA should consider all the observed cancer endpoints in its evaluation.
Nonlinear (aka threshold) Versus Linear Dose-Response Modeling
The Panel agreed that NTP bioassays appear to demonstrate nonlinear dose response, but
they expressed concern about using animal data to infer slope and dose response for humans.
The Panel pointed out that there are differences in slopes across different bioassays, and
specifically, that some appear linear while others appear nonlinear. Given the observation of
both nonlinear vs. linear, the Panel concluded that neither could be ruled out for extrapolation
below the POD simply based on the available data. One panelist noted that U.S. EPA Cancer
Guidelines (U.S. EPA, 2005) state that only if one can demonstrate that the MOA has a threshold
dose-response shape, and can exclude all other potential linear MO As, can one use a nonlinear
model. Lastly, the Panel noted that there are data and rationales to support use of both linear and
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-14 DRAFT—DO NOT CITE OR QUOTE
-------
nonlinear response below POD. From this discussion, the Panel raised one possibility to the U.S.
EPA:
• Both linear and nonlinear model functions should be considered in the dose-response
analysis.
Dose Metrics
In considering human data, the Panel expressed a preference for lipid-adjusted serum
levels over body burden (BB), and they expressed concerns over the assumptions used in the
back calculation of the BB in the epidemiologic cohorts. In considering the rat data, the Panel
supported the use of BB—especially lipid-adjusted BB. The Panel, however, did express
concern over the sequestering of TCDD in liver and then the use of liver levels in BB
calculations.
Supporting Information—Biologically-Based Dose-Response (BBDR) Models and MOA
The Panel discussed BBDR. Though once considered an attractive proposition, BBDR
models may mask uncertainty within the models, necessitating them to be used with greater
caution. The Panel suggested two issues for the U.S. EPA to consider:
• If there is a published model, use it if it is valid—do not generate a new model.
• Focus on the actual experimental data to drive the analysis.
References
Kociba, R.J., D.G. Keyes, J.E. Beyer et al. 1978. Results of a two-year chronic toxicity and
oncogenicity study of 2,3,7,8-tetrachlorodibenzo-p-dioxin in rats. Toxicol. Appl. Pharmacol.
46:279-303.
NAS (National Academy of Sciences). 2006. Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment. National Academies Press, Washington, DC
(July). Available at http://www.nap.edu/catalog.php7record id=l 1688.
NTP (National Toxicology Program). 2006. Toxicology and Carcinogenesis Studies of
2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) (CAS No. 1746-01-6) in Female Harlan Sprague-
Dawley Rats (Gavage Studies). U.S. Department of Health and Human Services. NTP TR 521.
Research Triangle Park, NC (April).
U.S. EPA (U.S. Environmental Protection Agency). 2003. Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds. NAS
Review Draft (EPA/600/P-00/001Cb). U.S. Environmental Protection Agency, National Center
for Environmental Assessment, Washington, DC. Available at
http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
U.S. EPA (U.S. Environmental Protection Agency). 2005. Guidelines for Carcinogen Risk
Assessment. U.S. Environmental Protection Agency Risk Assessment Forum.
EPA/630/P-03/001F.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-15 DRAFT—DO NOT CITE OR QUOTE
-------
SESSION 4B: DOSE-RESPONSE FOR REPRODUCTIVE/DEVELOPMENTAL
TOXICITY
The U.S. EPA plans to consider development of a quantitative dose-response assessment
for reproductive and developmental effects associated with TCDD exposure. Such an
assessment would be based on information in U.S. EPA (2003), NAS (2006) and key studies
identified in this workshop. The purpose of this session was to identify and discuss key issues
pertaining to dose-response assessment for dioxin-induced reproductive and developmental
effects.
Session 4B Panelists (Session Co-Chairs are identified by asterisk)
• Barbara Abbott, U.S. EPA
• Bruce Allen, Bruce Allen Consulting
• Roger Cooke, Resources for the Future
• George Daston, Procter & Gamble
• MikeDeVito, U.S. EPA
• Rob Goble, Clark University
• *Fumio Matsumura, University of California-Davis
• Paolo Mocarelli, University of Milan
• Brian Petroff, University of Kansas
• *Glenn Rice, U.S. EPA
• Marie Sweeney, National Institute of Occupational Safety and Health
• Mary Walker, University of New Mexico
• Bernie Weiss, University of Rochester
Please note that the use of the term "concluded" or "recommended" in this summary does not mean that a consensus
was reached. Session Summaries were written from the material prepared by the non-EPA/ANL Co-Chair and
represent a synopsis of the panel discussions.
A Major Question Posed During this Workshop Session was "Are Human Embryos and
Infants Less Sensitive to Dioxin Exposures Than Some Experimental Animals?"
The Panel recognized that animal data show a wide range of species sensitivity to dioxin
for a given developmental or reproductive endpoint. Presently, there are data for some endpoints
that show that human sensitivity is comparable to experimental animals (e.g., semen quality),
and for other endpoints the data demonstrate that humans are insensitive compared to other
species (e.g., cleft palate). Lastly, the Panel recognized that there are some endpoints for which
relative human sensitivity remains uncertain.
Key Study Selection
The Panel reviewed the charge questions (Appendix B), discussed them, and listed two
issues for the U.S. EPA to consider:
• Concerning key study determination, use a stepwise approach that is dependent upon the
information available and needed to address the question.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-16 DRAFT—DO NOT CITE OR QUOTE
-------
• Concerning the key studies informing the POD and the POD endpoint choice, use the
POD to depart from what is certain and use a high-confidence study that has found
effects at a low enough level at which other effects are protected.
The Panel also developed Table 1, based on the information presented in this session. Table 1
identifies specific reproductive and developmental effects of concern, listing whether an effect
has been observed in test animals and epidemiologic cohorts. It also identifies the EDi0
estimated by the U.S. EPA (2003) for health effects observed in rodent bioassays. If the U.S.
EPA did not report an EDi0 for an effect, the table identifies a study where the effect was
reported and the lowest study dose where the effect was observed. Table 1 also identifies the
epidemiologic cohort where the specific reproductive and developmental effects were observed.
Epidemiological Study Utility
The Panel reviewed the charge questions (Appendix B), discussed them, and made two
suggestions to the U.S. EPA:
• Concerning the ability of epidemiological studies to inform critical effects, start with
concordance across species (including humans) for the spectrum of effects.
• Concerning the ability of epidemiological studies to inform dose-response modeling, start
with the epidemiology and then go to animal data if the dose response has not been well
characterized for an endpoint of interest and compare to animal data as a reality check.
Animal Model Utility
The Panel reviewed and discussed the charge questions (Appendix B). Table 1, which
identifies the effects that occur in animals and also have relevance to humans, summarizes much
of this discussion. Regarding the influence of mode of action (MO A) on animal model choice,
the Panel concluded that by evaluating concordance among health effects reported in
epidemiologic and animal bioassay data, the U.S. EPA could identify a set of plausible
reproductive and developmental effects to consider. Actual animal and human MOA
information is helpful in that it creates comfort with the animal models and in defining the
boundaries of possible effects.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-17 DRAFT—DO NOT CITE OR QUOTE
-------
TABLE 1
Reproductive/Developmental Effects of Concern for Human Health
Endpoint
Rodent
(ED 10 ng/kg-d)
Human
Notes
Sperm Count/Motility
Yes (6.2-28;
66-200)
Yes
EDio bases Mabley et al. (1992a,b) caudal
sperm count and daily sperm production
range from 6.2-28; Gray et al. (1997)
epididymal sperm count and total testis sperm
counts range from 66-200.
Sex Ratio
No
Yes, Seveso
Delayed Puberty Males
Yes (94)
Yu-cheng
EDio basis rat male puberty delay Gray et al.
(1997). Need to qualify epidemiology data
because of cohort PCDD/PCDFs exposures.
Delayed Puberty in Females
Yes
No in Seveso
Gray and Ostby (2002) report delayed
puberty in female offspring of pregnant rats
receiving a single dose of 1 |ig TCDD/kg on
GD 15.
Cleft Palate
Yes(6300-6400)
No
EDio basis Birnbaum et al. (1989).
Premature Senescence
Yes
No, Seveso
Franczak et al. (2006) report that rats
prematurely entered reproductive senescence,
after receiving cumulative TCDD doses as
low as 1.7 jig TCDD/kg. They considered
first occurrence of prolonged interestrous
interval (>6 d) as evidence of onset of
reproductive senescence.
Hormones E2
Yes
Yes, Males—
Seveso
Li et al. (1995) report serum estradiol-17f^
(E2) concentrations induced by equine
Chorionic Gonadotropin injection were
significantly elevated in female rats orally
administered 10 (ig/kg TCDD onPND 22.
While E2 decreased dramatically in control
animals during the preovulatory LH surge, it
did not in TCDD-treated rats.
Low Birth Weight
Yes (190)
Suggestive
effect in Seveso
in first 8 years
after exposure
EDio basis Gray et al. (1997).
Reproductive Cycling
(prolongation)
Yes
Yes, Seveso
Prepubertal
exposure
Franczak et al. (2006) report loss of normal
cyclicity in female rats at 8 months of age
following a cumulative dose of 1.7 jig
TCDD/kg.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-18 DRAFT—DO NOT CITE OR QUOTE
-------
Supporting Information
The Panel reviewed the charge questions (Appendix B), discussed them, and made two
suggestions to the U.S. EPA:
• Concerning deviation from default approaches for noncancer endpoints, there needs to be
a careful assessment of the POD and the application of uncertainty factors in light of
PK/pharmacodynamics (PD), population characteristics and variability, and MOA
information.
• Concerning the MOA's ability to clarify endpoint and the incorporation of a cascade of
cellular event into dose-response for noncancer endpoint, any study that helps inform the
dose response should be considered—including studies not specific to dioxins.
Complicated mechanistic models need not be developed. Standard dose-response models
can be applied. One can look at the cascade of events in a stepwise, simple way.
References
Birnbaum, L.S., M.W. Harris, L.M. Stocking et al. 1989. Retinoic acid and 2,3,7,8-
tetrachlorodibenzo-p-dioxin selectively enhance teratogenesis in C57BL/6N mice. Toxicol.
Appl. Pharmacol. 98:487-500.
Franczak, A., A. Nynca, K.E. Valdez, K.M. Mizinga and B.K. Petroff 2006. Effects of acute
and chronic exposure to the aryl hydrocarbon receptor agonist 2,3,7,8-tetrachlorodibenzo-
p-dioxin on the transition to reproductive senescence in female Sprague-Dawley rats. Biol.
Reprod. 74:125-130.
Gray, L.E. and J.S. Ostby. 2002. In utero 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) alters
reproductive morphology and function in female rat offspring. Toxicol. Appl. Pharmacol.
133(2):285-294.
Gray, L.E., J.S. Ostby and W.R. Kelce. 1997. A dose-response analysis of the reproductive
effects of a single gestational dose of 2,3,7,8-tetrachlorodibenzo-p-dioxin in male Long Evans
Hooded rat offspring. Toxicol. Appl. Pharmacol. 146:11-20.
Li, X., D.C. Johnson and K.K. Rozman. 1995. Reproductive effects of 2,3,7,8-
tetrachlorodibenzo-p-dioxin (TCDD) in female rats: ovulation, hormonal regulation, and possible
mechanism(s). Toxicol. Appl. Pharmacol. 133:321-327.
Mably, T.A., D.L. Bjerke, R.W. Moore et al. 1992a. In utero and lactational exposure of male
rats to 2,3,7,8-tetrachlorodibenzo-p-dioxin. 3. Effects on spermatogenesis and reproductive
capability. Toxicol. Appl. Pharmacol. 114:118-126.
Mably, T.A., R.W. Moore, R.W. Goy et al. 1992b. In utero and lactational exposure of male
rats to 2,3,7,8-tetrachlorodibenzo-p-dioxin. 2. Effects on sexual behavior and the regulation of
luteinizing hormone secretion in adulthood. Toxicol. Appl. Pharmacol. 114:108-117.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-19 DRAFT—DO NOT CITE OR QUOTE
-------
NAS (National Academy of Sciences). 2006. Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment. National Academies Press, Washington, DC
(July). Available at http://www.nap.edu/catalog.php7record id=l 1688.
U.S. EPA (U.S. Environmental Protection Agency). 2003. Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds. NAS
Review Draft (EPA/600/P-00/001Cb). U.S. Environmental Protection Agency, National Center
for Environmental Assessment, Washington, DC. Available at
http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
SESSION 5: QUANTITATIVE UNCERTAINTY ANALYSIS OF DOSE-RESPONSE
This session addressed the uncertainty analysis to be considered for the dose-response
assessments. The session opened with a presentation on current estimates of dioxin exposure
levels. Then it focused on the factors to include in the scope of an uncertainty analysis including
dioxin kinetics.
Session 5 Panelists (Session Co-Chairs are identified by asterisk)
• Bruce Allen, Bruce Allen Consulting
• Lesa Aylward, Summit Toxicology
• Roger Cooke, Resources for the Future
• Kenny Crump, Louisiana Tech University
• MikeDeVito, U.S. EPA
• Dale Hattis, Clark University
• *Rick Hertzberg, Biomath Consulting
• Nancy Kerkvliet, Oregon State University
• Leonid Kopylev, U.S. EPA
• Rob McDowell, U.S. Department of Agriculture
• Lorenz Rhomberg, Gradient
• Woody Setzer, U.S. EPA
• Marie Sweeney, National Institute of Occupational Safety and Health
• *Linda Teuschler, U.S. EPA
Please note that the use of the term "concluded" or "recommended" in this summary does not mean that a consensus
was reached. Session Summaries were written from the material prepared by the non-EPA/ANL Co-Chair and
represent a synopsis of the panel discussions.
The Panel summarized the NAS comments regarding uncertainty. Areas for improvement
include:
• Ensure "transparency, thoroughness, and clarity in quantitative uncertainty analysis."
• Describe and define (quantitatively to the extent possible) the variability and uncertainty
for key assumptions used for each key endpoint-specific risk assessment, including
choices of data set, point of departure, dose-response model, and dose metric.
• Incorporate probabilistic models to represent the range of plausible values.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-20 DRAFT—DO NOT CITE OR QUOTE
-------
• Assess goodness-of-fit of dose-response models.
• Provide upper and lower bounds on central tendency estimates for all statistical estimates.
• When quantification is not possible, clearly state it, and explain what would be required
to achieve quantification.
Identification of Important Uncertainties
The Panel reviewed the charge questions (Appendix B), discussed them, and listed eight
issues for consideration by the U.S. EPA:
• Concerning species and strain differences in the U.S. EPA's Response to NAS, current
U.S. EPA procedures do not take this into account when selecting one data set for risk
assessment. Issues include "Where are humans in the distribution of potencies that can
be generated? How likely is it that human response is similar to the selected data? Can
we infer inter-individual variability from these differences?"
• Concerning the use of animal data for cross species extrapolation to humans (PK and PD
uncertainties), issues to consider include differences in distribution and responses
following bolus doses from those of subchronic and chronic protocols; uncertainty in
liver doses due to sequestration; differences in receptor binding affinity among
congeners; and age factors (e.g., assumption of a lifetime constant daily dose for a cancer
extrapolation).
• Concerning the description of AhR response, biochemical changes occur at lower doses
than toxicological changes. There should be an effort to identify the biochemical changes
that would mark Ah receptor binding to inform the BMR, and, thus, prevent toxicity.
• Concerning model uncertainty, the mathematical model choice depends on endpoint.
There should be an effort towards determining what is the most sensitive endpoint(s) for
humans and conducting animal studies to model that endpoint(s).
• Concerning exposure and dose response in human studies, ensure enough similarity to
current human exposure profiles (mixture composition) so that a dose-response
assessment can be done. Incorporate new epidemiological studies. Evaluate
concordance with animal data and consistency across studies. Panel-acknowledged
uncertainties include exposure estimates from person to person, shape of human dose-
response curve, healthy worker effect, and age dependence.
• Concerning POD determination, uncertainty factors are inherently mathematically
inconsistent and that should be conveyed in the discussion of uncertainties when
interpreting the POD.
• Concerning dose metric, tissue concentration is preferred. It should be evaluated against
a background of variability in AhR-binding expression. There is uncertainty in what
level of binding should be considered, in different cell types, tissues, life stage
(development). The relationship between dose metric and causation of adverse effects
should be examined.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-21 DRAFT—DO NOT CITE OR QUOTE
-------
Low-Dose Extrapolation
The Panel reviewed the charge questions and discussed them (Appendix B). The Panel
concluded that curve-fitting uncertainty (for a given dataset, dose metric, and model) can be
characterized and is useful, but, by itself, it is an incomplete characterization of uncertainty. The
Panel acknowledged the difficulty of fully characterizing uncertainty, especially quantitatively.
Some panelists argued that the problem is insurmountable and that no meaningful uncertainty
analysis is likely to be performable. Other panelists contended that, the difficulties
notwithstanding, "good-faith" efforts to do something practical and forthright to characterize
uncertainty in low-dose extrapolation would be useful and important. The Panel clarified "good
faith" as meaning a characterization that is useful and not misleading to decision makers and is
inclusive of approaches that have meaningful support in the scientific community as a whole.
Being in "good faith" is more important than being complete (i.e., addressing every uncertain
element), especially since completeness is not a realistic goal. From this discussion, the Panel
listed four issues for consideration by the U.S. EPA:
• Review alternative data sets, dose metrics, and models to see where consequential
uncertainties and impacts on low-dose implications arise.
• Consider the impacts of choices among plausible alternative data sets, dose metrics,
models, and other more qualitative choices—issues include how much difference the
choices make and also how much relative credence should be put to each alternative as a
way of gauging and describing the landscape of imperfect knowledge
regarding possibilities for the true dose-response.
¦ Hard to do quantitatively, since the factors are not readily expressed as statistical
distributions, but can describe the rationale for believing/doubting each alternative in
terms of available supporting evidence, contrary evidence, and needed assumptions.
¦ Expert judgment methods may be helpful in characterizing the relative weights of
scientific credibility among alternatives. The expert judgment process, when
conducted systematically, can be thought of as adding data to the assessment of
credibility of alternatives, rather than as just an opinion poll.
¦ Information on plausibility of alternative low-dose extrapolation approaches can
come from external considerations of mode of action, and not just from statistical
success at fitting particular (high-dose) data sets.
• Characterizing uncertainty through a variety of approaches could be tried, and their
relative merits and shortcomings discussed, as a way forward.
• Consider the sources of potential error, particularly in epidemiological data (e.g., TEF
uncertainty and variation in congener mixtures) and if possible quantify their impact on
the dose-response assessment.
Considerations for Conducting Uncertainty Analysis
Overall, the Panel was split on whether U.S. EPA should do quantitative uncertainty
analyses. The Panel noted that if done on only some of the uncertainties, then results would be
misleading and could be misused. Ultimately, the Panel listed seven issues for consideration by
the U.S. EPA:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-22 DRAFT—DO NOT CITE OR QUOTE
-------
• The Panel recapped what some consider as being the first integrated risk assessment, with
structured expert judgment and uncertainty analysis, i.e., the Rasmussen Report
(WASH-1400; U.S. Nuclear Regulatory Commission, 1975). In their discussion of the
report, the Panel noted that in addition to standard event tree/fault tree modeling, this
report also tackled difficult model uncertainty issues involved in accident progression,
dispersion of released pollutants in the atmosphere, environmental transport, exposure,
health, and economic impacts. And though the Panel also recognized that this method
was no longer state-of-the-art, the Panel contended that it represents a good example of a
structured approach and methodology that could be built upon.
• The Panel also discussed TEQs used in epidemiological studies, based on intake, and
recognized that the key uncertainty in what was measured was not just intake but also
involved PK/PD issues. The Panel acknowledged that the TEQ system is regularly used
on a concentration basis, but they expressed concern that the qualification becomes lost.
TEQs ignore pharmacokinetics and the common practice of rounding to orders of
magnitude introduces more error.
• Structure the risk assessment along MOA steps—identify key biochemical measures
(-5-10) common across toxic endpoints and identify the degree of meaningful change in
effect or effect variance. Make a table with all options for data set, model, etc.; make
best estimates/choices and determine which of these choices matter the most to the
answer.
• Use expert panels—expert judgment can be collected scientifically (procedures are
published). But there are known biases; central tendency estimates work much better
than extremes.
• Use supporting studies to fill in critical data gaps—Info filling methods do exist (e.g., PK
modeling). Put short-term studies into the "supporting info" category (unless, of course,
the risk assessment is for acute exposures, such as chemical spills).
• Be creative in the analysis of uncertainty. Intermediate steps between AhR binding and
the end processes can be hypothesized based on data, experiences, and analogies related
to other chemicals.
• The 2003 Reassessment presented potency estimates on wide variety of
endpoints/models; needed to be more transparent in that discussion. Statistical graphics
can be used to convey uncertainties.
Reference
U.S. Nuclear Regulatory Commission. 1975. Reactor Safety Study: An Assessment of Accident
Risks in U.S. Commercial Nuclear Power Plants. WASH-1400 (NUREG-75-014). Washington,
DC.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-23 DRAFT—DO NOT CITE OR QUOTE
-------
APPENDIX A: 2009 U.S. EPA DIOXIN WORKSHOP AGENDA
SCIENTIFIC WORKSHOP
TO INFORM THE TECHNICAL WORK PLAN FOR U.S. EPA'S RESPONSE TO
NAS COMMENTS ON THE HEALTH EFFECTS OF DIOXIN
PRESENTED IN U.S. EPA'S DIOXIN REASSESSMENT
Cincinnati, OH
Date: February 18-20, 2009
BACKGROUND/WORKSHOP OBJECTIVE
At the request of the U.S. Environmental Protection Agency (U.S. EPA), the National
Academy of Sciences (NAS) prepared a report, Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment (NAS, 2006), that made a number of
recommendations to improve the U.S. EPA's risk assessment for 2,3,7,8-tetrachlorodibenzo-
/;-dioxin (TCDD). In response, the U.S. EPA will prepare a technical report that addresses key
comments on the dose-response assessment for TCDD. The U.S. EPA intends to develop its
response through a transparent process that provides multiple opportunities for input.
To assist in this effort, a Workshop will be held to inform the U.S. EPA's evaluation of
the NAS recommendations. The Workshop will be open to the public. At the Workshop, the
U.S. EPA will solicit input from expert scientists and the public.
The goal of the Workshop is to ensure that the U.S. EPA's response to the NAS
comments focuses on the key issues and reflects the most meaningful science. The three main
objectives of the Workshop are to (1) identify and discuss the technical challenges involved in
addressing the NAS key comments on the TCDD dose-response assessment in the U.S. EPA
Reassessment (U.S. EPA, 2003), (2) discuss approaches for addressing these comments, and
(3) identify key published, independently peer-reviewed literature, particularly studies describing
epidemiologic and in vivo mammalian bioassays, which are expected to be most useful for
informing the U.S. EPA response.
Workshop participants will be encouraged to think broadly about the body of scientific
information that can be used to inform the U.S. EPA's response and to participate in open
dialogue regarding ways in which the science can best be used to address the key dose-response
issues. This Workshop is similar to scientific workshops being conducted under the new review
process for the National Ambient Air Quality Standards (NAAQS)1 that assess health-related
information for criteria pollutants.
1 Please see http://www.epa.gov/ttn/naaas/ for more information on the new NAAQS review process.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-24 DRAFT—DO NOT CITE OR QUOTE
-------
The Workshop discussions are expected to build upon two prior publications:
1. Exposure and Human Health Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin
(TCDD) and Related Compounds (U.S. EPA, 2003). This external review draft
provides a comprehensive reassessment of dioxin exposure and human health effects.
This "dioxin reassessment" was submitted in October 2004 to the National Academy
of Sciences (NAS) for review.
2. Health Risks from Dioxin and Related Compounds: Evaluation of the EPA
Reassessment (NAS, 2006).
Workshop participants are encouraged to review both of these documents and other
relevant materials (e.g., the National Toxicology Program report on TCDD [NTP, 2006]) before
the meeting because they provide important insights into the key questions and challenges.
There are a number of open comment periods that are intended to facilitate a broad discussion of
the issues.
Scientists with significant expertise and experience relevant to the health effects of
TCDD or dioxin-like compounds and associated topics will be asked to serve on "expert panels"
for discussions throughout the Workshop. Workshop panelists will include a wide range of
experts representing many scientific areas needed to assess TCDD dose-response (e.g.,
epidemiology, human and animal toxicology, nuclear receptor biology, dose-response modeling,
risk assessment, and uncertainty analysis). The Workshop panelists will be asked to highlight
significant and emerging research and to make recommendations to the U.S. EPA regarding the
design and scope of the technical response to NAS comments on the dose-response analysis for
TCDD—including, but not limited to, recommendations for evaluating associated uncertainty.
Open comment periods will follow each panel discussion session. Public participation will be
encouraged by way of these designated open comment periods and, also, by participation in the
scientific poster session planned for the second evening (February 19).
U.S. EPA will use the input received during this Workshop as the foundation for its
development of a technical work plan for responding to the NAS comments on the TCDD dose-
response analysis. The work plan will outline the schedule, process, and approaches for
evaluating the relevant scientific information and addressing the key issues. The work plan also
will identify the key literature to be utilized in U.S. EPA's response.
As a follow-on activity to this Workshop, a panel is being established under the Federal
Advisory Committee Act (FACA) to guide and review the U.S. EPA's response to NAS
comments. The FACA panel will be asked to conduct a consultation with the Agency on the
draft technical work plan. At the same time, the public will also have the opportunity to provide
comments to the FACA panel on the work plan. The final technical work plan will guide the
development of the technical report that will constitute the U.S. EPA's response to NAS
comments. During the development of this response, the U.S. EPA will seek advice from the
FACA panel and the public several times. Finally, the FACA panel will be asked to review the
technical report in a public forum.
The preliminary Agenda presented on the following pages may be revised prior to the
Workshop following review by the session Co-Chairs; the dates and general timing of the
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-25 DRAFT—DO NOT CITE OR QUOTE
-------
sessions, however, will not change. A final Agenda and a set of charge questions, intended to
provide general direction for the Workshop discussions, will be posted on the Workshop Internet
site (http://cfpub.epa.gov/ncea/cfrn/recordisplav.cfm?deid=199923) prior to the meeting.
A poster session will be held on the evening of the second day (February 19). The
purpose of this poster session is to provide a forum for scientists to present recent studies
relevant to TCDD dose-response assessment and to encourage open discussion about these
presentations.
REFERENCES
NAS (National Academy of Sciences). 2006. Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment. National Academies Press, Washington, DC
(July). Available at http://www.nap.edu/catalog.php7record id=l 1688.
NTP (National Toxicology Program). 2006. Toxicology and Carcinogenesis Studies of
2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) (CAS No. 1746-01-6) in Female Harlan Sprague-
Dawley Rats (Gavage Studies). U.S. Department of Health and Human Services. NTP TR 521.
Research Triangle Park, NC (April).
U.S. EPA (U.S. Environmental Protection Agency). 2003. Exposure and Human Health
Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds, NAS
review draft, Volumes 1-3 (EPA/600/P-00/001Cb, Volume 1). U.S. Environmental Protection
Agency, National Center for Environmental Assessment, Washington, DC (December).
Available at http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-26 DRAFT—DO NOT CITE OR QUOTE
-------
WORKSHOP AGENDA
Day 1
8:00-9:00 Registration
9:00-9:30 Welcome/Purpose of Meeting/Document Development Process
9:30-9:45 Panel Comments/Questions on Charge
9:45-2:45 Session 1: Quantitative Dose-Response Modeling Issues
(Hall of Mirrors)
9:45-10:10 Background/Introductory Remarks
10:10-10:35 TCDD Kinetics: Converting Administered Doses in Animals to
Human Body Burdens
Presenter: Michael Devito
10:35-11:30 Panel Discussion
11:30-1:00 Lunch
1:00-2:00 Panel Discussion cont.
2:00-2:45 Open Comment Period
2:45-3:05 Break
3:05-5:15 Session 2: Immunotoxicitv (Hall of Mirrors)
3:05-3:15 Background/Introductory Remarks
3:15-4:45 Panel Discussion
4:45-5:15 Open Comment Period
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-27 DRAFT—DO NOT CITE OR QUOTE
-------
Day 2
8:00-8:30 Report-Outs for Sessions 1 and 2 (Hall of Mirrors)
8:00-8:15 Report-Out for 1: Quantitative Dose-Response Modeling Issues
8:15-8:30 Report-Out for 2: Immunotoxicity
8:30-11:30 Sessions 3A and 3B (concurrent sessions)
8:30-11:30 Session 3A: Dose-Response for Neurotoxicity and
Nonreproductive Endocrine Effects (Hall of Mirrors)
8:30-8:45 Background/Introductory Remarks
8:45-11:00 Panel Discussion
11:00-11:30 Open Comment Period
8:30-11:30 Session 3B: Dose-Response for Cardiovascular Toxicity and
Hepatotoxicity (Rookwood Room)
8:30-8:45 Background/Introductory Remarks
8:45-11:00 Panel Discussion
11:00-11:30 Open Comment Period
11:30-1:00 Lunch
1:00-2:00 Report-Outs for Sessions 3A and 3B (Hall of Mirrors)
The structure of the session report-outs will include the following:
Summary of session presentation including minority opinion
¦ Public comments
¦ Discussion
1:00-l: 15 Report-Out for 3A: Dose-Response for Neurotoxicity and
Nonreproductive Endocrine Effects
1:15-1:30 Open Comment Period
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-28 DRAFT—DO NOT CITE OR QUOTE
-------
1:30—1:45 Report-Out for 3B: Dose-Response for Cardiovascular Toxicity and
Hepatotoxicity
1:45-2:00 Open Comment Period
2:00-5:15 Sessions 4A and 4B (concurrent sessions)
2:00-5:15 Session 4A: Dose-Response for Cancer (Hall of Mirrors)
2:00-2:15 Background/Introductory Remarks
2:15-4:45 Panel Discussion
4:45-5:15 Open Comment Period
2:00-5:15 Session 4B: Dose-Response for
Reproductive/Developmental Toxicity (Rookwood Room)
2:00-2:15 Background/Introductory Remarks
2:15-4:45 Panel Discussion
4:45-5:15 Open Comment Period
6:45-8:15 Poster Session (Rosewood Room)
Day 3
8:30-9:30 Report-Outs for Sessions 4A and 4B (Hall of Mirrors)
8:30-8:45 Report-Out for 4A: Dose-Response for Cancer
8:45-9:00 Open Comment Period
9:00-9:15 Report-Out for 4B: Dose-Response for Reproductive/Developmental
Toxicity
9:15-9:30 Open Comment Period
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-29 DRAFT—DO NOT CITE OR QUOTE
-------
9:30-3:30 Session 5: Quantitative Uncertainty Analysis of Dose-
Response (Hall of Mirrors)
9:30-9:40 Background/Introductory Remarks
9:40-10:10 Evidence of a Decline in Background Dioxin Exposures in Americans
Between the 1990s and 2000s
Presenter: Matt Lorber
10:10-10:30 Break
10:30-11:30 Panel Discussion
11:30-1:00 Lunch
1:00-2:15 Panel Discussion cont.
2:15-2:30 Break
2:30-3:00 Open Comment Period
3:00-3:15 Report-Out for 5: Quantitative Uncertainty Analysis of Dose-
Response
3:15-3:30 Closing Remarks
3:30 Adjourn
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-30 DRAFT—DO NOT CITE OR QUOTE
-------
APPENDIX B: 2009 U.S. EPA DIOXIN WORKSHOP
QUESTIONS TO GUIDE PANEL DISCUSSIONS
SESSION 1
Dose Metric
Considering all of the endpoints or target tissues, and species that U.S. Environmental Protection
Agency (U.S. EPA)'s dose-response modeling might evaluate, what are the best measures of
dose (e.g., ingested, tissue concentrations, body burden, receptor occupancy, other surrogate) and
why?
Developing Dose-Response Models from Mammalian Bioassays
How best can the point of departure (POD) be determined when the response range is
incompletely characterized (i.e., high response at the lowest dose or low response at the highest
dose; observed in several key 2,3,7,8-Tetrachlorodibenzo-p-Dioxin [TCDD] studies)?
If considered to be biologically plausible, how can a threshold be incorporated into a dose-
response function (e.g., for TCDD cancer data)?
How can nonmonotonic responses be incorporated into the dose-response function?
Developing Dose-Response Models from Epidemiological Studies
How can the epidemiological data be utilized best to inform the TCDD exposure-response
modeling? Which epidemiological studies are most relevant?
Supporting Information
For those toxicological endpoints that are Ah receptor-mediated, how would the receptor kinetics
influence the shape of the dose-response curve? How would downstream cellular events affect
the shape of the dose-response curve? How can this cascade of cellular events be incorporated
into a quantitative model of dose-response?
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-31 DRAFT—DO NOT CITE OR QUOTE
-------
SESSIONS 2, 3A, 3B, 4A, AND 4B
Key Study Selection
For this endpoint, what refinements should be made to the draft criteria for selection of key
studies?
What are the specific effects of concern for human health for this endpoint?
Based on the draft criteria for the selection of key studies, what are the key studies informing the
shape of the dose-response curve above the POD and the choice of the POD for this endpoint?
Epidemiological Study Utility
How and to what extent do the epidemiological data inform the choice of critical effect?
How can the epidemiological data inform the quantitative dose-response modeling?
Animal Model Utility
Are there types of effects observed in animal models that are more relevant to humans than
others? To what extent does information on mode of action (MOA) influence the choice of
animal model (species, strain, sex)?
Supporting Information
Are there studies that establish a sufficient justification for departure from the default procedures
that address the shape of the dose-response curve below the POD under the cancer guidelines?
Are there studies that establish a sufficient justification for departing from U.S. EPA's default
approaches for noncancer endpoints?
To what extent can MO A information clarify the identification of endpoints of concern and dose-
response metric for this endpoint? How can the cascade of cellular events for this endpoint be
incorporated into a quantitative model of dose response?
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-32 DRAFT—DO NOT CITE OR QUOTE
-------
SESSION 5
For cancer and noncancer TCDD dose-response assessments, U.S. EPA is interested in
developing a quantitative uncertainty analysis addressing both parameter and model uncertainty,
if feasible. Uncertainties will include, among others, choice of endpoint; underlying study
uncertainties; choice of dose metric; interspecies extrapolations such as kinetic uncertainties; and
choice of dose-response model, including threshold models. The U.S. EPA is currently
examining techniques and tools for uncertainty analysis—including Bayesian and frequentist
approaches.
Identification of Important Uncertainties
What are the major uncertainties pertaining to modeling the animal data?
Consider the dose metric (species or tissue specificity), vehicle of administration,
exposure frequency, exposure duration, and POD determination (e.g., benchmark
response selection or no-observed-adverse-effect level/lowest-observed-adverse-effect
level identification).
What are the major uncertainties pertaining to dose-response modeling below the POD?
Consider how receptor kinetics and downstream cellular event information might be used
to bound the uncertainties associated with dose-response modeling below the POD.
What are the major uncertainties in cross-species extrapolation (e.g., half-lives, tissue
distribution, and toxicodynamics)?
Consider the primary species dosed with TCDD: mice, hamsters, rats, guinea pigs, and
monkeys.
What are the major uncertainties pertaining to intrahuman variability?
Consider what data sets would be useful to represent sensitive subpopulations.
What are other significant sources of uncertainty for the cancer and noncancer assessments?
Considerations for Conducting Uncertainty Analysis
What data sets could be used to quantify uncertainties in cancer and noncancer TCDD dose-
response assessments?
Consider dioxin-like compound dose-response data.
Consider MOA information.
What are the appropriate techniques for the TCDD dose-response uncertainty analysis, and what
are their respective strengths and weaknesses of these approaches as applied to TCDD?
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 A-33 DRAFT—DO NOT CITE OR QUOTE
-------
' y
s?
Co'
§•
Cl
s;
3
s
>;*
>S
>S
>
0
3
1
§•
Co
a
o
o
o
a
Co
«Jk
s?
HH (/Q
H
-------
-------
DRAFT
DO NOT CITE OR QUOTE
January 2010
Agency/Interagency Review Draft
APPENDIX B
Evaluation of Cancer and Noncancer
Epidemiological Studies for Inclusion in
TCDD Dose-Response Assessment
NOTICE
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
CONTENTS—Appendix B: Evaluation of Cancer and Noncancer Epidemiological Studies
for Inclusion in TCDD Dose-Response Assessment
LIST OF TABLES B-iii
APPENDIX B. EVALUATION OF CANCER AND NONCANCER
EPIDEMIOLOGICAL STUDIES FOR INCLUSION IN
TCDD DOSE-RESPONSE ASSESSMENT B-l
B. 1. EVALUATION 01 CANCER STUDIES B-l
B.l.l. NIOSH Cohort Studies B-l
B.1.2. BASF Cohort Studies B-8
B.1.3. The Hamburg Cohort B-ll
B.1.4. The Seveso Cohort Studies B-16
B.1.5. The Chapaevsk Study B-22
B. 1.6. The Air Force Health ("Ranch Hands") Study B-23
B. 1.7. Other Studies of Potential Relevance to Dose-Response Modeling B-26
B.2. EVALUATION OF NONCANCER STUDIES B-31
B.2.1. NIOSH Cohort B-3 1
B.2.2. BASF Cohort B-33
B.2.3. Hamburg Cohort B-35
B.2.4. The Seveso Women's Health Study B-36
B.2.5. Other Seveso Noncancer Studies B-45
B.2.6. Chapaevsk Study B-54
B.2.7. Air Force Health ("Ranch Hands") Study B-55
B.2.8. Other Noncancer Studies of Dioxin B-56
B.3. REFERENCES B-60
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-ii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES
B-l. Fingerhut et al., 1991—All cancer sites, site-specific analysis B-l
B-2. Steenland et al., 1999—All cancer sites combined, site-specific analysis B-2
B-3. Steenland et al., 2001—All cancer sites combined B-4
B-4. Cheng et al., 2006—All cancer sites combined B-5
B-5. Collins et al., 2009—All cancer sites combined, site-specific analysis B-6
B-6. Zober et al., 1990—All cancer sites combined, site-specific analysis B-8
B-7. Ott and Zober, 1996—All cancer sites combined B-9
B-8. Manz et al., 1991—All cancer sites combined, site-specific analyses B-l 1
B-9. Flesch-Janys et al., 1995; Flesch-Janys et al., 1996 erratum—All cancer sites
combined B-12
B-10. Flesch-Janys et al., 1998—All cancer sites combined, site-specific analysis B-13
B-ll. Becher et al., 1998—All cancer sites combined B-14
B-12. Bertazzi et al., 2001—All cancer sites combined, site-specific analyses B-16
B-13. Pesatori et al., 2003—All cancer sites combined, site-specific analyses B-17
B-14. Consonni et al., 2008—All cancer sites combined, site-specific analyses B-18
B-15. Baccarelli et al., 2006—Site-specific analysis B-19
B-16. Warner et al., 2002—Breast cancer incidence B-21
B-17. Revich et al., 2001—All cancer sites combined, and site-specific analyses B-22
B-18. Akhtar et al., 2004—All cancer sites combined and site-specific analyses B-23
B-19. Michalek and Pavuk, 2008—All cancer sites combined B-24
B-20. 't Mannetje et al., 2005—All cancer sites combined, site specific analyses B-26
B-21. McBride et al., 2009b—All cancer sites combined, site-specific analysis B-27
B-22. McBride et al., 2009a—All cancer sites combined, site-specific analysis B-28
B-23. Hooiveld et al., 1998—All cancer sites combined, site-specific analysis B-29
B-24. Steenland et al., 1999—Mortality (noncancer) B-31
B-25. Collins et al., 2009—Mortality (noncancer) B-32
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-iii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
B-26. Ott and Zober, 1996—Mortality (noncancer) B-33
B-27. Flesch-Janys et al., 1995; Flesch-Janys et al., 1996 erratum—Mortality
(noncancer) B-35
B-28. Eskenazi et al., 2002a—Menstrual cycle characteristics B-36
B-29. Eskenazi et al., 2002b—Endometriosis B-37
B-30. Eskenazi et al., 2003—Birth outcomes B-39
B-31. Warner et al., 2004—Age at menarche B-40
B-32. Eskenazi et al., 2005—Age at menopause B-41
B-33. Warner et al., 2007—Ovarian function B-42
B-34. Eskenazi et al., 2007—Uterine leiomyoma B-43
B-35. Mocarelli et al., 2008—Semen quality B-45
B-36. Mocarelli et al., 2000—Sex ratio B-46
B-37. Baccarelli et al., 2008—Neonatal thyroid function B-47
B-38. Alaluusua et al., 2004—Oral hygiene B-48
B-39. Bertazzi et al., 2001—Mortality (Noncancer) B-50
B-40. Consonni et al., 2008—Mortality (Noncancer) B-51
B-41. Baccarelli et al., 2005—Chloracne B-52
B-42. Baccarelli et al, 2002 and 2004—Immunological effects B-53
B-43. Revich et al., 2001—Mortality (noncancer) and reproductive health B-54
B-44. Michalek and Pavuk, 2008—Diabetes B-55
B-45. McBride et al., 2009a—Mortality (Noncancer) B-56
B-46. McBride et al., 2009b—Mortality (noncancer) B-58
B-47. Ryan et al., 2002—Sex ratio B-59
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-iv DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. The data sources to ascertain vital status and cause of death
information were the Social Security death files, the National Death Index, and the Internal
Revenue Service. Vital status could be determined for 98% of the cohort.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. While the authors provide compelling arguments that suggest
risks are not unduly biased by lack of cigarette smoking data, they acknowledge potential
biases that could exist for other occupational exposure (e.g., asbestos) for which data were
lacking.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. There was not a statistically significant linear trend of increasing
mortality with increased duration of exposure.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. This study used duration of exposure, at an individual level, as a
surrogate measure of TCDD. Duration of exposure determined by number of years workers
were involved in processes involving TCDD contamination. Exposure was determined by
reviewing, at each plant, operating conditions, job duties, records of TCDD levels in industrial
hygiene samples, intermediate reactants, products, and wastes. Exposure assessment was
limited and the uncertainty related to exposure measures not fully addressed.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. This is the largest of the occupational cohorts that has been exposed
to TCDD. The cohort consisted of 5,172 workers and a total of 265 cancer deaths. Site-
specific mortality analyses, including soft tissue sarcoma (n = 4), was limited by small
numbers.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-l DRAFT—DO NOT CITE OR QUOTE
APPENDIX B. EVALUATION OF CANCER AND NONCANCER
EPIDEMIOLOGICAL STUDIES FOR INCLUSION IN TCDD
DOSE-RESPONSE ASSESSMENT
B.l. EVALUATION OF CANCER STUDIES
B.1.1. NIOSH Cohort Studies
Table B-l. Fingerhut et al., 1991—All cancer sites, site-specific analysis
-------
Response
Criteria satisfied. New England Journal of Medicine, 1991;324:212-218. Authors address
the possibility of bias from lack of control for potential confounders such as smoking and other
occupational exposures. They address limitations of using death certificates for identifying
certain causes of deaths, and limitations of using duration of employment as an exposure
metric.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. Since this study used duration of exposure as the exposure metric, dose-
response relationships cannot be quantified.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. Models incorporated period of latency, and a surrogate measure of
cumulative TCDD exposure was modeled. The follow-up interval was sufficiently long
(1942-1987).
Conclusion
Overall, quantitative exposure data are lacking on an individual-level basis. Further dose-
response analysis should consider updated data for this cohort that includes serum-based
measures of TCDD, in addition to an extension of the follow-up period. Given these
limitations, this study is not further evaluated for TCDD dose-response assessment.
Table B-2. Steenland et al., 1999—All cancer sites combined, site-specific
analysis
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. The study evaluated mortality from all cancer sites (combined). As
described in the paper, the sources of vital status and cause of death information were received
from the Social Security death files, the National Death Index, and the Internal Revenue
Service. Vital status was known for 99.4% of the cohort members, cause of death information
is available for 98% of the decedents.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Occupational exposure to asbestos and 4-aminobiphenyl contributed
to some excess cancer, but no evidence of confounding for the relationship between TCDD
and all cancer mortality was detected following removal of workers who died of bladder
cancer. No information is available for cigarette smoking, although dose-response patterns
were stronger for nonsmoking related cancers. This finding suggests that smoking is not
responsible for excess cancer risk that was observed in the cohort.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. When a 15-year lag interval was incorporated into the exposure
metric a statistically significant dose-response pattern was observed for all cancer sites
combined with both a continuous measure of TCDD (p = 0.05) as well as one that was log-
transformed (p < 0.001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-2 DRAFT—DO NOT CITE OR QUOTE
-------
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. The study conducted detailed sensitivity analyses and evaluated
different assumptions regarding latency, log-transformed TCDD exposures, and half-life
values for TCDD.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. This is the largest of the occupational cohorts with exposures to
TCDD. The cohort consisted of 5,132 male workers and a total of 377 cancer deaths. This
permits characterization of risk for all cancer sites (combined).
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Journal of the National Cancer Institute, 1999; 91(9):779-786. The authors
discussed the potential for bias from smoking, and other occupational exposures for which
data for both were lacking at an individual basis.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Exposure scores assigned on an individual level using a job-exposure
matrix. The job-exposure matrix was based on estimated factor of contact with TCDD in
each job, level of TCCD contamination of materials at each plant over time, and proportion of
day worker could be in contact with materials. These factors were multiplied together to
derive a daily exposure score, which was accumulated over the working history of each
worker to obtain a cumulative measure of TCDD.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. The follow-up of the cohort extended from 1942 until the end of 1993.
Greater than 25 years of follow-up have accrued in cohort allowing for latency to be
examined. Different assumptions on the half-life of TCDD were evaluated and produced
similar results. Latency intervals were incorporated, with strongest associations noted with an
interval of 15 years.
Conclusion
This study meets the criteria and considerations noted above but has been superseded and
updated by Steenland et al. (2001). Therefore, this study was considered for further dose-
response analyses.
1
2
3
4
5
6
7
8
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-3 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table B-3. Steenland et al., 2001—All cancer sites combined
2
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. The study evaluated mortality from all cancer sites (combined). As
described by Steenland et al., (1999) the sources of vital status and cause of death information
were received from the Social Security death files, the National Death Index, and the Internal
Revenue Service. Vital status was known for 99.4% of the cohort members, cause of death
information is available for 98% of the decedents.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Occupational exposure to asbestos and 4-aminobiphenyl contributed
to some excess cancer, but no evidence of confounding for the relationship between TCDD
and all cancer mortality was detected following removal of workers who died of bladder
cancer. No information is available for cigarette smoking, although dose-response patterns
were similar between smoking and nonsmoking related cancers.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Increased risk estimates were observed in the higher cumulative
exposure categories. The dose-response curve was not linear at higher doses.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied.
Exposure metrics considered included cumulative TCDD, loglOTCDD, average exposure, and
a cubic spline model was also evaluated. Exposure response relationships were also evaluated
using TEQs. Exposure scores were assigned on an individual level using a job-exposure
matrix. The job-exposure matrix was based on estimated factor of contact with TCDD in each
job, level of TCCD contamination of materials at each plant over time, and proportion of day
worker could be in contact with materials. Serum levels were measured in 199 workers at one
of 8 plants in 1998. Different estimate of the half-life of TCDD were used, and similar results
were produced. The paper presented a range in risk estimates thereby conveying the range of
uncertainties in risk estimates derived using different measures of exposure.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. This is the largest of the occupational cohorts with exposures to
TCDD. The cohort consisted of 3,538 male workers and a total of 256 cancer deaths.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied Am J Epidem, 2001, 154(5)451-458. However, additional details to assess
uncertainties associated with characterizing serum data in a subset of workers to remainder of
cohort are lacking.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-4 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria satisfied. The metrics considered included cumulative TCDD, loglOTCDD, average
exposure, and a cubic spline model was also evaluated. Exposure response relationships were
also evaluated using TEQs. Serum lipid TCDD measurements from 170 workers whose
TCDD levels were greater than 10 ppt (the upper ranges of a background level) were used
along with JEM information, work histories, and a pharmacokinetic elimination model to
estimate dose rates per unit exposure score. In this regression model, the estimated TCDD
level at the time of last exposure was modeled as a function of exposure scores. The
coefficient relating serum levels and exposure scores was then used to estimate serum TCDD
levels over time from occupational exposure (minus the background level) for all 3,538
workers. Time-specific serum levels were then integrated over time to derive a cumulative
serum lipid concentration due to occupational exposure for each worker.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. Greater than 25 years of follow-up have accrued in cohort allowing for
latency to be examined. Different assumptions on the half-life of TCDD were evaluated
producing similar results.
Conclusion
Overall, criterion has been satisfied and it is recommended that this study be considered for
dose-response analysis.
Table B-4. Cheng et al., 2006—All cancer sites combined
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. The study evaluated cancer mortality. The vital status and the
information regarding the cause of death were extracted from the Social Security death files,
the National Death Index, and the Internal Revenue Service (Steenland et al., 1999). Vital
status was known for 99.4% of the cohort members, while cause of death information is
available for 98% of the decedents.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. This is the same data set used in the Steenland et al., (2001) paper.
Occupational exposure to asbestos and 4-aminobiphenyl contributed to some excess cancer,
but no evidence of confounding for the relationship between TCDD and all cancer mortality
was detected following removal of workers who died of bladder cancer. No information is
available for cigarette smoking, although dose-response patterns were similar between
smoking and nonsmoking related cancers.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Slope coefficients are available for all cancers combined under a
varying set of assumptions. Little evidence of an association was found when lag interval was
not taken into account. Associations strengthened with incorporation of a 10 to 15 year lag
interval. Dose-response was nonlinear at higher exposures, suggesting a nonlinear
relationship or increased exposure misclassification at higher levels.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-5 DRAFT—DO NOT CITE OR QUOTE
-------
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Compared to the 1st order models, the concentration, and age
dependent model (CADM) provided a better fit for the serum sampling data. CADM model
exposure estimates are higher than those based on an age only, constant 8.7-year half-life
model. As discussed by Aylward et al. (2005b), model exposure estimates are influenced not
only by choice of elimination model, but also by choices in regression procedure (e.g., log
transformation, use of intercept, and incorporation of background dose term). Other
limitations or uncertainties in exposure assessment include the following
• Job-exposure matrix based on limited sampling data, and subjective judgment on contact
times and factors
• Inability to take into account inter-individual variability in TCDD elimination kinetics
• Dose-rate regressions are based on a small sample of the cohort with serum measures;
therefore, regression results may not be representative of remainder of the cohort.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Largest cohort of TCDD exposed workers. The risk estimates are
based on a total of 256 cancer deaths.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Risk Analysis, 2006; 4:1059-1071. Additional details to assess
uncertainties associated with characterizing serum data can be found in Aylward et al.
(2005b); Risk Anal. 25(4):945-956.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Cumulative serum lipid concentrations were estimated for each worker. No
other dioxin-like compounds were assessed in this analysis.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. Concentration and age-dependence of TCDD elimination and two
compartments (hepatic and adipose tissue) were taken into account when estimating TCDD
exposures. Nearly 50 years of follow-up were available permitting an evaluation of latency.
Conclusion
This study met the main criteria and considerations. These data were considered for further
dose-response analyses.
Table B-5. Collins et al., 2009—All cancer sites combined, site-specific
analysis
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Vital status complete for all but two workers.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-6 DRAFT—DO NOT CITE OR QUOTE
-------
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. No information collected on smoking status, but no excess in lung
cancer or nonmalignant respiratory diseases noted. Analyses took into account potential for
exposure to pentachlorophenol.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. No dose-response pattern was observed with all cancer sites
combined, however, a dose-response pattern was observed with soft tissue sarcoma. The study
found no association between TCDD and death from most types of cancer.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. The authors used these serum from 280 former TCP workers to
estimate historical exposure levels of TCDD, furans, and polychlorinated biphenyls for all
1,615 workers. Exposure assessment included detailed work history, industrial hygiene
monitoring, and the presence of chloracne cases among groups of workers. This data was
integrated into a 1-compartment, first-order pharmacokinetic to determine the average TCDD
dose associated with jobs in each group, after accounting for the presence of background
exposures estimated from the residual serum TCDD concentration in the sampled individuals.
The authors did not evaluate departures from linearity, or examine skewness at higher
exposures. Exposure levels were not provided.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Largest study of workers employed in one center, and a total of 177
deaths from cancer were observed. Limited precision for soft tissue sarcomas for which a
positive association has been demonstrated.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Published in Am J Epidemiol, 2009, 170(4):501-506. The authors discuss
limitations of using death certificates for identifying deaths from soft tissue sarcoma for which
a positive association was noted, assumptions in exposure characterization, and effects of
cigarette smoking.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. This study has the largest number of serum samples obtained from a specific
plant.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. While the cohort did have sufficient follow-up, no evaluation of possible
latent effects was presented.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-7 DRAFT—DO NOT CITE OR QUOTE
-------
Conclusion
Study is not suitable for further evaluation or dose-response modeling since an exposure-
response relationship was not demonstrated. The evaluation of exposure metrics and latency
considerations should be expanded beyond that presented in the paper. Previous analyses of
these same workers found positive associations between cancer mortality and TCDD
(Steenland et al., 2001). The reasons for the discrepancy in the findings from the two papers
may be due to Steenland et al.'s use of nonlinear exposure metrics, incorporation of a 15-year
lag interval, or differences in the TCDD exposure estimates themselves.
1
2 B.1.2. BASF Cohort Studies
3
4 Table B-6. Zober et al., 1990—All cancer sites combined, site-specific
5 analysis
6
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. A large component of the cohort (94 out of 247 workers) was
assembled by actively seeking out workers who were alive in 1986 through the "Dioxin
Investigation Programme." As a result, it is likely a number of deaths were missed due to the
recruitment of survivors. This underascertainment is supported by much lower all cancer
SMR one component of the cohort (SMR = 0.48, 95% CI: 0.13-1.23) relative to the general
population.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. See above discussion of underascertainment in mortality for some of
the cohort members. Although it is likely that other co-exposures occurred (e.g., among
firefighters), confounding could only occur if these co-exposures were associated with both
the endpoint and exposure (TCDD) being considered.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. Workers were not categorized on the basis of their exposure, but
rather their mortality experience compared to control cohort and the general population. The
design of the study does not allow for dose-response to be examined.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. Although years since first exposure was examined, exposure
assessment was based on working in various occupational cohorts. Since there was no
quantitative assignment of TCDD exposures, the associated uncertainties could not be
evaluated.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration not satisfied. There were only 23 cancer deaths in the entire cohort. As such,
this study lacked adequate statistical power to detect cancer mortality differences that were
moderate in magnitude.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-8 DRAFT—DO NOT CITE OR QUOTE
-------
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Int Arch Occup Envir Health, 1990,62:139-157. The authors address
issues related to the healthy worker effect, multiple comparisons, smoking, and small size of
the cohort.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. Risks were derived by comparing mortality rates of the three cohort
subsets relative to a control cohort and the general population by time since first exposure
categories. Workers were not assigned exposures. There were no quantitative estimates of
TCDD exposure.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. While the study was able to indirectly look at variations in risk estimates
related to latency by using time since exposure, there were no quantitative estimates of TCDD
exposure.
Conclusion
This study is not suitable for dose-response analysis, as it failed the inclusion criteria. Most
notably, the lack of exposure data does not permit the use of these data for a dose-response
analysis.
Table B-7. Ott and Zober, 1996—All cancer sites combined
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Mortality ascertainment appeared to be fairly complete. The
ascertainment of cancer incidence is more difficult to judge as geographical area not covered
by a cancer registry.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Information was collected on smoking status, body mass index, and
other occupational exposures, however a large portion of the cohort was firefighters who may
have been exposed to other occupational carcinogens. However, the recruitment of survivors
may results in under-ascertainment of mortality.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Increased cancer incidence was observed in the highest TCDD
cumulative exposure category. Risks were most pronounced when a period of 20 years since
first exposure was incorporated into the model.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-9 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. Cumulative measure of TCDD expressed was derived from serum
measures. Exposure was also estimated by chloracne status of the cohort members. The
authors have not addressed the potential implication of deriving TCDD exposure estimates for
the whole cohort using sera data that were available for only about half of the cohort.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For all cancer sites combined, there were 31 deaths. It is the smallest
of the occupational cohorts, but the deaths can be grouped into quartiles to allow for
evaluation of dose-response relationships.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Occupational and Environmental Medicine, 1996,53:606-612. A large
component of the cohort (94 out of 247 workers) was assembled by actively seeking out
workers who were alive in 1986 through the "Dioxin Investigation Programme." As a result,
it is likely a number of deaths were missed due to the recruitment of survivors. This
underascertainment is supported by much lower all cancer SMR one component of the cohort
(SMR = 0.48, 95% CI: 0.13-1.23) relative to the general population (Zober et al., 1990).
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum samples, taken in 1989, were available for 138 surviving workers out
of 254 and allowed for cumulative TCDD levels to be estimated using regression techniques in
the remainder of the cohort.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. Exposure assignment took into the affect that body mass index had on
TCDD half-lives. TCDD levels estimates through back-extrapolation of serum levels based on
half-life estimates obtained from previous studies. Latency was considered with stronger
association observed in external comparisons incorporating a latency of 20 years. The follow-
up of the cohort was lengthy (>50 years).
Conclusion
Given a part of the cohort was based solely on survivors in the in the mid-1980s, the SMR
statistic derived from this study underestimates excess mortality relative to the general
population. The cohort also includes a fair number of firefighters who are recognized to be
exposed to other carcinogenic agents—these exposures may be confounding the associations
that were reported. However, exposure to TCDD was quantified and the effective dose and
oral exposure estimable. Therefore, quantitative dose-response analyses were considered for
these data.
1
2
3
4
5
6
7
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-10 DRAFT—DO NOT CITE OR QUOTE
-------
1 B.1.3. The Hamburg Cohort
2
3 Table B-8. Manz et al., 1991—All cancer sites combined, site-specific
4 analyses
5
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Deaths were identified through medical records of the cohort
members. A review of death certificates of the identified cancer deaths found a high degree of
concordance (51/54). One of the 136 noncancer death certificates examined indicated an
"occult" neoplasm.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Smoking data were similar between exposed and nonexposed cohort
based on independent samples. Occupational exposure for which individual data are lacking
unlikely to explain dose-response with TCDD.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Dose-response patterns across three levels of exposure observed
among those who started work before 1954, and among those who worked for 20 years or
longer. Dose-response patterns not evident across whole cohort, among those with less than
20 years of employment, or among those who started after 1954.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Categorical exposures were based on TCDD concentrations in
precursor materials, products, waste, and soil from the plant grounds, measured after the plant
closed in 1984. Exposure uncertainty examined using a separate group of 48 workers who
provided adipose tissue samples. Other surrogate measures of exposure were considered in
this study, including duration of exposure and year of first employment.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For all cancer sites combined, there were 65 cancer deaths for the
comparison to the comparison cohort of gas workers. The study is underpowered to look at
site-specific cancers.
1 Ci'ileiia
Studs is published mi llie peei'-re\ ie\\ed scientific hieialuie and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Lance,1 1991, 338:959-964. The authors discussed potential for
misclassification using death certificates, healthy worker effect and their related use of a
comparison cohort of gas supply workers, other occupational exposures present at the plant,
potential impact and the lack of smoking data.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. Exposure consisted of a large DLC component that was not quantified.
Given crude TCDD exposure categorization data, no quantitative exposure metric was derived.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. Exposure metrics were constructed that took into account duration of
exposure, and periods when exposure was highest. However, exposure estimates did not
consider lagged exposure.
Conclusion
This study is not amenable to further TCDD dose-response analysis because it consisted of a
large DLC component that was quantified and no quantitative exposure metric was derived.
The dose-response patterns of risks observed across the three exposure groups provide
compelling support for an association between TCDD and cancer mortality, particularly, given
the associations observed when analyses restricted to those who were hired when TCDD
exposures were known to be much higher, and among those who worked for at least 20 years.
Subsequent studies improved the exposure assessment through the use of serum measures.
Table B-9. Flesch-Janys et al., 1995; Flesch-Janys et al., 1996 erratum—All
cancer sites combined
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Medical records used to identify deaths over the period 1952-1992.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Similarity in smoking rates between control cohort and the exposed
workers was similar based on independent surveys. Occupational exposures to benzene, and
dimethyl sulfate were unlikely to bias dose-response pattern observed as these exposures
occurred in production departments with low-medium levels of exposure.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Dose-response relationship observed across 6 exposure categories,
with the cohort of gas supply workers used as the referent.
4. Consideration
Consideration satisfied. Exposure assessment methodology is clear and adequately
characterizes individual-level exposures. The limitations and uncertainties in the exposure
assessment are considered.
Response
The exposure measure was an integrated TCDD concentration over time estimate that back-
calculated TCDD exposures to the end of the employment. Categorical and continuous TCDD
exposures were examined in relation to the health outcome. These efforts improve the
exposure assessment of earlier studies.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For all cancer sites combined, there were 124 deaths in the exposed
cohort, and 283 in the cohort of gas supply workers. No site-specific cancers were examined
in this paper.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-12 DRAFT—DO NOT CITE OR QUOTE
-------
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 1995, 1442:1165-1175. The authors discuss the potential
role of other occupational exposures (i.e., dimethyl sulfate, solvents, and benzene), smoking,
and suitability of the comparison cohort of gas supply workers.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum and adipose tissues were used to estimate TCDD exposure in 190
workers. A one-compartment first-order kinetic model was used to estimate exposure at end
of exposure for these workers. Regression methods were then used to estimates TCDD
exposures for all workers.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. Exposure was based on half-life estimates from individuals with
repeated serum measures. Other dioxin-like compounds were considered with the TOTTEQ
exposure metric. No consideration, however, was given to latency or lagged exposures.
Conclusion
The exposure data used within this study are well-suited to a dose-response analysis given the
associations observed, the characterization of exposure using serum, and quality of
ascertainment of cancer outcomes. However, subsequent methods have been applied to the
cohort to derive different exposures to TCDD using area under the curve approaches, which
updates the analysis herein. Therefore, subsequent studies (i.e., Becher et al., 1998) will
supersede this evaluation.
Table B-10. Flesch-Janys et al., 1998—All cancer sites combined, site-
specific analysis
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Mortality follow-up was extended until the end of 1992, an increase
in 3 years from previous analyses of the cohort.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Exposure was well characterized using sera data. While serum
samples provided only from a subsample of surviving workers, these levels were consistent
with expected levels in different production departments. The authors examined other
potential occupational co-exposures (e.g., p-hexachlorocyclohexane) and indirectly examined
the potential effect of smoking on the associations that were detected.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. A dose-response relationship across quartiles of TCDD was observed
with cancer mortality based on the SMR statistic (SMRs = 1.24, 1.34, 1.34, 1.73), and a linear
test for trend was statistically significant (p = 0.01).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-13 DRAFT—DO NOT CITE OR QUOTE
-------
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. The exposure measure was an integrated TCDD concentration over
time estimate that back-calculated TCDD exposures to the end of the employment.
Categorical and continuous TCDD exposures were examined in relation to the health outcome.
These efforts improve the exposure assessment of earlier studies.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For all cancer sites combined, there were 124 cancer deaths.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Environ Health Perspect, 1998, 106(2):655-662. The authors address
uncertainties in the estimation of exposure, describe the potential for confounding from (3-
2,4,5-T, hexachlorocyclohexane, and cigarette smoking. In fact, they showed that blood levels
of TCDD were not associated with smoking in a sub-sample suggesting little bias from lack of
smoking data.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum samples, taken from 190 workers were used to derive TCDD levels
for the entire cohort. Methods used to estimate exposure took into account elimination of
TCDD during employment periods when exposure took place, and the methods of the area
under the curve was used as it takes into account variations in concentration over time, and
reflects cumulative exposure.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. Exposure estimated based on half-lives observed in individuals with
repeated samples. Area under the curve approach was used which is an improvement from
past characterizations of exposure in this cohort.
Conclusion
The study provides data suitable for dose-response modeling. Derivation of exposure was
done using current understanding of elimination of TCDD. Estimates of risks were derived
from external comparisons to the general population that are unlikely to be biased by healthy
worker effect, but risks generated using internal cohort comparisons would be preferable.
Becher et al,. (1998) assessed this same data taking cancer latency into account, therefore
Flesch-Janys et al., (1998) will not be further considered for dose-response modeling.
Table B-ll. Becher et al., 1998—All cancer sites combined
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Medical records used to identify deaths over the period 1952-1992.
The follow-up interval was lengthy.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-14 DRAFT—DO NOT CITE OR QUOTE
-------
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Risks adjusted for exposures to TEQ, (3-hexachlorbenzene, and
employment characteristics. Smoking was shown to be similar to the comparison cohort of
gas workers.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. A variety of exposure measures for both TCDD and TEQs found
positive associations with cancer mortality.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. The exposure measure was an integrated TCDD concentration over
time estimate that back-calculated TCDD exposures to the end of the employment.
Categorical and continuous TCDD exposures were examined in relation to the health outcome.
Different models explored the shape of the dose-response curve. These efforts improve the
exposure assessment of earlier studies.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For all cancer sites combined, there were 124 cancer deaths.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Environ Health Perspect, 1998, 106(2):663-670. The authors discuss
uncertainties associated with their use of exposure metrics, inability to evaluate effects for
PCDD/Fs other than dioxin due to high correlations with (3-HCH, and inability to characterize
risks associated with exposures in children.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. The authors derived a measure of cumulative dose as a time-dependent
variable ("area under curve") using serum measures available in a sample of 275 workers.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. TCDD levels estimates through back-extrapolation of serum levels based on
half-life estimates obtained from previous studies. Latency was considered, and a variety of
exposure metrics including nonlinear relationships were evaluated.
Conclusion
In this paper, a variety of exposure metrics were found to be positively associated with cancer
mortality. The additional lifetime risk of cancer corresponded to a daily intake of lpg ranged
between .01 and 0.001. This study was modeled in the 2003 Reassessment and is considered
for further dose-response evaluations herein.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-15 DRAFT—DO NOT CITE OR QUOTE
-------
1 B.1.4. The Seveso Cohort Studies
2
3 Table B-12. Bertazzi et al., 2001—All cancer sites combined, site-specific
4 analyses
5
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Mortality appears to be well captured from the vital statistics
registries in the region (99% complete). Vital status was ascertained using similar methods for
both the exposed and reference populations. Both cancer and noncancer mortality outcomes
were evaluated. Ideally, would have evaluated incident rather than decedent outcomes for
cancer.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Individual-level data on potential confounders (i.e., age, calendar
period, and gender) were adjusted for. Information from other independent surveys suggests
similarity between smoking behaviors across the regions. Comparison of cancer mortality
rates before the time of the accident between the regions also revealed no differences.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied (for all cancers combined). No statistically significant excesses noted
in Zone A, or Zone B relative to reference area. Evidence of an exposure-response
relationship was detected for lymphatic and hematopoietic tissues by number of years since
first exposure.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. Subjects were assigned to one of the zones (A, B, R, or
reference) based on official residence on the day of the accident or at entry into the area.
Exposure misclassification is likely and lack of individual-level data precludes an examination
of this source of error.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. In total, 27, and 222, cancer deaths were found among residents of
Zones A, and B, respectively. This allowed examined of gender-specific effects.
1. Criicna
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 2001 Jun 1; 153(11): 1031—1044. Authors discuss
completeness of mortality ascertainment, diagnostic accuracy of death certificates particularly
with respect to diabetes, limited available of blood dioxin measures that did not permit
estimation of TCDD dose on an individual-level basis.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-16 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria not satisfied. Individual-level exposure data are unavailable. Exposure based on place
of residence at time of the explosion. Soil sampling performed indicated considerable
variability in TCDD levels within each region. In addition, place of residency at time of
explosion does not ensure individuals were at their home around the time of the accident.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. An ecological measure of exposure (region of residency at time of
accident) was used to categorize individuals according to their possible exposure. Latencies
were considered. While such an approach has value for identifying wherever excesses
occurred among highly exposed populations, it is not precise enough to conduct a quantitative
dose-response analysis.
Conclusion
The lack of individual-level exposure data precludes quantitative dose-response modeling
using these data.
Table B-13. Pesatori et al., 2003—All cancer sites combined, site-specific
analyses
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Mortality appears to be well captured from the vital statistics
registries in the region (99% complete).
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Individual-level data on potential confounders (i.e., age, calendar
period, and gender) were adjusted for.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. While excesses of mortality were observed for several health
conditions in Zone A, a dose-response pattern was not observed across Zones A, B and R.
Among men, excess mortality observed in zone A included chronic ischemic disease, and
chronic obstructive pulmonary diseases. Among females, an excess in Zone A was observed
with hypertension.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. Subjects were assigned to one of the zones (A, B, R, or
reference) based on official residence on the day of the accident or at entry into the area.
Exposure misclassification is likely and lack of individual-level data precludes an examination
of this source of error.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-17 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. Only 39 deaths observed among men in Zone A; 39 deaths observed
among their female counterparts. Among females, only 3 deaths from hypertension observed
in Zone A, and only 4 deaths observed among males for chronic obstructive pulmonary
disease.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. OccupEnvMed, 1998;55:126-131. Authors discuss limitations such as
residency-based exposure assignment, absence of smoking, differential and death certification
in exposed versus nonexposed areas.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. Individual-level exposure data are unavailable. Exposure based on place
of residence at time of the explosion. Soil sampling performed indicated considerable
variability in TCDD levels within each region. In addition, place of residency at time of
explosion does not ensure individuals were at their home around the time of the accident.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. An ecological measure of exposure (region of residency at time of
accident) was used to categorize individuals according to their possible exposure. Latencies
were considered. While such an approach has value for identifying wherever excesses
occurred among highly exposed populations, it is not precise enough to conduct a quantitative
dose-response analysis.
Conclusion
No dose-response patterns evident in the study, and the study lacked quantifiable measures of
TCDD at an individual-level basis. The data are not well suited for dose-response analysis.
Table B-14. Consonni et al., 2008—All cancer sites combined, site-specific
analyses
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Mortality appears to be well captured from the vital statistics
registries in the region (99% complete. Both cancer and noncancer mortality evaluated),
although diagnostic accuracy of death certificates is likely low. Ideally, would have evaluated
incident rather than decedent outcomes for cancer.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Individual-level data on potential confounders (i.e., age, calendar
period, and gender) were adjusted for. Comparison of cancer mortality rates before the time of
the accident between the regions also revealed no differences. Information from other
independent surveys suggests similarity between smoking behaviors across the regions.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-18 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied for some outcomes. For all cancer sites combined, no evidence of
dose-response was observed relative to general population across Zones A, B and R. Only
statistically significant excess found in Zone A was for chronic rheumatic disease but based on
only three deaths. Higher cancer excesses were found in Zone A after a latency period was
incorporated; however, no dose-response relationship observed with this latency period.
Evidence of an exposure-response relationship was detected for lymphatic and hematopoietic
tissues by zone of residence.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. Subjects were assigned to one of the zones (A, B, R, or
reference) based on official residence on the day of the accident or at entry into the area.
Exposure misclassification is likely and lack of individual-level data precludes an examination
of this source of error.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. In total, 42, 244, and 1,848 cancer deaths were found among residents
of Zones A, B, and R respectively.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 2008, 167:847-858. Authors discuss potential for
selection bias, limitation of residential based measure of exposure, similarities of mortality
ascertainment in exposed and referent populations, and multiple testing.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. Individual-level exposure data are unavailable. Exposure based on place
of residence at time of the explosion. Soil sampling performed indicated considerable
variability in TCDD levels within each region. In addition, place of residency at time of
explosion does not ensure individuals were at their home around the time of the accident.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. An ecological measure of exposure (region of residency at time of
accident) was used to categorize individuals according to their possible exposure. Latencies
were considered. While such an approach has value for identifying wherever excesses
occurred among highly exposed populations, it is not precise enough to conduct a quantitative
dose-response analysis.
Conclusion
The lack of individual-level exposure data precludes quantitative dose-response modeling
using these data.
Table B-15. Baccarelli et al., 2006—Site-specific analysis
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-19 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. Polymerase chain reaction (PCR) methods were used to describe
outcome measures. The prevalence of t(14; 18) was estimated as those individuals having a
t(14; 18) positive blood sample divided by the t(14; 18) frequency (number of copies per
million lymphocytes).
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Questionnaire data were used to collect information on cigarette
smoking. Other potential confounders (age, smoking status, and duration of smoking). In
addition, both exposure and outcome were objectively and accurately measured.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration was not satisfied. Associations were detected between the frequency of t(14;
18) and plasma TCDD levels as well as zone of residence at the time of the explosion. No
association was detected for these exposure measures and prevalence of t(14; 18). A dose-
response trend was detected for TCDD and the mean number of t(14; 18) translocations/106
lymphocytes, however the relevance of t(14; 18) in lymphocytes to Non-Hodgin's lymphoma
is uncertain.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. The authors highlight that exposure metrics represent both past and
current body burdens. They employ several different exposure metrics of TCDD: place of
residence (Zone A, B, R or reference), categorical serum measures, a linear term, log (base 10)
transformed TCDD, and individuals with chloracne diagnosed after the accident.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Analyses are made using 72 highly exposed, and 72 low exposed
individuals.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Carcinogenesis, 2006, 27(10):2001-2007. The authors discuss the
limitation of using t(14; 18) translocations as an outcome measure, and the uncertain role it
plays in the development of non-Hodgkin's lymphoma.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. A total of 144 subjects were included in the study. This included 72
subjects who had low exposures, and 72 who had high exposures based on serum
concentrations.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. A variety of measures were employed including current TCDD levels, as
well as siin oa.'ilcs of exposure at the time of the accident
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-20 DRAFT—DO NOT CITE OR QUOTE
-------
Conclusion
While an association was observed with the frequency of t(14; 18) translocation, it is uncertain
whether this translates into an increased risk of non-Hodgkin's lymphoma. Given the
speculative nature of this endpoint and lack of demonstrated adverse effect, dose-response
analyses for this outcome were not conducted.
Table B-16. Warner et al., 2002—Breast cancer incidence
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Diagnoses of incident breast cancer were based on interview and
information from medical records appears thorough. Of the 15 cases of breast cancer, 13 were
confirmed by pathology and the remaining 2 by surgery report only. Three cases of breast
cancer were excluded which represents a large proportion of the total cases identified. This
would reduce sample size and could result in bias if the exclusion was association with TCDD
exposure.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Information was collected on an extensive series of risk factors by
using an interviewer administered questionnaire. Participation rates for the survey were fairly
good (80%).
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Limited evidence (not statistically significant) of a dose-response
when TCDD was analyzed as a categorical variable; only one breast cancer case was in the
referent exposure category. In the analysis of TCDD as a continuous measure (logi0TCDD),
the hazard ratio associated with a 10-fold increase in TCDD serum levels was 2.1 (95% CI:
1.0^.6).
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Different exposure metrics were considered in these analyses
(categorical, continuous, measures on a log-scale). Exposure data are of high quality as they
are based on serum samples taken among women near the time of the accident. As such,
exposure assignment is not dependent on as many assumption as used in occupational cohorts
were back-extrapolation for many years had to be performed.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration somewhat satisfied. Inadequate follow-up for cancer limited the number of
cases available. Sample size also limited the conclusions draw from the categorical analysis
based on very few cases for some exposure categories.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Paper published in Environ Health Perspect, 2002 Jul, 110(7):625-628. .
A major limitation of the study is the small number of incident cases of breast cancer (n = 15),
discussed important strengths of the study including characterization of TCDD using serum
collected near the time of the accident.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-21 DRAFT—DO NOT CITE OR QUOTE
-------
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum was used to estimate TCDD levels in 981 of 1271 eligible women
who had lived in either of the two contaminated sites in 1976. Data represent an objective
measure of TCDD near the time of the exposure. Data obtained near the time of exposure
which minimized the potential for exposure misclassification.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. Exposure characterized using serum measures obtained close to the time of
the accident.
Conclusion
While characterization of exposure and availability of other risk factor data at an individual-
level basis are important strengths of this study, small sample size (n = 15 cases) based on
inadequate follow-up is a key limitation. Quantitative dose-response analyses were conducted
using this study, but continued follow-up of the study population or consideration of all cancer
outcomes would be valuable.
B.1.5. The Chapaevsk Study
Table B-17. Revich et al., 2001—All cancer sites combined, and site-specific
analyses
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration cannot be evaluated. Insufficient details are provided in the paper to gauge the
completeness and coverage of the cancer registry and mortality data. Health outcomes were
studied on the basis of information in the official medical statistics.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. Given that this is an ecological study, bias may be present.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration cannot be evaluated. Dose-response was not evaluated as exposure was based on
residency in the region vs. no residency.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. No individual-level exposure estimates were used.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. A total of 476 cancer deaths were observed among males, and 376
cancer deaths observed among females. The precision of the SMRs is demonstrated with fairly
narrow confidence intervals for many causes of death.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-22 DRAFT—DO NOT CITE OR QUOTE
-------
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria not satisfied. Published in Chemosphere, 2001, 43(4-7):951-966. Authors do not
address the completeness of the mortality follow-up, and whether there are differences in death
registrations between regions. The authors do acknowledge, however, that new investigations
being undertaken would characterize exposure using serum-based measures.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. It is a cross-sectional study that compares mortality rates between
regions. No individual-level exposure data available.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of exposure
are consistent with the current biological understanding of dose. The reported dose is
consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. No individual-level exposure estimates were used in the study.
Conclusion
These cancer data are cross-sectional in nature and not appropriate for a dose-response analysis.
1
2
3
4
5
6
7
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Cancer incidence and mortality based on information from repeated
medical examinations, medical records and death certificate.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. The risk estimates were adjusted for a number of factors
measured on an individual level including smoking. However, analyses are unable to
distinguish between exposure to TCDD and 2,4-D as both were used in equal parts in the
formulation of Agent Orange.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. There is evidence of a dose-response for all cancers and for some
site-specific cancers (i.e., malignant melanoma, and prostate cancer).
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. High quality exposure data for most veterans was collected, so
extrapolation to other members of the cohort was not required. The serum dioxin
measurements also correlated well with reported skin exposure to herbicide in Vietnam, but
collection of the samples 25 years later required back-extrapolation.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-23 DRAFT—DO NOT CITE OR QUOTE
B.1.6. The Air Force Health ("Ranch Hands") Study
Table B-18. Akhtar et al., 2004—All cancer sites combined and site-specific
analyses
-------
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. In total, 117 incidence cancers identified in the Ranch Hands cohort.
For those sites with a dose-response association, malignant melanoma and prostate cancer,
there were 16 and 34 incident cases, respectively.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Published in J Occup Environ Med, 2004, 46(2): 123-136. Authors
highlight that this is only cancer incidence study in US veterans, and the lengthy interval of
follow-up (35-40 years)-both important strengths of the study. They addressed potential bias
from healthy-worker effect, and uncertainties surrounding the estimation of TCDD exposure
(extrapolation 30 years after exposure), as well as exposure to other chemical exposures.
Study uses incident outcomes for cancer.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Individual exposure estimates are based on measurements of dioxin serum
lipid concentrations. They were available for 1,009 Ranch Hands and 1,429 in the
comparison cohort.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. TCDD exposures at the end of duty were estimated by back-extrapolating
1987 serum values.
Conclusion
The major limitation of the study is the inability to isolate effects of TCDD from other
chemicals used in the formulation of the herbicides. This limitation precludes dose-response
modeling of the TCDD and cancer outcomes data.
Table B-19. Michalek and Pavuk, 2008—All cancer sites combined
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Cancer incidence was ascertained through the use of medical records.
Death certificate were used to identify some malignancies. Little data is provided on the
number of individuals lost to follow-up, however the same mechanisms of case ascertainment
were applied to both the comparison and Ranch Hand cohorts.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. Information collected from repeated physical examinations
allowed for the adjustment of risk factors such as smoking. Agent Orange was a 50% mixture
of 2,4-D and TCDD; therefore, potential for confounding by other coexposures is likely.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-24 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied for some comparisons. Statistically significant associations were
noted with cancer incidence and TCDD when analyses were restricted to workers who served
at most two years in Southeast Asia and those who sprayed more than 30 days before 1967.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Initial TCDD dose were estimated at the end of the tour of duty for
the Ranch Hands. Individual-level serum dioxin measurements correlated well with
correlated with days of spraying and calendar period of service, but collection of the samples
roughly 20 years later required back-extrapolation.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. A total of 347 incident cases of cancer were used in the analyses.
For stratified analyses, statistical power is more limited. For example, only 67 incident
cancer in the subset of workers who spent less than 2 years in Southeast Asia, and sprayed for
at least 30 days before 1967.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied J Occup Environ Med 2008; 50:330-340. The authors discuss issues related
to exposure misclassification error, and suggest approaches for improving characterization of
days of spraying. Congener specific data were unavailable, thereby not allowing for congener
specific risks or adjustments to be made.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. TCDD data was available for 986 veterans in the Ranch Hand cohort, and
1,597 members of the comparison cohort.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. TCDD exposures at the end of duty were estimated by back-extrapolating
1987 serum values.
Conclusion
Ranch Hand veterans were exposed to other contaminants in the herbicides that were mixed,
thereby making it difficult to determine independent effects of TCDD on cancer. In
particular, 2,4-D has been shown to be associated with some cancers, notable cancer of the
prostate. In our view, this limitation precludes dose-response modeling of TCDD and cancer
using data from this cohort.
1
2
3
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-25 DRAFT—DO NOT CITE OR QUOTE
-------
1 B.1.7. Other Studies of Potential Relevance to Dose-Response Modeling
2
3 Table B-20. 't Mannetje et al., 2005—All cancer sites combined, site specific
4 analyses
5
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. National records for death registrations through the New
Zealand Health Information Service (NZHIS). Subjects not registered as having died during
the study period were confirmed to be actually alive and resident in New Zealand using the
New Zealand Electoral Roll, drivers' license, and social security records.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. Seventeen percent of workers were lost to follow up but it is
unclear if bias resulted. The dichotomous exposure measure was based on exposure to
TCDD, chlorinated dioxins and phenoxy herbicides, so confounding is a possibility by these
coexposures.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Dose-response evidence for duration of employment and elevated
mortality noted only in synthesis workers.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Exposure measures were limited to duration of employment and exposed/unexposed.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For all cancer sites combined, there were 43 cancer deaths among the
production workers, and 35 such deaths among the sprayers. Site-specific cancer analyses are
limited by small sample sizes.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria not satisfied Occup Env Med, 2005; 62:34-40. A high percentage of the cohort was
lost to follow-up (17%). The authors fail to mention this important limitation in this paper.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. This study used duration of exposure, at an individual level, as a
surrogate measure of TCDD.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. Exposure was defined according to duration, and not concentrations of
TCDD. Latency intervals were not evaluated.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-26 DRAFT—DO NOT CITE OR QUOTE
-------
Conclusion
Overall, quantitative exposure data are lacking for TCDD and limited dose-response
relationships were observed across duration of exposure categories. Furthermore confounding
by coexposures is a possibility. Taken together, these data are not suitable for inclusion in a
dose-response analysis
Table B-21. McBride et al., 2009b—All cancer sites combined, site-specific
analysis
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. The New Zealand Health Information Service Mortality Collection
and the Registrar-General's Index to Deaths. Additional searches were based on the last
known address from the work record; the electoral roll and the habitation index; the telephone
book; the internet; and Terranet property information database. An additional search was
carried out through the Births, Deaths, and Marriages office of the New Zealand Department
of Internal Affairs. Lastly, automated personnel and pension records were also used to locate
past New Plymouth workers and identify some deaths.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. Considerable amount of workers were lost to follow up (22%), but
it is unclear if bias resulted. The dichotomous exposure measure was based on exposure to
TCDD, chlorinated dioxins and phenoxy herbicides, so confounding is a possibility by these
coexposures.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. There was no examination of dose-response effects.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Dichotomous exposure (exposed/unexposed) and duration of
employment were examined from job exposure classification assessed via occupational history
records industrial hygienists/factory personnel knowledge and questionnaires. Authors discuss
limitations in the assignment of exposure among cohort members.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration not satisfied. A low number of deaths (n = 76) may have limited ability to
detect effects small in magnitude and exposure-response relationships.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Published in Occup Medicine, 2009; 59(4):255-263. The authors highlight
cohort lost to follow-up, the limited size of the cohort, differences in cohort definitions
between sprayers and producers, and the potential for other exposures during employment at
the plant.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-27 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria not satisfied. TCDD exposures were not quantified.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. Effective dose could not be estimated given the lack of individual-level
exposure data.
Conclusion
The study lacks the quantification of exposures at an individual level precluded dose-response
analysis.
Table B-22. McBride et al., 2009a—All cancer sites combined, site-specific
analysis
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. The New Zealand Health Information Service Mortality Collection
and the Registrar-General's Index to Deaths were used to identify deaths. Additional searches
were based on the last known address from the work record; the electoral roll and the
habitation index; the telephone book; the internet; and several other public databases in New
Zealand. An additional search was carried out through the Births, Deaths, and Marriages
office of the New Zealand Department of Internal Affairs. Lastly, automated personnel and
pension records were also used to locate past New Plymouth workers and identify some
deaths.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Workers lost to follow-up were an unlikely source of bias especially
for internal analyses. Confounding by other coexposures (e.g., 2,4,6-TCP) unlikely to have
resulted in bias, due to presumed poor correlation with TCDD.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. The linear test for trend for TCDD exposure was not statistically
significant for all cancer sites (combined), as well as lung cancer mortality. Dose-response
relationships were not apparent across quartiles of TCDD exposure for all cancer sites
combined, digestive cancers, lung cancer, soft tissue sarcomas orNon-Hodgkin's Lymphoma.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Cumulative exposure to TCDD as a time-dependent metric was
estimated for each worker from serum samples, but the authors did not examine a continuous
measure of TCDD exposure (lagged orunlagged).
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-28 DRAFT—DO NOT CITE OR QUOTE
-------
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Published in J Occup Environ Med 51:1049-1056. This paper discussed the
22% of the cohort lost to follow-up, differences in cohort definitions between sprayers and
producers, and the potential for other exposures during employment at the plant.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum measures available for 346 workers were used to derive TCDD
exposures for the entire cohort using the area under the curve approach.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria satisfied. Effective dose could be estimated from serum-derived cumulative exposure
estimates.
Conclusion
Given that no dose-response associations were found, the data are not suited to dose-response
analysis.
Table B-23. Hooiveld et al., 1998—All cancer sites combined, site-specific
analysis
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Outcomes were mortality. Few deaths expected to be missed since
only 5% of the cohort was lost to follow-up or had emigrated.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. Although dioxin-like compounds (PCDDs, PCDFs, and PCBs)
were measured in the serum samples, these were not incorporated into the analysis. Therefore,
confounding cannot be ruled out as an explanation of the reported association.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. A dose-response pattern was observed for internal cohort comparison
for all cancer mortality, with RRs of 5.0 and 5.6 for the medium and high exposure,
respectively. Dose-response patterns evident fort lung cancer as well.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Detailed occupational histories to assign dichotomous exposures
(exposed/unexposed) based on maximum exposure levels. Although serum data also collected
for TCDD and other coexposures (PCDDs, PCDFs, and PCBs), study only presents data for
TCDD exposure. TCDD exposures at time of maximum exposure were extrapolated from
measured serum.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-29 DRAFT—DO NOT CITE OR QUOTE
-------
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration not satisfied for internal cohort comparisons in either men or women. Among
men, only 7 cancer deaths were observed among those in the unexposed part of the cohort, and
51 among exposed workers. For external cohort comparisons, a total of 20 deaths were
observed.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 1998, 147:891-901. The authors address potential
limitations of estimating TCDD exposure from a sub-sample of surviving workers, lack of
smoking data, the healthy worker effect, and relevance of other occupational exposures.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum samples were obtained from 94 of 144 subjects who were asked to
participate in serum measurement study. Of these, a further 44 excluded due to absence due to
holiday or work (n = 22), and nonexposed workers excluded because matching exposed
worker not participating (n = 20). TCDD levels were extrapolated to the time of maximum
exposure.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined.
Response
Criteria not satisfied. Exposures assigned based on levels at maximum exposure. Assignment
of exposure based on nonrepresentative sample of 50 survivors among the occupational cohort.
Conclusion
The small number of identified cancer deaths, limitations in terms of the exposure assignment
(based on nonrepresentative sample, and maximum exposure level) and concern over potential
confounding by coexposures preclude using these data for a dose-response analysis.
1
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-30 DRAFT—DO NOT CITE OR QUOTE
-------
1 B.2. EVALUATION OF NONCANCER STUDIES
2 B.2.1. NIOSH Cohort
3
4 Table B-24. Steenland et al., 1999—Mortality (noncancer)
5
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. The study evaluated mortality from all cancer sites (combined). As
described in the paper, the sources of vital status and cause of death information were received
from the Social Security death files, the National Death Index, and the Internal Revenue
Service. Vital status was known for 99.4% of the cohort members, cause of death information
is available for 98% of the decedents.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. External comparisons for all-cause and cardiovascular mortality
do not appear to be affected by the "healthy worker effect" as similar patterns were observed
with internal cohort comparisons. Nonetheless, internal cohort comparisons are unable to
adjust for many of the individual-level risk factors for cardiovascular disease.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. A dose-response relationship was observed with ischemic heart
disease (linear test for trendp = 0.05), and with TCDD on a log-transformed scale the p-value
was <0.001.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. The study conducted detailed sensitivity analyses and evaluated
different assumptions regarding latency, log-transformed TCDD exposures, and half-life
values for TCDD. Associations were stronger for log-transformed values, and latency
intervals of 15 years.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. This is the largest of the occupational cohorts with exposures to
TCDD. The cohort consisted of 5,132 male workers and a total of 456 deaths from ischemic
heart disease. This permits characterization of risk for all cancer sites (combined)
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied Journal of the National Cancer Institute, 1999, 91(9):779-786. The authors
discussed the potential for bias from smoking, and other occupational exposures for which
data for both were lacking at an individual basis.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-31 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria not satisfied. Exposure scores assigned at an individual level based on job-exposure
matrix (JEM). The JEM was based on estimated factor of contact with TCDD in each job,
level of TCCD contamination of materials at each plant over time, and proportion of day
worker could be in contact with materials. These factors were multiplied together to derive a
daily exposure score, which was accumulated over the working history of each worker to
obtain a cumulative measure of TCDD.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. The follow-up of the cohort extended from 1942 until the end of 1993.
Greater than 25 years of follow-up have accrued in cohort allowing for latency to be
examined. Different assumptions on the half-life of TCDD were evaluated and produced
similar results. Latency intervals were incorporated, with strongest associations noted no lag.
Suggests mechanisms occur at the same time as exposure. However, noncancer mortality is
ik»I a \ iable endpoiiil U» consider for (mllier dose-response anaUsis
Conclusion
TCDD exposures were quantified in this study, and a dose-response relationship was observed
with ischemic heart disease mortality. The sample size was sufficient, and the follow-up
interval was lengthy. However, no individual-level data were available for cardiovascular
conditions, and the inability to adjust for these exposures introduces considerable uncertainty
into the risk estimates. Furthermore, noncancer mortality is not considered a viable endpoint
for dose-response analysis.
Table B-25. Collins et al., 2009—Mortality (noncancer)
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Vital status complete for all but two workers.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. No information collected on smoking status, but no excess in lung
cancer or nonmalignant respiratory diseases noted. Analyses took into account potential for
exposure to pentachlorophenol. External cohort comparisons should be interpreted cautiously
due to healthy worker effect, but internal cohort comparisons should not be influence by this
bias.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. No statistically significant mortality excess for any noncancer
mortality outcome evaluated. This included ischemic heart disease, stroke, nonmalignant
respiratory disease, ulcers, cirrhosis, and external causes of death (accidents). Modeling of
continuous measure of TCDD was not related to diabetes, ischemic heart disease, or
nonmalignant respiratory mortality.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-32 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. The authors used these serum from 280 former TCP workers to
estimate historical exposure levels of TCDD, furans, and polychlorinated biphenyls for all
1,615 workers. Exposure assessment included detailed work history, industrial hygiene
monitoring, and the presence of chloracne cases among groups of workers. This data was
integrated into a 1-compartment, first-order pharmacokinetic to determine the average TCDD
dose associated with jobs in each group, after accounting for the presence of background
exposures estimated from the residual serum TCDD concentration in the sampled individuals.
The authors did not evaluate departures from linearity, or examine skewness at higher
exposures. No presentation of exposure levels was provided.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. A total of 662 deaths were observed. Of these, 218 were from
ischemic heart disease, and 16 from diabetes (two outcomes for which associations have been
noted elsewhere).
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Published in Am J Epidemiol, 2009, 170(4):501-506. The authors discuss
potential for exposure misclassification, large size of the cohort, lengthy follow-up interval,
and large number of workers who provided serum from which TCDD exposures were
estimated.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. This study has the greatest number of serum samples obtained from a
specific plant.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. Noncancer mortality is not a viable endpoint to consider for further
dose-response analysis.
Conclusions
No dose-response associations were noted for noncancer mortality outcomes. The data are,
therefore, not suited for dose-response modeling.
B.2.2. BASF Cohort
Table B-26. Ott and Zober, 1996—Mortality (noncancer)
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Mortality ascertainment appeared to be fairly complete.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-33 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. Information was collected on smoking status, body mass index, and
other occupational exposures, however a large portion of the cohort was firefighters who may
have been exposed to other occupational carcinogens. However, the recruitment of survivors
may results in under-ascertainment of mortality.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. For external cohort comparisons across the three TCDD exposure
categories, there was no dose-response pattern observed for any of the noncancer causes of
death. Cox regression risk estimates for all cause or circulatory disease mortality when TCDD
was modeled as a continuous variable were not statistically significant.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Cumulative measure of TCDD expressed was derived from serum
measures. Exposure was also estimated by chloracne status of the cohort members. The
authors have not addressed the potential implication of deriving TCDD exposure estimates for
the whole cohort using sera data that were available for only about half of the cohort.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For all causes of death, there were 92 deaths, while 37 circulatory
deaths. Many of the cause-specific death had less than 5 deaths in the upper exposure
category.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Occup Environ Med, 1996, 53:606-612. A large component of the cohort
was assembled by actively seeking out workers who were alive in the mid 1980s. As a result,
it is likely a number of deaths were missed. This is supported by much lower SMRs in this
component of the cohort published in earlier studies of the cohort. This underascertainment of
mortality results in biased SMR statistics (underestimated). The authors do highlight the value
of the serum based measures to estimate TCDD exposure
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum samples, taken in 1989, were available for 138 surviving workers out
of 254 and allowed for cumulative TCDD levels to be estimated using regression techniques in
the remainder of the cohort.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. Exposure assignment took into the affect that body mass index had on
TCDD half-lives. TCDD levels estimates through back-extrapolation of serum levels based on
half-life estimates obtained from previous studies. Latency was considered with stronger
association observed in external comparisons incorporating a latency of 20 years. The follow-
up of the cohort was lengthy (>50 years). However, noncancer mortality is not a viable
endpoint to consider for further dose-response analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-34 DRAFT—DO NOT CITE OR QUOTE
-------
Conclusion
No associations noted with any noncancer deaths. External comparisons should be treated
cautiously especially for cardiovascular mortality which is recognized to often be biased by
the healthy-worker effect. In the absence of any outcome with an association with TCDD
exposure, no dose-response analyses of these data are recommended.
1 B.2.3. Hamburg Cohort
2
3 Table B-27. Flesch-Janys et al., 1995; Flesch-Janys et al., 1996 erratum—
4 Mortality (noncancer)
5
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Medical records used to identify deaths over the period 1952-1992.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Similarity in smoking rates between control cohort and the exposed
workers was similar based on independent surveys. Occupational exposures to benzene, and
dimethyl sulfate were unlikely to bias dose-response pattern observed as these exposures
occurred in production departments with low to medium levels of TCDD exposure.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Dose-response relationship observed for all-cause mortality,
cardiovascular mortality, and ischemic heart disease mortality across 6 exposure categories,
with the cohort of gas supply workers used as the referent. The linear tests for trend for these
three outcomes were all statistically significant (p < 0.05).
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. The exposure measures was an integrated TCDD concentration over
time estimate that back-calculated TCDD exposures to the end of the employment.
Categorical and continuous TCDD exposures were examined in relation to the health outcome.
These efforts improve the exposure assessment of earlier studies.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For all causes of death combined, there were 414 deaths in the
exposed cohort, and 943 in the cohort of gas supply workers. A total of 157 and 76 deaths
from cardiovascular disease, and ischemic heart disease were noted. The corresponding
number in the cohort of gas supply workers was 459, and 205, respectively.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 1995, 1442:1165-1175. The authors discuss the potential
role of other occupational exposures (i.e., dimethyl sulfate, solvents, benzene), smoking, and
suitability of the comparison cohort of gas supply workers.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-35 DRAFT—DO NOT CITE OR QUOTE
-------
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum and adipose tissues were used to estimate TCDD exposure in 190
workers. A one-compartment first-order kinetic model was used to estimate exposure at end
of exposure for these workers. Regression methods were then used to estimates TCDD
exposures for all workers.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. Exposure based on half-life estimates from individuals with repeated
serum measures. Other dioxin-like compounds were considered with the TOTTEQ exposure
metric. Noncancer mortality, however, is not a viable endpoint to consider for further dose-
response analysis.
Conclusion
Although, the exposure data used within this study are well-suited to a dose-response analysis
for all-cause and cardiovascular mortality given the associations observed, use of noncancer
mortality endpoint is not amenable for further dose-response analysis.
1 B.2.4. The Seveso Women's Health Study
2
3 Table B-28. Eskenazi et al., 2002a—Menstrual cycle characteristics
4
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Information was also obtained from medical records for all obstetric
and gynecologic conditions. Information on menstrual cycles was obtained from
questionnaires. Women were asked about length of cycles, regularity, how many days flow
lasted, and heaviness of menstrual flow (scanty, moderate, or heavy). Measurement error is
likely for the subjective nature of self-reported menstrual parameters but specificity and
sensitivity is difficult to ascertain due to lack of validation data for these measures.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Detailed risk factor information was collected from questionnaire,
allowing for the potential confounding influence of many risk factors to be controlled for. The
length of cycle study findings may have been affected by the presence of a few outliers.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. A positive dose-response relationship was found with TCDD among
women who were premenarcheal at time of the explosion and longer menstrual cycle. Increased
TCDD resulted in a reduced odds of scanty menstrual flow. No association was noted with
these two outcomes among postmenarcheal women. A decreased risk of irregular cycles was
observed with higher TCDD levels.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-36 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria satisfied. Serum concentrations of TCDD offer improved exposure assessment,
although delineating the critical exposure window is challenging given the nature of the very
high initial exposure.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure adequate
statistical power.
Response
Consideration satisfied. Cohort was large enough as analyses were conducted on 301 women.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion of
the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 2002; 156(4) 383-392. Limitations included an inability to
assess affects on menstrual cycle at time body burdens were the highest (at time of the
accident). Also, TCDD was estimated for 1976, not concurrent with their cycles in the previous
year, and a large number of women were excluded due to intrauterine device or oral
contraceptive use. Strengths included population-based nature of study, with characterization
of exposure using serum, and levels of other polychlorinated dibenzo-p-dioxins and
dibenzofurans were at background levels. Findings for length of menstrual cycle may be
unduly influenced by the presence of some outliers.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. The study population was based on 301 women as those who were over the
age of 44 were excluded, as well as women with surgical of natural menopause, women with
Turner's syndrome, those who had been pregnant or breastfed in the past year, and those who
had used an intrauterine device or oral contraceptives. For 272 women, TCDD levels were
based on serum data provided in 1976; TCDD levels were back-extrapolated to 1976 levels for
the other 29 women.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of exposure
are consistent with the current biological understanding of dose. The reported dose is consistent
with a toxicologically relevant dose. Latency and appropriate window(s) of exposure
examined. Response had to be a nonfatal endpoint.
Response
Criteria satisfied. Ideally, TCDD exposures would be concurrent with reporting of cycle
characteristics. Herein, TCDD exposures were based on levels in 1976; however, given the
long half-life of TCDD and the same follow-up interval for all women, TCDD exposures in
1976 should correlate well with levels near the time of interview. Further, the critical window
of exposure can be estimated for the women that were premenarcheal at the time of the accident
(13 years).
Conclusion
This study meets all of the criteria and considerations for further dose-response analysis. The
determination of the relevant time interval over which TCDD dose should be considered is
uncertain.
Table B-29. Eskenazi et al., 2002b—Endometriosis
1. Consideration
(
Vlethods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
(
Consideration not satisfied. Results of a pilot study showed that ultrasounds had excellent
specificity and sensitivity for ovarian endometriosis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-37 DRAFT—DO NOT CITE OR QUOTE
-------
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design or
statistical analysis.
Response
Consideration not satisfied. More than half of the women were classified as 'uncertain' with
respect to endometriosis disease status.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of an
exposure-response relationship.
Response
Consideration not satisfied. While an increased risk of endometriosis was observed across the 3
TCDD categories, these risks were not statistically significant relative to the lowest exposure
category. The test for trend based on a continuous measure (logi0TCDD) was also not statistically
significant.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Criteria satisfied. Serum concentrations of TCDD offer improved exposure assessment, although
delineating the critical exposure window is challenging given the nature of the very high initial
exposure.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure adequate
statistical power.
Response
Consideration not satisfied. Only a total of 19 cases of endometriosis were identified.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion of
the strengths and limitations.
Response
Criteria satisfied. Environ Health Perspect 2002; 110(7)629-634. Author's highlight that this is
the first study to examine the relationship between TCDD and endometriosis, and the availability
of sera data to estimate TCDD levels. Limitations included the small number of women with
endometriosis, and inability to confirm disease status using laparoscopy. Finally, young women
may have been underrepresented due to cultural difficulties in examining women who had never
been sexually active.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response relationships
can be assessed.
Response
Criteria satisfied. Eligible study subjects were women between 1 month and 40 years of age at
time of accident. These analyses excluded virgins, those with Turner's syndrome, and women
who refused the examination of ultrasound. Serum data were available for the 601 participants on
which the analyses are based. Of these, 559 had serum measures taken in 1976/77, 25 between
1978 and 1981, and 17 women in 1996.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of exposure
are consistent with the current biological understanding of dose. The reported dose is consistent
with a toxicologically relevant dose. Latency and appropriate window(s) of exposure examined.
Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. TCDD exposure was estimated at the time of "conception attempt" using
serum measures, with extrapolation from 1976 levels using half-life assumptions. It is difficult to
identify the relevant time interval over which TCDD dose should be considered for dose-response
analysis. The critical window of exposure is unknown.
Conclusion
The lack of a statistically significant association coupled with a large number of women for which
endometriosis disease status was "uncertain", precludes the use of these data to conduct dose-
response analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-38 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table B-30. Eskenazi et al., 2003—Birth outcomes
2
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration not satisfied. Outcomes were identified through self-reported questionnaires.
Women were found to over-report birth weight, and have a tendency to underreport birth
defects in children. As a large number of women in Seveso underwent voluntary abortion in
the first year after the explosion, an awareness bias may have contributed to differential
reporting of pregnancy histories.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. See above.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. There was no association between spontaneous abortions and
logioTCDD, or with births small for gestational age. An inverse association with birth weight
was noted in first eight years following the accident as were the number of births small for
gestational age; however, none achieved statistical significance atp< 0.05.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Criteria satisfied. Serum concentrations of TCDD offer improved exposure assessment,
although delineating the critical exposure window is challenging given the nature of the very
high initial exposure.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For spontaneous abortions there were 769 pregnancies. Fetal growth
and gestational age analysis was carried out on 608 singleton births that occurred post-
explosion.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Environ Health Perspect, 2003, lll(7):947-953. The authors highlight
potential limitation of reliance on self-reported data to ascertain pregnancy outcomes. They
also address the relevance of paternal exposures to TCDD on the developing fetus—such
exposure data were not considered in this study.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. A total of 745 women in the SWHS had reported getting pregnant, of these
510 women were pregnant after the explosion (888 pregnancies). Analyses of spontaneous
abortions based on 476 women (excludes those with voluntary abortion, ectopic pregnancy, or
molar pregnancy. TCDD measured for 413 women in 1976/77, 12 women between 1978 and
1981, and 1996 for 19 women.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-39 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria not satisfied. TCDD exposures were extrapolated to 1976 values. However, it is
difficult to identify the relevant time interval over which TCDD dose should be considered for
dose-response analysis.
Conclusion
The findings of the study are somewhat limited due to the reliance on self-reported information
for pregnancy outcomes, and lack of paternal exposures. The findings were not statistically
significant. Taken together, quantitative dose-response analyses for this study population is
not recommended.
Table B-31. Warner et al., 2004—Age at menarche
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. In this study age at menarche was based on retrospective recall 5 to
19 years before the interview. Previous work suggests moderate to high correlations between
actual and recalled menarche, misclassification of outcome would bias risk estimates towards
the null (assuming nondifferential misclassification.)
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Data collected from self-reported questionnaires allow for the
potential confounding influence of many risk factors to be taken into account. Some
misclassification of outcome may bias risk estimates towards the null.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. There was no association between TCDD levels and the age at
menarche with either the continuous or categorical measures of TCDD.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Criteria satisfied. Serum concentrations of TCDD offer improved exposure assessment,
although delineating the critical exposure window is challenging given the nature of the very
high initial exposure.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Cohort was large enough as analyses were performed using 282
women who were premenarcheal at the time of the explosion.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Environ Health Perspect, 2004, 112:1289-1292. Authors discuss use of
pooled serum from residents of the unexposed zone, and that those in lowest exposure group
had high exposures relative with contemporary levels for the area. Strengths of study include
use of serum to estimate TCDD exposure.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-40 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria satisfied. The SWHS included women between 1 month and 40 years of age at time of
accident who attempted to get pregnant after the explosion (n = 463). This study is restricted
to those who were premenarcheal at the time of the explosion (n = 282). Serum was collected
for these women, primarily in 1976-1977 (n = 257), between 1978 and 1981 for 23, and in
1996-1997 for the 2 remaining women.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. TCDD exposures in 1976 were estimated by extrapolation serum levels
obtained after this date using the Filser model. Both categorical and continuous measures of
exposure were modeled. In utero measures of exposure are likely most relevant exposure
based on findings from animal studies.
Conclusion
No association between TCDD levels and age at menarche was found. There may be some
misclassification of age at menarche based on self-report, and biologically, the most relevant
dose as suggested by animal studies occurs in utero. Additionally, it is difficult to identify the
relevant time interval over which TCDD dose should be considered for dose-response analysis.
For these reasons, these data are not suited to a dose-response analysis.
Table B-32. Eskenazi et al., 2005—Age at menopause
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Outcome measures were obtained based on self-reported data
collected from questionnaires. Studies have shown that self-reports of age at menopause are
reported with accuracy and reliability, and among women with surgical menopause, the self-
reported age correlated well with that on the medical records.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Data obtained from the questionnaire allow for the potential
confounding influence of several potential confounders to be controlled for.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. Although risks of earlier menopause increased in the first four
quintiles, with a statistically significant trend, no increased risk was noted in the highest
exposure category (hazard ratio = 1.0 relative to lowest exposure group). Study authors
suggest this is due to the "inverted U" dose response often seen with hormonally active
compounds. Additionally, no statistically significant association was noted with logi0TCDD
for the individual quintiles.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Criteria satisfied. Serum concentrations of TCDD offer improved exposure assessment,
although delineating the critical exposure window is challenging given the nature of the very
high initial exposure.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-41 DRAFT—DO NOT CITE OR QUOTE
-------
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. The study included 616 women. Of these, 260 were premenopausal,
169 classified as natural menopause, 83 as surgical menopause, 24 as impending menopause,
33 as premenopausal, and 58 in an "othef' category.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Environ Health Perspect, 113:858-862 (2005). Authors highlight this is
first study to look at relationship between dioxin and age at menopause. Other limitations of
the study include lowest exposure group (< 20.4 ppt) includes exposures level that are far
higher than background, and age at menopause was based on retrospective recall. Strength of
study is ability to characterize TCDD using serum measures.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. The Seveso Women's Health Study collected serum sample which allowed
TCDD exposures to be characterized. Those women (n = 616) who had not reached natural
menopause at the time of the accident were included in the study. Serum measures collected
in 1976/77 were available for 564 women, for 28 women, sera was collected between 1978
and 1981, while for 24 women, sera was collected in 1996/97.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. TCDD levels were estimated at the time of the explosion using available
information on TCDD half-life. However, it is difficult to identify the relevant time interval
over which TCDD dose should be considered for dose-response analysis. The critical window
of exposure can be estimated but is large and highly uncertain.
Conclusion
The findings do not provide strong support for a dose-response relationship. As such, they are
not well suited to a quantitative dose-response analysis.
Table B-33. Warner et al., 2007—Ovarian function
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Ovarian cyst analysis based on women who underwent ultrasound
(n = 310). Ovarian follicle analysis based on self-report on menstrual cycle and done in
women in pre-ovulatory cycle (n = 96) at time of ultrasound. Hormonal analysis based on
women in last 14 days of cycle (n = 129).
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Data collected from self-reported questionnaires allow for the
potential confounding influence of many risk factors to be taken into account. Some
misclassification of outcome based on self-reports of menstrual cycle may bias risk estimates
towards the null.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-42 DRAFT—DO NOT CITE OR QUOTE
-------
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. There was no association between serum TCDD levels and the
number or size of ovarian follicles. TCDD was also not associated wit the odds of ovulation.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Criteria satisfied. Serum concentrations of TCDD offer improved exposure assessment,
although delineating the critical exposure window is challenging given the nature of the very
high initial exposure.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Cohort was large enough as analyses were performed using 129
women for ovulation outcome, and hormone analyses based on 87 women in luteal, and 55 in
midluteal phases.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Environ Health Perspect, 2007,115:336-340. An important limitation cited
by the authors was that women may not have been exposed at critical period (prenatally).
Phases of the cycle may also have been misclassified as this was based on self-reported data.
Strength, first study to have examined ovarian function and TCDD exposures.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. The SWHS included women between 1 month and 40 years of age at time of
accident who were between 20-40 years of age and not using oral contraceptives at follow-up
(n = 363).Of these, serum was collected for 330 women between 1976 and 1977, between
1978 and 1982 for 25 women, and between 1996 and 1997 for 8 women.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. The women may not have been exposed at critical period (prenatally).
Conclusion
No association between TCDD levels and ovarian function was found. There may be some
misclassification of period of the cycle based on self-report, and biologically, the most relevant
dose as suggested by animal studies occurs in utero. For these reasons, these data are not
suited to a dose-response analysis.
Table B-34. Eskenazi et al., 2007—Uterine leiomyoma
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-43 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. Outcomes were determined using two definitions: current fibroids, or
past diagnosis of fibroids. For past diagnosis of fibroids, self-reported data and medical
records were used to determine whether women were previously diagnosed with fibroids, these
were confirmed with medical records. A total of 25 women indicated they had never been
diagnosed with fibroids. Medical records indicate a past diagnosis for these women, and they
were classified as such. For current fibroids, this was determined at the time of the interview
for 634 women using transvaginal ultrasound examinations.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. In the SWHS questionnaires were administered to the participants and
detailed data for reproductive characteristics, smoking, body mass index, and alcohol use were
collected so risks could readily be adjusted for these covariates.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied, but inversely. An inverse dose-response pattern with the percentage of
women diagnosed (current & past history-combined) with fibroids across 3 categories of
exposure. Namely, the percentages of women with fibroids in the <20, 20.1-75.0, and >75.0
ppt categories were 41.1%, 26.8%, and 20.0%, respectively.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. A variety of different exposure metrics were considered including
linear, categorical, splines, and logi0TCDD.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. A total of 251 women were found to have fibroids, and there were 62,
110, and 79 women with fibroids diagnosed in the 3 TCDD exposure categories.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 2007, 166:79-87. In this study, the authors found an
inverse association between TCDD and uterine leiomyoma risk. The authors highlighted
strengths of the study that included the longitudinal design, serum measures taken at an
individual-level basis and most taken within 2 years of the accident, ability to include
outcomes among those who did not take an ultrasound by using an adapted statistical
approach. An important limitation that was the differences in risk by the stage of development
could not be assessed as all women were exposed postnatally, and only 4 cases were observed
among those who were premenarcheal at the time of exposure.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Final sample consisted of 956 women in the Seveso Women's Health Study
without a history of fibroids. For 872 of these women, serum was collected in 1976 and 1977.
For 56 women, TCDD was measured in women between 1978 and 1981, and for 28 women
the serum was collected in 1996.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-44 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria not satisfied. TCDD exposures were back extrapolated to expected levels in 1976 (at
the time of the accident). However, it is difficult to identify the relevant time interval over
which TCDD dose should be considered for dose-response analysis. The critical window of
exposure is unknown.
Conclusion
The data suggest an inverse (protective) effect between fibroids and exposure to TCDD. As
such, these data are not suited to further dose-response analyses.
B.2.5. Other Seveso Noncancer Studies
Table B-35. Mocarelli et al., 2008—Semen quality
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Serum levels of TCDD were measured on an individual basis for men
in exposed areas; pooled samples from men in uncontaminated areas were measured to assess
background TCDD exposure levels.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. While compliance rates may have introduced some possible bias, this
does not seem likely as different effects noted between the 22-31 and 32-39 year old age
groups. Information collected for other risks factors, which have been used as adjustment
factors in the models.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Figure 3 suggests dose-response relationship among those aged 1-9 at
the time of the accident for sperm concentration and motility.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Serum concentrations of TCDD offer improved exposure assessment,
although delineating the critical exposure window is challenging.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Analyses are based on 135 males exposed to TCDD.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Environmental Health Perspective s, 2008, 116(l):70-77. The authors
describe strengths associated with characterization of exposure (using serum samples), and
representativeness of study population. Limitation of study includes low compliance (but high
for semen sample studies), namely, 60% among a group of healthy men. The compliance rate
was higher among exposed group (69%).
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-45 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria satisfied. Involved males, < 16 years old at time of accident.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria satisfied. TCDD exposures were based on serum samples. Serum samples were
drawn (in 1997/1998) from participants whose 1976 samples were above 15 ppt. Pooled
samples obtained in 1997/98 were used to describe background TCDD levels in
uncontaminated areas. The associated between TCDD exposure and semen quality was found
statistically significant for the boys with 1 and 9 years of age at the time of the accident. This
provides a critical window of exposure to estimate TCDD concentration.
Conclusion
Health outcomes are exposures are well characterized using serum data. However, the men
exposed between the ages of 1 and 9 to elevated TCDD levels had reduced semen quality 22
years later. It is difficult to discern whether this effect is a consequence of the initial high
exposure between 1 and 9 years of age or a function of the cumulative exposure for this entire
exposure window beginning at the early age. Nonetheless, quantitative dose-response analyses
for this outcome were conducted.
Table B-36. Mocarelli et al., 2000—Sex ratio
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Birth records examined for those who lived in parents who lived in
the area and who provided serum samples.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Paternal TCDD exposures were associated with an increased
probability of female births (p = 0.008).
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Serum samples were used to estimate maternal and paternal TCDD
levels. No discussion of exposure levels in reference population.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Statistically significant findings achieved.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria not satisfied. The Lancet, 2000, 355:1858-1863. There is no discussion on the
strengths and limitations of this study.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-46 DRAFT—DO NOT CITE OR QUOTE
-------
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum levels of TCDD were obtained from parents using samples provided
in 1976/77. Serum measures available for 296 mothers and 239 fathers.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. Serum based measures of TCDD were obtained shortly after the
accident. TCDD levels were also extrapolated to the time of conception. However, it is
difficult to identify the relevant time interval over which TCDD dose should be considered for
dose-response analysis. The critical window of exposure is unknown.
Conclusion
The data from this study demonstrate a positive dose-response relationship with paternal
TCDD levels at the time of the accident and increased likelihood for female births. However,
It is difficult to identify the relevant time interval over which TCDD dose should be
considered; specifically, it is difficult to discern whether this effect is a consequence of the
initial high exposure during childhood or a function of the cumulative exposure for this entire
exposure window beginning at the early age. Using the initial exposures in a dose-response
model would yield LOAELs that are too high to be relevant to factor into the RfD calculation.
Dose-response analysis for this outcome is, therefore, was not conducted.
Table B-37. Baccarelli et al., 2008—Neonatal thyroid function
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Measures of b-TSH are taken using a standardized protocol 72 hours
after birth. These b-TSH measures are taken on all newborns born in the region of Lombardy of
which Seveso if a part of.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design or
statistical analysis.
Response
Consideration satisfied for component of the study based on plasma dioxin measures. For the
comparisons involving place of residence at the time of the accident, exposure misclassification is
likely given variability in soil TCDD exposure levels within these areas.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of an
exposure-response relationship.
Response
Consideration satisfied. Mean neonatal b-TSH was 0.98|iU/ml [0.90-1.08] in the reference area,
1,35|iU/ml [1.22-1.49] in zone B, and 1,66|iU/ml [1.19-2.31] in zone A (p < 0.001). The plotted
frequency distributions have similar shapes, but have shifted to the right for areas of higher
exposures. Neonatal b-TSH was correlated with current maternal plasma TCDD ((3-0.47, p <
0.001) in the 51 newborns for which individual maternal serum TCDD values were available.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-47 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. TEQs were measured among the 38 women for which serum samples
were available and were defined for a mixture of dioxin-like compounds. Maternal mean total
TEQs (PCDDs, PCDFs, coplanar PCBs, and noncoplanar PCBs) was 41.8 ppt. Two measures of
exposure included place of residence at time of accident and plasma samples obtained from
mothers at the time of delivery. Similarities in positive dose-response relationships give stronger
weight to the findings.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure adequate
statistical power.
Response
Consideration satisfied for exposure metric that was based on 'place of residence'. For plasma
based estimate of maternal TCDD there were only 51 mother-child pairs. Only seven children in
total were found to have b-TSH levels in excess of 5 uU/ml; this implies limited statistical power
involving this health outcome.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion of
the strengths and limitations.
Response
Criteria satisfied. PLOS Medicine 2008; 5(7)1133-1142. The authors discuss the strength of the
study related to characterization of exposure using serum sampling, and ability to adjust for
factors related to b-TSH or TCDD levels (gender, birth weight, birth order, maternal age, hospital
and type of delivery. They also highlight that a limitation of study was that the influence of
mother-child dioxin transfer through colostrum could not be assessed because no information on
breastfeeding before b-TSH measurement was available.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response relationships
can be assessed.
Response
Criteria satisfied. In the population-based study, eligible women who resided in zones A and B at
the time of the accident (n = 1,772) were matched to nonexposed women. In the study based on
plasma dioxin measurements, participants were the 51 children born to 38 women from zones A,
B, R, or a reference zone for which plasma dioxin measurements were available.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of exposure
are consistent with the current biological understanding of dose. The reported dose is consistent
with a toxicologically relevant dose. Latency and appropriate window(s) of exposure examined.
Response has to be a nonfatal endpoint.
Response
Criteria satisfied. Maternal TCDD levels were estimated at the time of delivery based on plasma
samples, and the critical window of exposure can be defined as the 9 month gestation period.
Conclusion
The data provide an opportunity for quantitative dose-response analyses.
1
2
3
4
Table B-38. Alaluusua et al., 2004—Oral hygiene
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Ascertainment of dental health was done blind to place of residence,
used standard protocol for caries developed by the WHO, and the clinical examination
supplemented by radiographic examination.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-48 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. Additional risk factor information was collected on questionnaires.
These factors were considered as adjustment factors. Findings potentially susceptible to
participation biases.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Increased prevalence of developmental enamel effects found with
increased TCDD serum measures. Namely, prevalence in unexposed region was 26%,
whereas in the low, middle, and high TCCD groups the prevalence was 10, 40, and 60%,
respectively.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. TCDD exposure level based on serum lipids. No discussion of
exposure levels in reference population.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Criteria satisfied. Despite small numbers, statistically significant findings were achieved.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Environmental Health Perspectives, 2004, 112(13)1313-1318. Authors
mention two important strength of the study: characterization of TCDD exposure using serum
collected shortly after the time of the accident, and the fact that developmental defects are
permanent in nature. Therefore, they represent a health outcome can evaluated years later.
Little discussion was made of the impact of differential compliance rates between the exposed
(74%) and nonexposed (58%) groups. Authors mention two important strength of the study:
characterization of TCDD exposure using serum collected shortly after the time of the
accident, and the fact that developmental defects are permanent in nature. Therefore, they
represent a health outcome can evaluated years later. Little discussion was made of the impact
of differential compliance rates between the exposed (74%) and nonexposed (58%) groups.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum levels of TCDD could be estimated for children in exposed areas.
No serum levels were available for reference group of children, and assumption of zero
exposure was made. This seems reasonable.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria satisfied. It is difficult to discern whether this effect is a consequence of the initial
high exposure during childhood or a function of the cumulative exposure of the entire
exposure window beginning at early age. However, assumptions can be made regarding the
critical window of exposure and the relevant dose can be calculated.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-49 DRAFT—DO NOT CITE OR QUOTE
-------
Conclusion
The considerations for conducting a dose-response analysis have been satisfied with the study
population of only those subjects who lived in the ABR zone at the time of the accident;
exposure data are unavailable for those in the referent area. While is difficult to identify the
relevant time interval over which TCDD dose should be considered, quantitative dose-
response analysis for this outcome was conducted.
Table B-39. Bertazzi et al., 2001—Mortality (Noncancer)
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied for some causes of death, but not others. Mortality appears to be well
captured from the vital statistics registries in the region (99% complete). Some health
outcomes (e.g., diabetes) are subject to misclassification using death certificate data.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Although individual-level data for individual risk factors are not
available, the potential for confounding is likely minimal. For e.g., independent surveys
suggests similarity between smoking behaviors across the regions. Exposure misclassification
based on place of residency likely to bias risk estimates towards the null.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied for most causes of death. An exception was the dose-response
relationship was observed for chronic obstructive pulmonary disease across Zones A, and B.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. Exposure classification was based on the address of the
residence on the date of the accident or when the person first entered the area. Although
TCDD blood levels were also measured, these were not examined with respect to health
outcomes. The lack of individual-level data also precluded an examination of these
uncertainties.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. A total of 494 non-cancer deaths were found among residents of
Zones A, and B, respectively. This allowed examined of gender-specific effects.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 2001, 153:1031-1044. Authors discuss lack of individual-
level exposure data and other risk factors (e.g., smoking), difficulties in extrapolating to
background levels, diagnostic accuracy of using death certificates. Strengths included
similarities between exposed and comparison population for several risk factors, completeness
of follow-up, and consistent methods to identify mortality outcomes in the exposed and
comparison populations.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-50 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria not satisfied. Individual-level exposure data are unavailable. Exposure based on place
of residence at time of the explosion. Soil sampling performed indicated considerable
variability in TCDD levels within each region. In addition, place of residency at time of
explosion does not ensure individuals were at their home around the time of the accident.
3. Critieria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. An ecological measure of exposure (region of residency at time of
accident) was used to categorize individuals according to their possible exposure. Latencies
were considered. While such an approach has value for identifying whether excesses occurred
among highly exposed populations, it is not precise enough to conduct a quantitative dose-
response analysis. Furthermore, noncancer mortality is not a viable endpoint to consider for
further dose-response analysis.
Conclusion
Study is not suitable for dose-response analysis due to mortality as endpoint and lack of
individual-level exposure data.
Table B-40. Consonni et al., 2008—Mortality (Noncancer)
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied for some causes of death, but not others. Mortality appears to be well
captured from the vital statistics registries in the region (99% complete). Some health
outcomes (e.g., diabetes) are subject to misclassification using death certificate data.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Although individual-level data for individual risk factors are not
available, the potential for confounding is likely minimal. For e.g., information from other
independent surveys suggests similarity between smoking behaviors across the regions.
Exposure misclassification based on place of residency is likely to bias risk estimates towards
the null.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. Statistically significant association noted in most highly exposed
area for chronic rheumatic disease and chronic obstructive pulmonary disease. Dose-response
pattern noted across Zones A, B and R for circulatory disease mortality 5-9 years after the
accident.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. Lack of individual-level data precludes an examination of these
uncertainties.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-51 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied for some causes of death but not others. For example, only three
deaths from diabetes occurred among residents of Zone A. The limitation related to statistical
power is exacerbated for stratified analyses carried out by number of years since the accident.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Am J Epidemiol, 2008, 167:847-858. Authors discuss potential for selection
bias, limitation of residential based measure of exposure, similarities of mortality
ascertainment in exposed and referent populations, and multiple testing.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. Individual-level exposure data are unavailable. Exposure based on place
of residence at time of the explosion. Soil sampling performed indicated considerable
variability in TCDD levels within each region. In addition, place of residency at time of
explosion does not ensure individuals were at their home around the time of the accident.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. An ecological measure of exposure (region of residency at time of
accident) was used to categorize individuals according to their possible exposure. Latencies
were considered. While such an approach has value for identifying whether excesses occurred
among highly exposed populations, it is not precise enough to conduct a quantitative dose-
response analysis. Furthermore, noncancer mortality is not a viable endpoint to consider for
further dose-response analysis.
Conclusion
Study is not suitable further dose-response evaluation due to noncancer morality endpoint.
Table B-41. Baccarelli et al., 2005—Chloracne
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Chloracne cases identified using standardized criteria.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Plasma TCDD was associated with an increased risk of chloracne.
The odds ratios increased in a dose-response pattern across zone of residence.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Authors discussed implications of differential elimination rates by age
and body growth.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-52 DRAFT—DO NOT CITE OR QUOTE
-------
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. A total of 101 chloracne cases were identified, and 211 controls were
selected. Statistically significant findings were observed in several comparisons.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. British Journal of Dermatology, 2005, 152,459-465. The authors detail the
limited statistical power they had available in the study. They also highlight a strength of the
study that included uniqueness of age and sex distribution of chloracne cases, characterization
of TCDD that could be done using sera samples, and availability of both clinical and
epidemiological data.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. TCDD was estimated in both chloracne cases and control using serum
measures.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria satisfied. Serum based measures of TCDD were obtained shortly after the accident.
Chloracne is thought to be caused by the initial high exposure.
Conclusion
Exposure to TCDD at sufficiently high levels is recognized to cause chloracne. This study
provides limited relevance to dose-response modeling of TCDD as exposure levels typically
observed in the general population are much lower.
Table B-42. Baccarelli et al, 2002 and 2004—Immunological effects
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Common methods were used to describe blood levels of plasma
immunoglobulins (IgA, IgG, and IgM) and complement components (C3 and C4).
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Both exposure and outcome were objectively and accurately measured.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration satisfied. Plasma IgG levels were inversely related with TCDD.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Both categorical (quintiles) and continuous measures of TCDD were
examined in the dose-response analysis.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-53 DRAFT—DO NOT CITE OR QUOTE
-------
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Analyses are made using 72 highly exposed, and 72 low exposed
individuals.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Toxicology letters, 2004, 149:287-293 and Environ Health Perspect, 2002,
110(12): 1169-1173. The authors highlight that few studies have looked at immunological
effects of TCDD in humans, that the current study was able to exclude those with concurrent
medical conditions, and the ability to characterize exposure using serum measures. Limitations
addressed were the uncertainty about the clinical relevance of the dose-response pattern found,
and the relatively small size of the study population.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. A total of 120 subjects were included in the study. This included 62
randomly selected from the high exposed zone, and 58 selected from the reference area.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of exposure
are consistent with the current biological understanding of dose. The reported dose is
consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. Dose-response relationships were examined using current TCDD levels.
However, it is difficult to identify the relevant time interval over which TCDD dose should be
considered for dose-response analysis.
Conclusion
An inverse dose-response association between IgG and TCDD was observed, however, because
the relationship can not be described in terms of clinical relevance with respect to a specific
health outcome, it is our view that these data are not suited to dose-response modeling.
B.2.6. Chapaevsk Study
Table B-43. Revich et al., 2001—Mortality (noncancer) and reproductive
health
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration cannot be evaluated. Insufficient details are provided in the paper to gauge the
completeness and coverage of the cancer registry and mortality data. Health outcomes were
studied on the basis of information in the official medical statistics
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. It is an ecological study.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-54 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration cannot be evaluated. Dose-response was not evaluated as exposure was based on
residency in the region vs. no residency.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. No individual-level exposure estimates were used.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. Population-based data over several years were used to make ecological
comparisons.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Published in Chemosphere, 2001, 43(4-7):951-966.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. It is a cross-sectional study that compares mortality rates between
regions. No individual-level exposure data available.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of exposure
are consistent with the current biological understanding of dose. The reported dose is
consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. No exposure estimates were used in the study.
Conclusion
These cancer data are cross-sectional in nature and not appropriate for a dose-response analysis.
B.2.7. Air Force Health ("Ranch Hands") Study
Table B-44. Michalek and Pavuk, 2008—Diabetes
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. Prevalent diabetes identified from medical records from repeated
medical check-ups. Preferred method of ascertaining outcome relative to use of death
certificates.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. Adjustment was made for a number of risk factors related to
diabetes (e.g., BMI, family history, smoking). However, Agent Orange was a 50% mixture of
2,4-D and TCDD; therefore, potential for confounding by other coexposures is likely.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-55 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. The RR for an increase in 10 units was 1.29 (p < 0.001), and the risks
across the background, low and high exposure categories, relative to the unexposed were 0.86,
1.45, and 1.68.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Initial TCDD dose were estimated at the end of the tour of duty for
the Ranch Hands. Individual-level serum dioxin measurements correlated well with correlated
with days of spraying and calendar period of service, but collection of the samples roughly 20
years later required back-extrapolation.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. There were a total of 439 cases of diabetes identified.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. J Occup Environ Medicine, 2008, 50:330-340. The authors address
strengths and limitations related to the accuracy of the one-compartment pharmacokinetic
model, impact of the covariate time spent in Southeast Asia, and potential exposure
misclassification on days sprayed.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. TCDD estimates were derived using serum samples. However, Ranch
Hand veterans were exposed to other compounds in the herbicides, such as 2,4-D.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria satisfied. TCDD levels at the end of service were estimated. Extrapolation was done
using a half-life of 7.6 years. Exposures were grouped into comparison, background, low and
high. This allows for a shape of the dose-response curve to be evaluated. A continuous
measure of TCDD was also examined (logi0TCDD).
Conclusion
Ranch Hand veterans were exposed to other contaminants in the herbicides that were mixed,
thereby making it difficult to determine independent effects of TCDD on diabetes. In our
view, this limitation precludes dose-response modeling of TCDD and diabetes using data from
this cohort.
B.2.8. Other Noncancer Studies of Dioxin
Table B-45. McBride et al., 2009a—Mortality (Noncancer)
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-56 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Consideration satisfied. The New Zealand Health Information Service Mortality Collection
and the Registrar-General's Index to Deaths were used to identify deaths. Additional searches
were based on the last known address from the work record; the electoral roll and the
habitation index; the telephone book; the internet; and Terranet property information database.
An additional search was carried out through the Births, Deaths, and Marriages office of the
New Zealand Department of Internal Affairs. Lastly, automated personnel and pension
records were also used to locate past New Plymouth workers and identify some deaths.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration satisfied. Workers lost to follow-up were an unlikely source of bias especially
for internal analyses. Confounding by other coexposures (e.g., 2,4,6-TCP) unlikely to have
resulted in bias, due to presumed poor correlation with TCDD.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. There was no cause of death among those considered for which a
dose-response trend was observed across four exposure categories of TCDD.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Dichotomous exposure (exposed/unexposed) and duration of
employment were examined from job exposure classification assessed via occupational history
records industrial hygienists/factory personnel knowledge and questionnaires.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration not satisfied.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Published in J Occup Environ Med, 2009, 51:1049-1056. The other studies
in the cohort highlight the 22% of the cohort lost to follow-up, the limited size of the cohort
tissue sarcomas, differences in cohort definitions between sprayers and producers, and the
potential for other exposures during employment at the plant.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria satisfied. Serum measures available for 346 workers were used to derive TCDD
exposures for the entire cohort using the area under the curve approach.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. Dichotomous exposure assessment did not allow individual estimates of
dose to be developed. However, noncancer mortality is not a viable endpoint to consider for
further dose-response analysis.
Conclusion
A considerable portion of the cohort was lost to follow-up, and no dose-response associations
noted. As a result, the data are not suited to dose-response analysis.
1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-57 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table B-46. McBride et al., 2009b—Mortality (noncancer)
2
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration satisfied. The New Zealand Health Information Service Mortality Collection
and the Registrar-General's Index to Deaths were used to identify deaths. Additional searches
were based on the last known address from the work record; the electoral roll and the
habitation index; the telephone book; the internet; and Terranet property information database.
An additional search was carried out through the Births, Deaths, and Marriages office of the
New Zealand Department of Internal Affairs. Lastly, automated personnel and pension
records were also used to locate past New Plymouth workers and identify some deaths.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. Considerable amount of workers were lost to follow up (22%),
but it is unclear if bias resulted. The dichotomous exposure measure was based on exposure
to TCDD, chlorinated dioxins and phenoxy herbicides, so confounding is a possibility by
these coexposures.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. Because no individual exposure estimates were available for
these analyses, dose-response could not be evaluated.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration satisfied. Consideration satisfied. Dichotomous exposure
(exposed/unexposed) and duration of employment were examined from job exposure
classification assessed via occupational history records industrial hygienists/factory personnel
knowledge and questionnaires. Authors discuss limitations in the assignment of exposure
among cohort members.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria satisfied. Published in Occup Medicine, 2009, 59(4):255-263. The authors highlight
cohort lost to follow-up, the limited size of the cohort, differences in cohort definitions
between sprayers and producers, and the potential for other exposures during employment at
the plant.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
Response
Criteria not satisfied. Exposures were not quantified. The dichotomous exposure measure
was based on exposure to TCDD, chlorinated dioxins and phenoxy herbicides.
3. Critiera
The effective dose and oral exposure can be reasonably estimated and the measures of
exposure are consistent with the current biological understanding of dose. The reported dose
is consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-58 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Effective dose could not be estimated given the lack of individual-level exposure data.
Noncancer mortality is not a viable endpoint to consider for further dose-response analysis.
Conclusion
The study lacks the quantification of exposures at an individual level, and a considerable
portion of the cohort was lost to follow-up. As a result, the data are not suited to dose-
response analysis.
Table B-47. Ryan et al., 2002—Sex ratio
1. Consideration
Methods used to ascertain health outcomes identified were unbiased, highly sensitive, and
specific.
Response
Consideration not satisfied. Company records were used to identify births, the date of birth,
and the sex of the child. No information was provided on the expected completeness of
identifying births in this manner. Moreover, the study was expanded to include workers who
heard about the study in a public forum. Therefore, the study could be influenced by
participation bias.
2. Consideration
Risk estimates are not susceptible to biases from confounding exposures or from study design
or statistical analysis.
Response
Consideration not satisfied. See above.
3. Consideration
Study demonstrates an association between TCDD and adverse health effect with evidence of
an exposure-response relationship.
Response
Consideration not satisfied. The study compared birth ratios among men and women employed
at the plant to the general population. No categories of exposure were examined.
4. Consideration
Exposure assessment methodology is clear and adequately characterizes individual-level
exposures. The limitations and uncertainties in the exposure assessment are considered.
Response
Consideration not satisfied. This is not relevant as no analyses were done in relation to
exposure levels.
5. Consideration
Study size and follow-up are large enough to yield precise estimates of risk and ensure
adequate statistical power.
Response
Consideration satisfied. For the categories of exposure used (yes/no), and the stratified
analyses by sex and subcohort, the study allows for the birth ratios to be estimated with
sufficient precision.
1. Criteria
Study is published in the peer-reviewed scientific literature and has an appropriate discussion
of the strengths and limitations.
Response
Criteria not satisfied. Published in Environ Health Perspect, 2002, 110(11):A699-A701. The
authors discussed the limitations of using serum collected many years after they stopped
working to estimate TCDD exposures when the preferred metric would be TCDD levels at the
time of conception. They did not address issues about the representativeness of the study
participants to the entire cohort of workers, nor did they address the limitation of not being able
to conduct dose-response analyses using individual-level TCDD data.
2. Criteria
Exposure must be primarily TCDD and is properly quantified so that dose-response
relationships can be assessed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-59 DRAFT—DO NOT CITE OR QUOTE
-------
Response
Criteria not satisfied. While serum measures were available for 84 of the 198 participants of
the study, birth ratios were compared between the cohort of 2,4,5-T and 2,4,5-trichlorphgenol
workers relative to the city of Ufa. There was no attempt to derive birth ratios in relation to
exposure levels. The serum data were only used to demonstrate that these workers, on average,
had TCDD levels 30 times higher than Ufa residents.
3. Criteria
The effective dose and oral exposure can be reasonably estimated and the measures of exposure
are consistent with the current biological understanding of dose. The reported dose is
consistent with a toxicologically relevant dose. Latency and appropriate window(s) of
exposure examined. Response has to be a nonfatal endpoint.
Response
Criteria not satisfied. TCDD exposures were based on serum measures taken in some cases
many years after children were born; no attempt was made to back-extrapolate to the time of
conception.
Conclusion
The data are not suitable for dose-response modeling. Risk estimates have not been derived in
relation to TCDD exposure levels. There exist uncertainties about the representativeness of the
participants in relation to the cohort as a whole, and insufficient details are provided to evaluate
the extent in which all births were identified. While these data should not be used for
quantitative dose-response modeling, the much lower M/F birth ratio among exposed fathers is
consistent with the finding by Mocarelli et al, and lends support to those findings.
1
2
3 B.3. REFERENCES
4 Akhtar, FZ; Garabrant, DH; Ketchum, NS; et al. (2004) Cancer in US Air Force veterans of the Vietnam War. J
5 Occup Environ Med 46(2): 123-136.
6 Alaluusua, S; Calderara, P; Gerthoux, PM; et al. (2004) Developmental dental aberrations after the dioxin accident
7 inSeveso. Environ Health Perspect 112(13): 1313-1318.
Aylward, LL; Brunet, RC; Starr, TB; et al. (2005a) Exposure reconstruction for the TCDD-exposed NIOSH cohort
using a concentration- and age-dependent model of elimination. Risk Anal 25(4):945-956.
Aylward, LL; Brunet, RC; Carrier, G; et al. (2005b) Concentration-dependent TCDD elimination kinetics in
humans: toxicokinetic modeling for moderately to highly exposed adults from Seveso, Italy, and Vienna, Austria,
and impact on dose estimates for the NIOSH cohort. J Expo Anal Environ Epidemiol 15(1):51-65.
8 Baccarelli, A; Mocarelli, P; Patterson, DG, Jr.; et al. (2002) Immunologic effects of dioxin: new results from Seveso
9 and comparison with other studies. Environ Health Perspect 110(12): 1169—1173.
10 Baccarelli, A; Pesatori, AC; Masten, SA; et al. (2004) Aryl-hydrocarbon receptor-dependent pathway and toxic
11 effects of TCDD in humans: a population-based study in Seveso, Italy. Toxicol Lett 149(l-3):287-293.
12 Baccarelli, A; Pesatori, AC; Consonni, D; et al. (2005) Health status and plasma dioxin levels in chloracne cases 20
13 years after the Seveso, Italy accident. Br J Dermatol 152(3):459-465.
14 Baccarelli, A; Hirt, C; Pesatori, AC; et al. (2006) t(14; 18) translocations in lymphocytes of healthy dioxin-exposed
15 individuals from Seveso, Italy. Carcinogenesis 27(10):2001-2007.
16 Baccarelli, A; Giacomini, SM; Corbetta, C; et al. (2008) Neonatal thyroid function in Seveso 25 years after maternal
17 exposure to dioxin. PLoS Med 5(7): 1133-1142.
18 Becher, H; Steindorf, K; Flesch-Janys, D. (1998) Quantitative cancer risk assessment for dioxins using an
19 occupational cohort. Environ Health Perspect 106(Suppl 2):663-670.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-60 DRAFT—DO NOT CITE OR QUOTE
-------
1 Bertazzi, PA; Consonni, D; Bachetti, S; et al. (2001) Health effects of dioxin exposure: a 20-year mortality study.
2 Am J Epidemiol 153(11): 1031-1044.
3 Cheng, H; Aylward, L; Beall, C; et al. (2006) TCDD exposure-response analysis and risk assessment. Risk Anal
4 26:1059-1071.
5 Collins, JJ; Bodner, K; Aylward, LL; et al. (2009) Mortality rates among trichlorophenol workers with exposure to
6 2,3,7,8-tetrachlorodibenzo-p-dioxin. Am J Epidemiol 170(4):501-506.
7 Consonni, D; Pesatori, AC; Zocchetti, C; et al. (2008) Mortality in a population exposed to dioxin after the Seveso,
8 Italy, accident in 1976: 25 years of follow-up. Am J Epidemiol 167(7):847-858.
9 Eskenazi, B; Warner, M; Mocarelli, P; et al. (2002a) Serum dioxin concentrations and menstrual cycle
10 characteristics. Am J Epidemiol 156(4):383-392.
11 Eskenazi, B; Mocarelli, P; Warner, M; et al. (2002b) Serum dioxin concentrations and endometriosis: a cohort study
12 in Seveso, Italy. Environ Health Perspect 110(7):629-634.
13 Eskenazi, B; Mocarelli, P; Warner, M; et al. (2003) Maternal serum dioxin levels and birth outcomes in women of
14 Seveso, Italy. Environ Health Perspect 111(7), 947-953.
15 Eskenazi, B; Warner, M; Marks, AR; et al. (2005) Serum dioxin concentrations and age at menopause. Environ
16 Health Perspect 113(7):858-862.
17 Eskenazi, B; Warner, M; Samuels, S; et al. (2007) Serum dioxin concentrations and risk of uterine leiomyoma in the
18 Seveso Women's Health Study. Am J Epidemiol 166(l):79-87.
19 Fingerhut, MA; Halperin, WE; Marlow, DA; et al. (1991) Cancer mortality in workers exposed to
20 2,3,7,8-tetrachlorodibenzo-p-dioxin. N Engl J Med 324(4):212-218.
21 Flesch-Janys, D; Berger, J; Gurn, P; et al. (1995) Exposure to polychlorinated dioxins and furans (PCDD/F) and
22 mortality in a cohort of workers from a herbicide-producing plant in Hamburg, Federal Republic of Germany. Am J
23 Epidemiol 142(11): 1165-1175.
24 Flesch-Janys, D; Becher, H; Gurn, P; et al. (1996) Elimination of polychlorinated dibenzo-p-dioxins and
25 dibenzofurans in occupationally exposed persons. J Tox Environ Health 47(4):363-378.
26 Flesch-Janys, D; Steindorf, K; Gurn, P; et al. (1998) Estimation of the cumulated exposure to polychlorinated
27 dibenzo-p-dioxins/furans and standardized mortality ratio analysis of cancer mortality by dose in an occupationally
28 exposed cohort. Environ Health Perspect 106(Suppl 2):655-662.
29 Hooiveld, M; Heederik, DJ; Kogevinas, M; et al. (1998) Second follow-up of a Dutch cohort occupationally
30 exposed to phenoxy herbicides, chlorophenols, and contaminants. Am J Epidemiol 147(9):891-901.
31 Manz, A; Berger, J; Dwyer, JH; et al. (1991) Cancer mortality among workers in chemical plant contaminated with
32 dioxin. Lancet 338(8773):959-964.
33 McBride, DI; Collins, JJ; Humphry, NF; et al. (2009a) Mortality in workers exposed to 2,3,7,8-tetrachlorodibenzo-
34 p-dioxin at a trichlorophenol plant in New Zealand. J Occup Environ Med 51(9): 1049-1056.
35 McBride, DI; Burns, CJ; Herbison, GP; et al. (2009b) Mortality in employees at a New Zealand agrochemical
36 manufacturing site. Occup Med (Oxford, England) 59(4):255-263.
3 7 Michalek, JE; Pavuk, M. (2008) Diabetes and cancer in veterans of Operation Ranch Hand after adjustment for
3 8 calendar period, days of spraying, and time spent in Southeast Asia. J Occup Environ Med 50(3):330-340.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-61 DRAFT—DO NOT CITE OR QUOTE
-------
1 Mocarelli, P; Gerthoux, PM; Ferrari, E; et al. (2000) Paternal concentrations of dioxin and sex ratio of offspring.
2 Lancet 355(9218): 1858-1863.
3 Mocarelli, P; Gerthoux, PM; Patterson, DG, Jr.; et al. (2008) Dioxin exposure, from infancy through puberty,
4 produces endocrine disruption and affects human semen quality. Environ Health Perspect 116(l):70-77.
5 Ott, MG; Zober, A. (1996) Cause specific mortality and cancer incidence among employees exposed to
6 2,3,7,8-TCDD after a 1953 reactor accident. Occup Environ Med 53(9):606-612.
7 Pesatori, AC; Consonni, D; Bachetti, S; et al. (2003) Short- and long-term morbidity and mortality in the population
8 exposed to dioxin after the "Seveso accident". Ind Health 41(3): 127—138.
9 Revich, B; Aksel, E; Ushakova, T; et al. (2001) Dioxin exposure and public health in Chapaevsk, Russia.
10 Chemosphere 43(4-7):951-966.
11 Ryan, JJ; Amirova, Z; Carrier, G. (2002) Sex ratios of children of Russian pesticide producers exposed to dioxin.
12 Environ Health Perspect, 110(11):A699-701.
13 Steenland, K; Piacitelli, L; Deddens, J; et al. (1999) Cancer, heart disease, and diabetes in workers exposed to
14 2,3,7,8-tetrachlorodibenzo-p-dioxin. J Natl Cancer I 91(9):779-786.
15 Steenland, K; Deddens, J; Piacitelli, L. (2001) Risk assessment for 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)
16 based on an epidemiologic study. Am J Epidemiol 154(5):451-458.
17 't Mannetje, A; McLean, D; Cheng, S; et al. (2005) Mortality in New Zealand workers exposed to phenoxy
18 herbicides and dioxins. Occup Environ Med 62(l):34-40.
19 Warner, M; Eskenazi, B; Mocarelli, P; et al. (2002) Serum dioxin concentrations and breast cancer risk in the
20 Seveso Women's Health Study. Environ Health Perspect 110(7):625-628.
21 Warner, M; Samuels, S; Mocarelli, P; et al. (2004) Serum dioxin concentrations and age at menarche. Environ
22 Health Perspect 112(13): 1289-1292.
23 Warner, M; Eskenazi, B; Olive, DL; et al. (2007) Serum dioxin concentrations and quality of ovarian function in
24 women of Seveso. Environ Health Perspect 115(3):336-340.
25 Zober, A; Messerer, P; Huber, P. (1990) Thirty-four-year mortality follow-up of BASF employees exposed to
26 2,3,7,8-TCDD after the 1953 accident. Int Arch Occup Environ Health 62(2): 139-157.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 B-62 DRAFT—DO NOT CITE OR QUOTE
-------
DRAFT
DO NOT CITE OR QUOTE
January 2010
Agency/Interagency Review Draft
APPENDIX C
Kinetic Modeling
NOTICE
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
CONTENTS—APPENDIX C: Kinetic Modeling
APPENDIX C. KINETIC MODELING C-l
C. 1. LITERATURE SEARCH STRATEGY AND RESULTS—IDENTIFYING
RECENT PUBLICATIONS FOR UPDATING TCDD TOXICOKINETIC
MODEL INPUT PARAMETERS C-l
C.l.l. Data Bases Searched C-l
C.1.2. Literature Search Strategy and Approach C-2
C.1.2.1. Chemical Search Terms—DIALOG Search C-2
C.1.2.2. Primary Search Terms (Species)—DIALOG Search C-2
C.1.2.3. Secondary Search Terms (Toxicology)—DIALOG Search C-3
C.1.3. Citation Screening Procedures and Results C-3
C. 1.4. References Selected for More Detailed Review for Updating the PBPK
Models C-6
C.1.5. Brief Descriptions of DIALOG Bibliographic Data Bases Searched C-8
C.2. TOXICOKINETIC MODELING CODE (EMOND ET AL., 2005) C-10
C.2.1. Human Standard Model C-10
C.2.1.1. Model Code C-10
C.2.1.2. Input File C-20
C.2.2. Human Gestational Model C-21
C.2.2.1. Model Code C-21
C.2.2.2. Input File C-35
C.2.3. Rat Standard Model C-36
C.2.3.1. Model Code C-36
C.2.3.2. Input Files C-46
C.2.3.2.1. Cantoni et al. (1981) C-46
C.2.3.2.2. Chu et al. (2007) C-46
C.2.3.2.3. Crofton et al. (2005) C-47
C.2.3.2.4. Fattore et al. (2000) C-48
C.2.3.2.5. Hassoun et al. (2000) C-48
C.2.3.2.6. Kitchin and Woods (1979) C-49
C.2.3.2.7. Kociba et al. (1976) (13 weeks) C-50
C.2.3.2.8. Kociba et al. (1978) (female) (104 weeks) C-51
C.2.3.2.9. Kociba et al. (1978) (male) (104 weeks) C-51
C.2.3.2.10. Latchoumycandane and Mathur. (2002) C-52
C.2.3.2.11. Li etal. (1997) C-53
C.2.3.2.12. Murray et al. (1979) C-53
C.2.3.2.13. NTP (1982) (female) (chronic) C-54
C.2.3.2.14. NTP (1982) (male) (chronic) C-55
C.2.3.2.15. NTP (2006) 31 weeks C-55
C.2.3.2.16. NTP (2006) 53 weeks C-56
C.2.3.2.17. NTP (2006) 2 year C-57
C.2.3.2.18. Sewall et al. (1995) C-58
C.2.3.2.19. Shi etal. (2007), adult portion C-58
C.2.3.2.20. Van Birgelen et al. (1995) C-59
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-ii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
CONTENTS (continued)
C.2.3.2.21. Vanden Heuvel et al. (1994) C-60
C.2.4. Rat Gestational Model C-60
C.2.4.1. Model Code C-60
C.2.4.2. Input Files C-74
C.2.4.2.1. Bell et al. (2007) C-74
C.2.4.2.2. Hojo et al. (2002) C-75
C.2.4.2.3. Ikeda et al. (2005) C-76
C.2.4.2.4. Kattainen et al. (2001) C-77
C.2.4.2.5. Markowski et al. (2001) C-78
C.2.4.2.6. Miettinen et al. (2006) C-78
C.2.4.2.7. Murray et al. (1979) C-79
C.2.4.2.8. Nohara et al. (2000) C-80
C.2.4.2.9. Ohsako et al. (2001) C-81
C.2.4.2.10. Schantz et al. (1996) and Amin et al. (2000) C-82
C.2.4.2.11. Seoetal. (1995) C-82
C.2.4.2.12. Shi et al. (2007) C-83
C.2.5. Mouse Standard Model C-84
C.2.5.1. Model Code C-84
C.2.5.2. Input Files C-94
C.2.5.2.1. Hassoun et al. (1998) (13 weeks) C-94
C.2.5.2.2. NTP (1982) (male) (chronic) C-96
C.2.5.2.3. Smialowicz et al. (2008) C-96
C.2.5.2.4. Toth et al. (1979) (1 year) C-97
C.2.5.2.5. Toth et al. (1979) (2 year) C-98
C.2.5.2.6. White et al. (1986) C-99
C.2.6. Mouse Gestational Model C-100
C.2.6.1. Model Code C-100
C.2.6.2. Input Files C-113
C.2.6.2.1. Keller etal. (2007) C-113
C.2.6.2.2. Li et al. (2005) C-l 14
C.3. TOXICOKINETIC MODELING RESULTS FOR KEY ANIMAL
BIO AS SAY STUDIES C-l 15
C.3.1. Nongestational Studies C-115
C.3.1.1. Cantoni et al. (1981) C-115
C.3.1.2. Chu et al. (2007) C-l 17
C.3.1.3. Crofton et al. (2005) C-l 19
C.3.1.4. Fattore etal. (2000) C-l22
C.3.1.5. Hassoun et al. (1998) C-l24
C.3.1.6. Hassoun et al. (2000) C-l26
C.3.1.7. Kitchin and Woods (1979) C-l29
C.3.1.8. Kocibaetal. (1976) C-l32
C.3.1.9. Kociba etal. (1978) Female C-l 34
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-iii DRAFT—DO NOT CITE OR QUOTE
-------
1 CONTENTS (continued)
2
3
C.3.1.10. Kocibaetal. (1978) Male C-136
C.3.1.11. Latchoumycandane and Mathur (2002) C-138
C.3.1.12. Li etal. (1997) C-139
C.3.1.13. NTP (1982)—Female Rats, Chronic C-143
C.3.1.14. NTP (1982)—Male Rats, Chronic C-144
C.3.1.15. NTP (1982)—Female Mice, Chronic C-146
C.3.1.16. NTP (1982)—Male Mice, Chronic C-148
C.3.1.17. NTP (2006) 31 Weeks C-150
C.3.1.18. NTP (2006) 53 Weeks C-152
C.3.1.19. NTP (2006) 2 Years C-154
C.3.1.20. Sewalletal. (1995) C-156
C.3.1.21. Smialowicz et al. (2008) C-158
C.3.1.22. Toth et al., 1 Year (1979) C-160
C.3.1.23. Van Birgelen (1995) C-162
C.3.1.24. Yanden IIeuvel etal. (1994) C-164
C.3.1.25. White etal. (1986) C-167
4 C.3.2. Gestational Studies C-170
C.3.2.1. Bell et al. (2007) C-170
C.3.2.2. Hojo et al. (2002) C-171
C.3.2.3. Ikeda et al. (2005) C-173
C.3.2.4. Kattainen et al. (2001) C-174
C.3.2.5. Keller et al. (2007) C-175
C.3.2.6. Li et al. (2006) 3-Day C-177
C.3.2.7. Markowski et al. (2001) C-178
C.3.2.8. Mietinnen et al. (2006) C-180
C.3.2.9. Murray et al. (1979) Gestational Portion C-181
C.3.2.10. Murray et al. (1979) Adult Portion C-183
C.3.2.11. Nohara et al. (2000) C-185
C.3.2.12. Ohsako et al. (2001) C-186
C.3.2.13. Schantzetal. (1995) and Amin et al. (2000) C-188
C.3.2.14. Seo etal. (1995) C-189
C.3.2.15. Shi et al. (2007) Gestational Portion C-191
C.3.2.16. Shi etal. (2007) Adult Portion C-192
5 C.4. REFERENCES C-196
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-iv DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
APPENDIX C. KINETIC MODELING
C.l. LITERATURE SEARCH STRATEGY AND RESULTS—IDENTIFYING RECENT
PUBLICATIONS FOR UPDATING TCDD TOXICOKINETIC MODEL INPUT
PARAMETERS
The purpose of this literature search was to identify recent publications that address the
input parameters for the physiologically based pharmacokinetic (PBPK) models Aylward and
colleagues described in 2004-2005 and Emond and colleagues described in 2004-2006. This
literature search was part of the U.S. Environmental Protection Agency (EPA)'s preparation of a
response to the National Academy of Sciences' review (Health Risks from Dioxin and Related
Compounds: Evaluation of the EPA Reassessment, NAS, 2006]) of EPA Reassessment of Health
Risks From Dioxin and Related Compounds (2003 Reassessment, U.S. EPA, 2003). English-
only references from 2003 to May 2009 were searched using bibliographic data bases relevant to
health effects and toxicology of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD). The search
focused on toxicokinetic data that could be used to update the dynamic disposition of
2,3,7,8-TCDD in mice, rats, guinea pigs, monkeys, and humans.
In the primary search, EPA identified 775 distinct citations based on the literature search
criteria described below. EPA also performed an independent supplemental search to avoid
missing key studies. EPA identified 28 papers for further analysis that appeared on first review
to report data to update the input parameters of the Aylward and Emond PBPK models;
considerations for selection are described in Section C.1.3.
C.l.l. Data Bases Searched
EPA used the following DIALOG bibliographic data bases in the primary search. Brief
descriptions of the DIALOG data bases searched are provided in Section C.l.5.
1. File 6: NTIS
2. File 41: Pollution Abstracts
3. File 55: Biosis
4. File 153: IPA Toxicology
5. File 155: MedLine
6. File 156: ToxFile
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
7. File 157: Biosis Toxicology
8. File 159: CancerLit
9. File 336: RTECS
The PUB MED data base was used for the supplemental search.
C.1.2. Literature Search Strategy and Approach
The primary search used a tiered key-word approach, as documented below. The
principal search term was the Chemical Abstract Service Registry Number (CASRN) or specific
chemical name, 2,3,7,8-tetrachlorodibenzo-p-dioxin or 2,3,7,8-TCDD. The next tier of search
terms was species, and finally toxicokinetic keywords, as listed below. The period of the search
was 2003 through May 2009, and articles were limited to English language.
The supplemental PUB MED search was limited to the most recent five years (2004 to
present) and used four combinations of key words:
• TCDD + pharmacokinetic + humans,
• TCDD + toxicokinetic + humans,
• TCDD + pharmacokinetic + animals, and
• TCDD + toxicokinetic + animals.
C.l.2.1. Chemical Search Terms—DIALOG Search
• CASRN: 1746-01-6
• 2,3,7,8 -tetrachl orodib enzo-p-di oxin
• 2,3,7,8-TCDD
C. 1.2.2. Primary Search Terms (Species)—DIALOG Search
• Guinea pig(s)
• Human(s)
• Monkey(s)
• Mouse
• Mice
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-2 DRAFT—DO NOT CITE OR QUOTE
-------
1 • Rodent(s)
2 • Rat(s)
3
4 C.l.2.3. Secondary Search Terms (Toxicology)—DIALOG Search
5
1. Absor*
2. ADME
3. Aryl hydrocarbon
receptor
4. AhR
5. Bioavail*
6. Biliar*
7. Biotransform*
8. Cytochrome
9. CYP*
10. CYP1A1
11. CYP1A2
12. Diet, dietary, diets
13. Disposit*
14. Distrib*
15. Drink*
1
2 ADME = absorption, distribution, metabolism, elimination; AhR = aryl hydrocarbon receptor; CYP =
3 cytochrome P450; * = truncated; lw = terms are within 1 word of each other and in the order
4 specified (see search term 32
5
6
7 C.1.3. Citation Screening Procedures and Results'
8 Initial DIALOG searches resulted in a very large number of citation hits. Therefore,
9 some title and key word restrictions were applied iteratively to screen out less relevant citations
10 (e.g., requiring some search terms in title, requiring 2,3,7,8-TCDD rather than just TCDD).
11 Then, using reference management software, pooled information obtained from the various
12 DIALOG data bases was screened to remove duplicates. Citations then were numbered
13 sequentially (as a unique identifier). Information retrieved included the following (when
14 available): author(s), publication year, title, source document name, volume, and page numbers.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-3 DRAFT—DO NOT CITE OR QUOTE
16.
Elimin*
31.
Lymph*
17.
Excret*
32.
Mechanism (lw)
18.
Epidemiolog*
action
19.
Feces
33.
Metabo*
20.
Feed*
34.
Oral*
21.
First order kinetics
35.
P450
22.
Food*
36.
Partition coefficient
23.
Gastro*
37.
PBPK
24.
Gavage*
38.
Pharmacody nami c *
25.
Half-life
39.
Pharmacokineti c *
26.
Induct*
40.
Physiologically
based
27.
Ingest*
41.
pharmacokinetic
28.
In silico
Protein bind*
42.
29.
Kinetic*
Toxicokinetic*
43.
30.
Liver
Urin*
44.
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
The DIALOG search and duplicate removal procedure produced 775 unique citations. In
the next step, all 775 citations were screened for potential applicability to updating parameters in
the Aylward and Emond PBPK models. Of these 775 citations, 26 were selected for more
detailed review to determine their potential applicability, and full publications were retrieved.
Two citations were added from the supplemental search, giving a total of 28 articles identified
for further review.
Bibliographic information for the 28 articles selected for full review is provided in the
reference list at the end of this section. Table C-l summarizes the model input parameters
potentially addressed by the selected articles.
During 2003 to May 2009, the authors of the two kinetic models under consideration
published several articles. For the Emond model, which was first published in 2004 (Emond et
al., 2004), two subsequent papers have been published (Emond et al., 2005, 2006). The Aylward
model, which originated from the 1995 papers by Carrier et al. (1995a, b), was later updated by
the same group (Aylward et al., 2004, 2005). The major change implemented in the last two
papers was the description of a desorption process in the digestive tract. The transfer rate
described is slow, but for a low body burden of TCDD, this process remains significant. This
concept was reported in 2002 by Moser and McLachlan (2002). The major modifications
expected to update the Emond model are (1) consideration of the desorption process in the
gastrointestinal tract and (2) rearrangement of the elimination constant, which will have a
negligible impact on the simulation. These changes are motivated by plausible observations
reported in the literature.
Because of the body burden found in humans and the importance of selecting an
appropriate dose metric in human risk assessment, the physiological model is an important tool
for assessing the kinetics following exposure to TCDD (Kim et al., 2003). Based on the
literature identified in this search, the major contributions that should be reviewed with respect to
the Aylward and Emond kinetic models are not modes of action or pharmacokinetic mechanisms,
but rather information for verifying or improving the accuracy of some model parameters.
Pharmacokinetics typically refers to four distinct steps including absorption, distribution,
metabolism, and excretion. Physiologically-based models consider each step. In the model each
step is parameterized to reflect better predictions of the real observations. Occasionally,
reviewing these models is essential to determine if any key processes or parameters might be
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
described with better accuracy. This perspective underlies the review of the literature described
here. The review indicates TCDD disposition has become recognized as relatively significant
since the publication of the Emond and Aylward models. The literature that provides
information related to improving these models, however, is limited. For the benefit of this
exercise, EPA selected the literature that would likely contribute significantly to model response,
or to clarify or confirm different key issues driving the model results. Regarding the two TCDD
models, the two major issues that should be evaluated with respect to the recent literature
identified are the elimination profile and the induction of CYP1A2.
Reviewing the elimination variation in different species and testing variable elimination
with a data set appears to be appropriate. The literature reports that various factors might
influence elimination rate. Recent publications report the influence of diverse predictors such
age, body fat, or smoking habit on the elimination half-life (Milbrath et al., 2009; Kerger et al.,
2006, 2007). Determining whether using the Milbrath et al. information would help account for
intraspecies variability in elimination rate in the Emond and Aylward kinetic models would be
useful. In 2006, Emond et al. reviewed the influence of body fat mass and CYP1A2 induction on
the pharmacokinetics of TCDD. These two factors appear to contribute significantly to
elimination and their influences seem to be driven by TCDD body burden. Mullerova and
Kopecky (2007) discussed the influence of adipose tissue and the "yoyo" effects on various
diseases that might be influenced by persistent organic pollutant distribution. One group
explored the importance of variable elimination and compared these predictions to first-order
elimination using the Aylward and Emond models and supported these approaches for risk
assessment (Heinzl et al., 2007). Two groups of authors considered a one-compartment model to
derive the elimination half-life (Aylward et al., 2009; Nadal et al., 2008). Comparing the
half-life they obtained using this approach for a range of body burden to the variable elimination
half-life would be interesting.
The second important mechanism driving the distribution and elimination of TCDD is the
induction of CYP1A2, identified as the major ligand protein in liver (Diliberto et al., 1997). For
that process, authors suggested different aspects that should be investigated, including the
importance of the dose metrics in the target tissue and the inducible level of CYP1A2 (Wilkes
et al., 2008; Staskal et al., 2005). Other papers address the intraspecies variability of lethal
potency in mature species versus the developing fetus (Kransler et al., 2007; Korkalainen et al.,
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
2004). Still others point out pronounced differences among species (namely, guinea pigs,
hamsters, mice, and rats) (Bohonowych and Denison, 2007), as observed in studies of long-term
effects of low TCDD dose in liver and in studies comparing hepatic accumulation and clearance
of TCDD (Korenaga et al., 2007; Boverhof et al., 2005). The interspecies variation of the
binding affinity constant of AhR also has been reported (Connor and Aylward, 2006; Nohara
et al., 2006).
The articles identified in this literature review should be adequate to update the Aylward
and Emond models, which need to be evaluated according to the same structure of compartments
described in the literature by the two model authors.
C.1.4. References Selected for More Detailed Review for Updating the PBPK Models
Aylward, LL; Brunet, RC; Carrier, G; et al. (2004) Concentration-dependent TCDD elimination kinetics in humans:
toxicokinetic modeling for moderately to highly exposed adults from Seveso, Italy, and Vienna, Austria, and impact
on dose estimates for the NIOSH cohort. J Expo Anal Environ Epidemiol 15(1):51-65.
Aylward, LL; Brunet, RC; Starr, TB; et al. (2005) Exposure reconstruction for the TCDD-exposed NIOSH cohort
using a concentration- and age-dependent model of elimination. Risk Anal 25(4):945-956.
Aylward, LL; Bodner, KM; Collins, JJ; et al. (2009) TCDD exposure estimation for workers at a New Zealand
2,4,5-T manufacturing facility based on serum sampling data. J Expo Sci Environ Epidemiol, doi:
10.1038/jes.2009.31.
Bohonowych, JE; Denison, MS. (2007) Persistent binding of ligands to the aryl hydrocarbon receptor. Toxicol Sci
98(1):99-109.
Boverhof, DR; Burgoon, LD; Tashiro, C; et al. (2005) Temporal and dose-dependent hepatic gene expression
patterns in mice provide new insights into TCDD-mediated hepatotoxicity. Toxicol Sci 85(2): 1048-1063.
Connor, KT; Aylward, LL. (2006) Human response to dioxin: aryl hydrocarbon receptor (AhR) molecular structure,
function, and dose-response data for enzyme induction indicate an impaired human AhR. J Toxicol Environ Health
B 9(2):147-171.
Heinzl, H; Mittlback, M; Edler, L. (2007) On the translation of uncertainty from toxicokinetic to toxicodynamic
models - the TCDD example. Chemosphere 67(9):S365-S374.
Irigaray, P; Mejean, L; Laurent, F. (2005) Behaviour of dioxin in pig adipocytes. Food Chem Toxicol
43(3)457-460.
Kerger, BD; Leung, HW; Scott, P; et al. (2006) Age- and concentration-dependent elimination half-life of 2,3,7,8-
tetrachlorodibenzo-p-dioxin in Seveso children. Environ Health Perspect 114(10): 1596-1602.
Kerger, BD; Leung, HW; Scott, PK; et al. (2007) Refinements on the age-dependent half-life model for estimating
child body burdens of polychlorodibenzodioxins and dibenzofurans. Chemosphere 67(9):S272-S278.
Kim, AH; Kohn, MC; Nyska, A; et al. (2003) Area under the curve as a dose metric for promotional responses
following 2,3,7,8-tetrachlorodibenzo-p-dioxin exposure. Toxicol Appl Pharmacol 191(1): 12-21.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-6 DRAFT—DO NOT CITE OR QUOTE
-------
Korenaga, T; Fukusato, T; Ohta, M; et al. (2007) Long-term effects of subcutaneously injected 2,3,7,8-
tetrachlorodibenzo-p-dioxin on the liver of rhesus monkeys. Chemosphere 67(9):S399-S404.
Korkalainen, M; Tuomisto, J; Pohjanvirta, R. (2004) Primary structure and inducibility by 2,3,7,8-
tetrachlorodibenzo-p-dioxin (TCDD) of aryl hydrocarbon receptor repressor in a TCDD-sensitive and a TCDD-
resistant rat strain. Biochem Biophys Res Communications 315(1): 123—131.
Kransler, KM; McGarrigle, BP; Olson, JR. (2007) Comparative developmental toxicity of 2,3,7,8-
tetrachlorodibenzo-p-dioxin in the hamster, rat and guinea pig. Toxicology 229(3):214-225.
Maruyama, W; Yoshida, K; Tanaka, T; et al. (2002) Determination of tissue-blood partition coefficients for a
physiological model for humans, and estimation of dioxin concentration in tissues. Chemosphere 46(7):975-985.
Maruyama, W; Yoshida, K; Tanaka, T; et al. (2003) Simulation of dioxin accumulation in human tissues and
analysis of reproductive risk. Chemosphere 53(4):301-313.
Maruyama, W; Aoki, Y. (2006) Estimated cancer risk of dioxins to humans using a bioassay and physiologically
based pharmacokinetic model. Toxicol Appl Pharmacol 214(2): 188-198.
Milbrath, MO; Wenger, Y; Chang, C-W; et al. (2009) Apparent Half-Lives of Dioxins, Furans, and Poly chlorinated
Biphenyls as a Function of Age, Body Fat, Smoking Status, and Breast-Feeding. Environ Health Perspect
117(3):417-425.
Moser, GA; McLachlan, MS. (2002) Modeling digestive tract absorption and desorption of lipophilic organic
contaminants in humans. Environ Sci Technol 36(15):3318—25.
Mullerova, D; Kopecky, J. (2007) White adipose tissue: storage and effector site for environmental pollutants.
Physiol Res 56(4):375-381.
Nadal, M; Perello, G; Schuhmacher, M; et al. (2008) Concentrations of PCDD/PCDFs in plasma of subjects living
in the vicinity of a hazardous waste incinerator: Follow-up and modeling validation. Chemosphere 73(6):901-906.
Nohara, K; Ao, K; Miyamoto, Y; et al. (2006) Comparison of the 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)-
induced CYP1A1 gene expression profile in lymphocytes from mice, rats, and humans: Most potent induction in
humans. Toxicology 225(2-3):204-213.
Olsman, H; Engwall, M; Kammann, U; et al. (2007) Relative differences in aryl hydrocarbon receptor-mediated
response for 18 polybrominated and mixed halogenated dibenzo-p-dioxins and -furans in cell lines from four
different species. Environ Toxicol Chem 26(11):2448-2454.
Saghir, SA; Lebofsky, M; Pinson, DM; et al. (2005) Validation of Haber's Rule (doseX time=constant) in rats and
mice for monochloroacetic acid and 2,3,7,8-tetrachlorodibenzo-p-dioxinunder conditions of kinetic steady state.
Toxicology 215(l-2):48-56.
Schecter, A; Pavuk, M; Popke, O; et al. (2003) Dioxin, dibenzofuran, and coplanar PCB Levels in Laotian blood and
milk from Agent Orange-sprayed and nonsprayed areas, 2001. J Toxicol Environ Health A 66(21):2067-2075.
Staskal, DF; Diliberto, JJ; Devito, MJ; et al. (2005) Inhibition of human and rat CYP1A2 by TCDD and dioxin-like
chemicals. Toxicol Sci 84(2):225-231.
Toyoshiba, H; Walker, NJ; Bailer, AJ; et al. (2004) Evaluation of toxic equivalency factors for induction of
cytochromes P450 CYP1A1 and CYP1A2 enzyme activity by dioxin-like compounds. Toxicol Appl Pharmacol
194(2): 156-168.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
Wilkes, JG; Hass, BS; Buzatu, DA; et al. (2008) Modeling and assaying dioxin-like biological effects for both
dioxin-like and certain non-dioxin-like compounds. Toxicol Sci 102(1): 187-195.
C.1.5. Brief Descriptions of DIALOG Bibliographic Data Bases Searched
The National Technical Information Service (NTIS) database comprises summaries of
U.S. government-sponsored research, development, and engineering, plus analyses prepared by
federal agencies, their contractors, or grantees. It is the means through which unclassified,
publicly available, unlimited distribution reports are made available for sale from 240 agencies.
Additionally, some state and local government agencies contribute summaries of their reports to
the database. NTIS also provides access to the results of government-sponsored research and
development from countries outside the United States. Organizations that currently contribute to
the NTIS database include but are not limited to the following: the Japan Ministry of
International Trade and Industry (MITI); laboratories administered by the United Kingdom
Department of Industry; the German Federal Ministry of Research and Technology (BMFT); and
the French National Center for Scientific Research (CNRS).
Pollution Abstracts provides access to environmental information that combines
information on scientific research and government policies in a single resource. Topics of
growing concern are extensively covered from the standpoints of atmosphere, emissions,
mathematical models, effects on people and animals, and environmental action in response to
global pollution issues. This database also contains material from conference proceedings and
hard-to-find summarized documents along with information from primary journals in the field of
pollution.
BIOSIS Previews® contains citations from Biological Abstracts® (BA) and Biological
Abstracts/Reports, Reviews, and Meetings® (BA/RRM) (formerly BioResearch Index®), the
major publications of BIOSIS®. These publications constitute the major English-language
service providing comprehensive worldwide coverage of research in the biological and
biomedical sciences. Biological Abstracts includes approximately 350,000 accounts of original
research yearly from nearly 5,000 primary journal and monograph titles. BA/RRM includes an
additional 200,000+ citations a year from meeting abstracts, reviews, books, book chapters,
notes, letters, and selected reports.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
IPA Toxicology provides focused toxicology information on all phases of the
development and use of drugs and on professional pharmaceutical practice. The scope of the
database ranges from the clinical and practical to the theoretical aspects of toxicology literature.
A unique feature of abstracts reporting clinical studies is the inclusion of the study design,
number of patients, dosage, dosage forms, and dosage schedule.
Medical Literature, Analysis, and Retrieval System Online (MEDLINE®), produced by
the U.S. National Library of Medicine (NLM), is NLM's premier bibliographic database. It
contains more than 15 million references to journal articles in life sciences with a concentration
on biomedicine. The broad coverage of the database includes basic biomedical research and the
clinical sciences since 1950, including nursing, dentistry, veterinary medicine, pharmacy, allied
health, and pre-clinical sciences. MEDLINE® also covers life sciences that are vital to
biomedical practitioners, researchers, and educators, including some aspects of biology,
environmental science, marine biology, and plant and animal science, as well as biophysics and
chemistry. MEDLINE® is indexed using NLM's controlled vocabulary, Medical Subject
Headings (MeSH®). Approximately 400,000 records are added per year, of which more than 76
percent are in English. MEDLINE® contains AIDSLINE, HealthSTAR, Toxline, In Process
(formerly known as Pre-MEDLINE®), In Data Review, and POPLINE.
ToxFile covers the toxicological, pharmacological, biochemical, and physiological
effects of drugs and other chemicals. Adverse drug reactions, chemically induced diseases,
carcinogenesis, mutagenesis, teratogenesis, environmental pollution, waste disposal, radiation,
and food contamination are typical areas of coverage. The databases Environmental Mutagen
Information Center (EMIC), Developmental and Reproductive Toxicology (DART), and Toxic
Substances Control Act Test Submissions (TSCATS) are included in ToxFile. It is not clearly
stated whether the Chemical Carcinogenesis Research Information System (CCRIS), Hazardous
Substances Data Bank (HSDB), or Genetic Toxicology Data Bank (GENE-TOX) are included in
ToxFile. Consequently, a separate, on-line search was conducted to ensure that these databases
were searched.
BIOSIS® Toxicology contains citations from BA and BA/RRM (formerly BioResearch
Index®), the major publications of BIOSIS®, that focus on toxicology and related topics.
Records are drawn from journal articles, conference papers, monographs and book chapters,
notes, letters, and reports, as well as original research. U.S. patent records are also included.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
CANCERLIT® is produced by the International Cancer Research DataBank Branch
(ICRDB) of the U.S. National Cancer Institute. The database consists of bibliographic records
referencing cancer research publications dating from 1963 to 2002. Most records contain
abstracts, and all records contain citation information and additional descriptive fields such as
document type and language. Beginning with the June 1983 CANCERLIT update, records from
the MEDLINE® database dealing with cancer topics have been added to CANCERLIT.
The Registry of Toxic Effects of Chemical Substances (RTECS®) is a comprehensive
database of basic toxicity information for over 150,000 chemical substances including
prescription and non-prescription drugs, food additives, pesticides, fungicides, herbicides,
solvents, diluents, chemical wastes, reaction products of chemical waste, and substances used in
both industrial and household situations. Reports of the toxic effects of each compound are
cited. In addition to toxic effects and general toxicology reviews, data on skin and/or eye
irritation, mutation, reproductive consequences and tumorigenicity are provided. Federal
standards and regulations, National Institute for Occupational Safety and Health (NIOSH)
recommended exposure limits and information on the activities of EPA, NIOSH, National
Toxicology Program (NTP), and Occupational Safety and Health Administration (OSHA)
regarding the substance are also included. The toxic effects are linked to literature citations from
both published and unpublished governmental reports, and published articles from the scientific
literature. The database corresponds to the print version of the RTECS®, formerly known as the
Toxic Substances List, which was started in 1971. Originally prepared by the NIOSH, the
RTECS® database is now produced and distributed by Symyx Technologies, Inc.
C.2. TOXICOKINETIC MODELING CODE (Emond et al., 2005)
C.2.1. Human Standard Model
C.2.1.1. Model Code
PROGRAM: 'Three Compartment PBPK Model for TCDD in Human: Standard Model
(Non-Gestation)'
!HUM_NON_GEST_ICF_F083109.csl
INITIAL ! INITIALIZATION OF PARAMETERS
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
! SIMULATION PARAMETERS
CONSTANT EXP TIME ON
(HOUR)
CONSTANT EXPTIMEOFF
(HOUR)
CONSTANT DAY CYCLE
(HOUR)
CONSTANT BCK TIME ON
EXPOSURE BEGINS (HOUR)
CONSTANT BCKTIMEOFF
EXPOSURE ENDS (HOUR)
= 0. ! TIME AT WHICH EXPOSURE BEGINS
= 6.132e5 ! TIME AT WHICH EXPOSURE ENDS
= 24.0 ! NUMBER OF HOURS BETWEEN DOSES
= 6.132e5 ! TIME AT WHICH BACKGROUND
= 6.132e5 ! TIME AT WHICH BACKGROUND
!EXPO SURE DOSES
CONSTANT MSTOTBCKGR
(NG/KG)
CONSTANT MSTOT
CONSTANT DOSEIV
CONSTANT MW
MSTOT NM = MSTOT/MW
= 0.0 ! ORAL BACKGROUND EXPOSURE DOSE
1 OE-7 ! ORAL EXPOSURE DOSE (NG/KG)
0.0 ! INJECTED DOSE (NG/KG)
322.0 ! MOLECULAR WEIGHT (G/MOL)
! CONVERTS THE DOSE TO NMOL/KG
MSTOT NMBCKGR = MSTOTBCKGR/MW ! CONVERTS THE BACKGROUND DOSE
TO NMOL/KG
DOSEIV NM = DOSEIV/MW ! CONVERTS THE INJECTED DOSE TO
NMOL/KG
!INITIAL GUESS OF THE FREE CONCENTRATION IN THE LIGAND
(COMPARTMENT INDICATED BELOW) ====
CONSTANT CFLLI0 = 0.0 ! LIVER (NMOL/L)
IBINDING CAPACITY (AhR) FOR NON LINEAR BINDING (COMPARTMENT
INDICATED BELOW) ===
CONSTANT LIBMAX = 0.35 ! LIVER (NMOL/L)
! PROTEIN AFFINITY CONSTANTS (1A2 OR AhR, COMPARTMENT INDICATED
BELOW) ===
CONSTANT KDLI = 0.1 ! LIVER (AhR) (NMOL/L) WANG ET AL.. 1997
CONSTANT KDLI2 = 40.0 ! LIVER (1A2) (NMOL/L) EMOND ET AL. 2004
!EXCRETION AND ABSORPTION CONSTANTS
CONSTANT KST = 0.01 ! GASTRIC RATE CONSTANT (HR-1), EMOND
ET AL., 2005
CONSTANT KABS = 0.06 ! INTESTINAL ABSORPTION CONSTANT (HR-1),
EMOND ET AL. 2005
!ELIMINATION CONSTANTS
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
CONSTANT CLURI = 4.17D-8 ! URINARY CLEARANCE (L/HR), EMOND ET
AL., 2005
CONSTANT KELV = l .le-3 ! INTERSPECIES VARIABLE ELIMINATION
CONSTANT (1/HOUR)
!CONSTANT TO DIVIDE THE ABSORPTION INTO LYMPHATIC AND PORTAL
FRACTIONS
CONSTANT A
(1997)
0.7
!PARTITION COEFFICIENTS
CONSTANT PF = 1.0e2
1997
CONSTANT PRE = 1.5
1997
CONSTANT PLI = 6.0
LYMPHATIC FRACTION, WANG ET AL.
ADIPOSE TISSUE/BLOOD, WANG ET AL.
! REST OF THE BODY/BLOOD, WANG ET AL.
! LIVER/BLOOD, WANGET AL. 1997
!PARAMETERS FOR INDUCTION OF CYP1A2
CONSTANT PAS INDUC = 1.0 ! INCLUDE INDUCTION? (1 = YES, 0 = NO)
CONSTANT CYP1A210UTZ = 1.6e3 ! DEGRADATION CONCENTRATION
CONSTANT OF 1A2 (NMOL/L)
CONSTANT CYP1A21A1 = 1.6e3 ! BASAL CONCENTRATION OF 1A1
(NMOL/L)
CONSTANT CYP1A21EC50 = 1.3e2 ! DISSOCIATION CONSTANT TCDD-CYP1A2
(NMOL/L)
CONSTANT CYP1A21A2 = 1.6e3
(NMOL/L)
CONSTANT CYP1A21KOUT = 0.1
(H-l)
CONSTANT CYP1A2 1TAU = 0.25
! BASAL CONCENTRATION OF 1A2
! FIRST ORDER RATE OF DEGRADATION
! HOLDING TIME (H)
CONSTANT CYP1A21EMAX = 9.3e3 ! MAXIMUM INDUCTION OVER BASAL
EFFECT (UNTITLES S)
CONSTANT HILL = 0.6 !HILL CONSTANT; COOPERATIVELY LIGAND
BINDING EFFECT CONSTANT (UNITLESS)
! DIFFUSION AL PERMEABILITY FRACTION
CONSTANT PAFF = 0.12 ! ADIPOSE (UNITLESS)
CONSTANT PAREF = 0.03 ! REST OF BODY (UNITLESS)
CONSTANT PALIF = 0.35 ! LIVER (UNITLESS)
!TISSUE BLOOD FLOW EXPRESSED AS A FRACTION OF CARDIAC OUTPUT
CONSTANT QFF = 0.05 ! ADIPOSE TISSUE BLOOD FLOW FRACTION
(UNITLESS), KRISHNAN 2008
CONSTANT QLIF = 0.26 ! LIVER (UNITLESS), KRISHNAN 2008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
ICOMPARTMENT TISSUE BLOOD EXPRESSED AS A FRACTION OF THE TOTAL
COMPARTMENT VOLUME =========
CONSTANT WFBO = 0.050 ! ADIPOSE TISSUE, WANG ET AL. 1997
CONSTANT WREB0 = 0.030 ! REST OF THE BODY, WANG ET AL. 1997
CONSTANT WLIB0 = 0.266 ! LIVER, WANG ET AL. 1997
! EXPO SURE SCENARIO FOR UNIQUE OR REPETITIVE WEEKLY OR MONTHLY
EXPOSURE
INUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK = 0.0 ! DELAY BEFORE EXPOSURE ENDS
(WEEK)
CONSTANT WEEK PERIOD = 168.0 ! NUMBER OF HOURS IN THE WEEK
(HOURS)
CONSTANT WEEK FINISH = 168.0 ! TIME EXPOSURE ENDS (HOURS)
INUMBER OF EXPOSURES PER MONTH
CONSTANT MONTH LACK = 0.0 ! DELAY BEFORE EXPOSURE BEGINS
(MONTH)
! SET FOR BACKGROUND EXPOSURE===========
!TIME CONSTANT FOR BACKGROUND EXPOSURE===========
CONSTANT Day LACK BG = 0.0 ! DELAY BEFORE EXPOSURE BEGINS
(HOUR)
CONSTANT Day PERIOD BG = 24.0 ! LENGTH OF EXPOSURE (HOUR)
!TIME CONSTANT FOR WEEKLY EXPOSURE
CONSTANT WEEK LACK BG = 0.0 ! DELAY BEFORE BACKGROUND
EXPOSURE BEGINS (WEEK)
CONSTANT WEEK PERIOD BG = 168.0 ! NUMBER OF HOURS IN THE WEEK
(HOURS)
CONSTANT WEEK FINISH BG = 168.0 ! TIME EXPOSURE ENDS (HOURS)
! CONSTANT USED IN CARDIAC OUTPUT EQUATION
CONSTANT QCC = 15.36 ! (L/KG-H), EMOND ET AL. 2004
! COMPARTMENT LIPID EXPRESSED AS THE FRACTION OF TOTAL LIPID
!Data from Emonds Thesis 2001
CONSTANT F TOTLIP = 0.8000 ! ADIPOSE TISSUE (UNITLESS)
CONSTANT B TOTLIP = 0.0057 ! BLOOD (UNITLESS)
CONSTANT RE TOTLIP = 0.0190 ! REST OF THE BODY (UNITLESS)
CONSTANT LI TOTLIP = 0.0670 ! LIVER (UNITLESS)
CONSTANT MEANLIPID = 974.0
END ! END OF THE INITIAL SECTION
DYNAMIC ! DYNAMIC SIMULATION SECTION
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
ALGORITHM IALG = 2 ! GEAR METHOD
CINTERVAL CINT = 10.0 ! COMMUNICATION INTERVAL
MAXTERVAL MAXT = 1.0e+10 ! MAXIMUM INTERVAL CALCULATION
MINTERVAL MINT = 1.0E-10 ! MINIMUM INTERVAL CALCULATION
VARIABLE T = 0.0
CONSTANT TIMELIMIT = 1.752e5 ! SIMULATION LIMIT TIME (HOUR)
CONSTANT Y0 = 0.0 ! AGE (YEARS) AT BEGINNING OF SIMULATION
CONSTANT GROWON = 1.0 ! INCLUDE BODY WEIGHT AND HEIGHT
GROWTH? (1 = YES, 0 = NO)
CINTXY = CINT
PFUNC = CINT
DAY=T/24.0
WEEK =T/168.0
MONTH =T/730.0
YEAR=Y0+T/8760.0
GYR =Y0 + growon*T/8760.0
! TIME IN DAYS
! TIME IN WEEKS
! TIME IN MONTHS
! TIME IN YEARS
! TIME FOR USE IN GROWTH EQUATION (YEARS)
DERIVATIVE ! PORTION OF CODE THAT SOLVES DIFFERENTIAL EQUATIONS
! CHRONIC OR SUBCHRONIC EXPOSURE SCENARIO =======
! NUMBER OF EXPOSURES PER DAY
DAY LACK = EXP TIME ON ! DELAY BEFORE EXPOSURE BEGINS (HOURS)
DAY PERIOD = DAY CYCLE ! EXPOSURE PERIOD (HOURS)
DAY FINISH = CINTXY ! LENGTH OF EXPOSURE (HOURS)
MONTH PERIOD = TIMELIMIT ! EXPOSURE PERIOD (MONTHS)
MONTH FINISH = EXP TIME OFF ! LENGTH OF EXPOSURE (MONTHS)
! NUMBER OF EXPOSURES PER DAY AND MONTH
DAYFINISHBG = CINTXY
MONTH LACK BG = BCK TIME ON IDELAY BEFORE BACKGROUD EXPOSURE
BEGINS (MONTHS)
MONTH PERIOD BG = TIMELIMIT ! BACKGROUND EXPOSURE PERIOD
(MONTHS)
MONTH FINISH BG = BCK TIME OFF ! LENGTH OF BACKGROUND EXPOSURE
(MONTHS)
B = 1.0-A ! FRACTION OF DIOXIN ABSORBED IN THE PORTAL FRACTION OF THE
LIVER
!HUMAN BODY WEIGHT GROWTH EQUATION========
! POLYNOMIAL REGRESSION EXPRESSION WRITTEN
! APRIL 10 2008, OPTIMIZED WITH DATA OF PELEKIS ET AL. 2001
! POLYNOMIAL REGRESSION EXPRESSION WRITTEN WITH
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
!HUH AND BOLCH 2003 FOR BMI CALCULATION
! BODY WEIGHT CALCULATION
WTO = (0.0006*GYR**3 - 0.0912*GYR**2 + 4.32*GYR + 3.652)
! BODY MASS INDEX CALCULATION
BH = -2D-5 *GYR**4+4.2D-3*GYR* *3.0-0.315 *GYR* *2.0+9.7465 *GYR+72.098
HEIGHT EQUATION FORMULATED FOR USE FROM 0 TO 70 YEARS
BHM= (BH/100.0) !HUMAN HEIGHT IN METERS (BHM)
HBMI= WT0/(BHM**2.0) ! HUMAN BODY MASS INDEX (BMI)
! ADIPOSE TISSUE FRACTION
WT0GR= WTO* 1 0e3 ! BODY WEIGHT IN GRAMS
WF0= -6.36D-20*WT0GR**4.0 +1.12D-14*WT0GR**3.0 -5.8D-10*WT0GR**2.0 +1.2D-
5*WT0GR+5.91D-2
! LIVER,VOLUME,
! APPROACH BASED ON LUECKE (2007)
WLI0= (3.59D-2 -(4.76D-7*WT0GR)+(8.50D-12*WT0GR**2.0)-(5.45D-17*WT0GR**3.0))
WRE0 = (0.91 -(WLIB 0 * WLI0+WFB 0 * WF 0+WLI0+WF 0))/( 1.0+WREB 0)
!REST OF THE BODY FRACTION; UPDATED FOR EPA
ASSESSMENT
QREF = 1 0-(QFF+QLIF) !REST OF BODY BLOOD FLOW
QTTQF = QFF+QREF+QLIF ! SUM MUST EQUAL 1
! COMPARTMENT VOLUME (L OR KG) =========
WF = WF0 * WTO ! ADIPOSE
WRE = WRE0 * WTO ! REST OF THE BODY
WLI = WLI0 * WTO ! LIVER
WB=0.075*WT0 ! BLOOD
!COMPARTMENT TISSUE BLOOD (L OR KG) =========
WFB = WFB0 * WF ! ADIPOSE
WREB = WREB0 * WRE ! REST OF THE BODY
WLIB = WLIB0 * WLI ! LIVER
! CARDIAC OUTPUT FOR THE GIVEN BODY WEIGHT
QC= QCC*(WT0**0.75) ! [L BLOOD/HOUR]
QF = QFF*QC ! ADIPOSE TISSUE BLOOD FLOW RATE [L/HR]
QLI = QLIF*QC ! LIVER TISSUE BLOOD FLOW RATE [L/HR]
QRE = QREF*QC !REST OF THE BODY BLOOD FLOW RATE [L/HR]
QTTQ = QF+QRE+QLI ! TOTAL FLOW RATE [L/HR]
! PERMEABILITY ORGAN FLOW [L/HR]=======
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
PAF =PAFF*QF
PARE = PAREF*QRE
PALI = PALIF*QLI
! ADIPOSE
! REST OF THE BODY
! LIVER TISSUE
! ABSORPTION SECTION
! INTRAVENOUS
IV = DOSEIV NM * WTO ! AMOUNT IN NMOL
MSTTBCKGR = MSTOT NMBCKGR *WT0 ! AMOUNT IN (NMOL)
MSTT = MSTOT NM * WTO ! AMOUNT IN NMOL
!REPETITIVE ORAL BACKGROUND EXPOSURE SCENARIOS
DAYEXPOSUREBG = PULSE(DAY_LACK_BG,DAY_PERIOD_BG,DAY_FINISH_BG)
WEEKEXPOSUREBG =
PUL SE(WEEK_L ACKB G, WEEKPERIODB G, WEEKFINISHB G)
MONTHEXPO SUREB G =
PUL SE(MONTH_L ACKB G,MONTH_PERIOD_B G,MONTH_FINI SH_B G)
MSTTCHBG =
(D AYEXPO SUREB G* WEEKEXPO SUREB G*MONTH_EXPO SUREB G)*MSTTBCK
GR
MSTTFRBG = MSTTBCKGR/CINT
CYCLE BG =DAY EXPOSURE BG*WEEK EXPOSURE BG*MONTH EXPOSURE BG
! CONDITIONAL ORAL EXPOSURE (BACKGROUND EXPOSURE)
IF (MSTTCH BG.EQ.MSTTBCKGR) THEN
ABSMSTT_GB= MSTTFRBG
ELSE
ABSMSTTGB = 0.0
END IF
!REPETITIVE ORAL MAIN EXPOSURE SCENARIO
DAYEXPOSURE = PULSE(DAY_LACK,DAY_PERIOD,DAY_FINISH)
WEEKEXPOSURE = PULSE(WEEK_LACK,WEEK_PERIOD,WEEK_FINISH)
MONTHEXPOSURE = PULSE(MONTH_LACK,MONTH_PERIOD,MONTH_FINISH)
MSTTCH = (DAY EXPO SURE* WEEK EXPO SURE*MONTH EXPO SURE) *MSTT
CYCLE = DAYEXPO SURE* WEEKEXPO SURE*MONTH_EXPO SURE
MSTTFR=MSTT/CINT
! CONDITIONAL ORAL EXPOSURE
IF (MSTTCH.EQ.MSTT) THEN
ABSMSTT= MSTTFR
ELSE
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
ABSMSTT = 0.
END IF
C Y CLETOT=INTEG(C Y CLE, 0.0)
! MASS Balance CHANGE IN THE LUMEN
RMSTT= -(KST+KABS)*MST+ABSMSTT +ABSMSTT GB ! RATE OF CHANGE
(NMOL/H)
MST = INTEG(RMSTT,0.) ! AMOUNT REMAINING IN GI TRACT (NMOL)
! ABSORPTION IN LYMPH CIRCULATION
LYRMLUM = KABS*MST*A
LYMLUM = INTEG(L YRMLUM, 0.0)
! ABSORPTION IN PORTAL CIRCULATION
LIRMLUM = KABS*MST*B
LIMLUM = INTEG(LIRMLUM, 0.0)
! PERCENT OF DOSE REMAINING IN THE GI TRACT
PRCTremainGIT = 100.0*MST/(MSTT+lE-30)
!IV ABSORTPION SCENARIO
IVR= IV/PFUNC ! RATE FOR IV INFUSION IN BLOOD
EXPIV= IVR * (1 O-STEP(PFUNC))
IVDOSE = integ(EXPIV,0.0)
! SYSTEMIC BLOOD COMPARTMENT
! MODIFICATION OCT 8 2009
CB=(QF * CFB+QRE * CREB+QLI * CLIB+EXPIV+LYRMLUM)/(QC+CLURI) !
CA = CB ! CONCENTRATION (NMOL/L)
! CB=(QF * CFB+QRE* CREB+QLI* CLIB+EXPIV+LYRMLUM-RAURI)/QC !
! CA = CB ! CONCENTRATION (NMOL/L)
!URINARY EXCRETION BY KIDNEY
! MODIFICATION OCT 8 2009
RAURI = CLURI *CB
AURI = INTEG(R AURI, 0.0)
! CONCENTRATION UNIT
PRCT B = 100.0*CB/(MSTT+lE-30) ! PERCENT OF DOSE
CBSNGKGLIADJ = CB*MW/(0.55*B_TOTLIP) !serum concentration in lipid adjust (PG/G
LIPID=PPT)
CBPPT = CBSNGKGLIADJ
CBNGKG = CB*MW
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CBpptRH = CB*MW* 10000/(0.55*MEANLIPID) ! SERUM CONCENTRATION IN LIPID
ADJUST (PG/G LIPID=PPT)
AUCCB SNGKGLIAD J=INTEG(CB SNGKGLIAD J,0.0)
! ADIPOSE TISSUE COMPARTMENT
RAFB= QF*(CA-CFB)-PAF*(CFB-CF/PF) ! (NMOL/HR)
AFB = INTEG(RAFB ,0.0) ! (NMOL)
CFB = AFB/WFB ! (NMOL/KG)
!TISSUE SUBCOMPARTMENT
RAF = PAF*(CFB-CF/PF) ! (NMOL/HR)
AF = INTEG(RAF, 0.0) ! (NMOL)
CF = AF/WF ! (NMOL/KG)
IPO ST SIMULATION UNIT CONVERSION
CFTOTAL = (AF + AFB)/(WF + WFB) ! TOTAL CONCENTRATION NMOL/ML
PRCTF = 100.0*CFTOTAL/(MSTT+1E-30)
CFNGKG =CFTOTAL*MW
IREST OF THE BODY COMPARTMENT========
RAREB= QRE*(CA-CREB)-PARE*(CREB-CRE/PRE) ! (NMOL/HR)
AREB = INTEG(RAREB ,0.0) !(NMOL)
CREB = AREB/WREB ! (NMOL/KG)
!TISSUE SUBCOMPARTMENT
RARE = PARE* (CREB -CRE/PRE) ! (NMOL/HR)
ARE = INTEG(RARE,0.0) ! (NMOL)
CRE = A RE/W RE ! (NMOL/KG)
IPO ST SIMULATION UNIT CONVERSION
CRETOTAL = (ARE + AREB)/(WRE + WREB) ! TOTAL CONCENTRATION IN NMOL/ML
PRCT RE = 100.0*CRETOTAL/(MSTT+1E-30) ! PERCENT OF DOSE
!LIVER COMPARTMENT
!TISSUE BLOOD SUBCOMPARTMENT
RALIB = QLI*(CA-CLIB)-PALI*(CLIB-CFLLIR)+LIRMLUM ! (NMOL/HR)
ALIB = INTEG(RALIB ,0.0) ! (NMOL)
CLIB = ALIB/WLIB
!TISSUE SUBCOMPARTMENT
RALI = PALI * (CLIB -CFLLIR)-REXCLI ! (NMOL/HR)
ALI = INTEG(RALI, 0.0) ! (NMOL)
CLI = ALI/WLI ! (NMOL/KG)
!FREE TCDD IN LIVER
! MODIFICATION OCTOBER 8 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
CFLLI= IMPLC(CLI-(CFLLIR*PLI+(LIBMAX*CFLLIR/(KDLI+CFLLIR)) &
+((CYP1A2_103*CFLLIR/(KDLI2+CFLLIR)*PAS_INDUC)))-CFLLI,CFLLIO) !
CONCENTRATION OF FREE TCDD IN LIVER
CFLLIR=DIM(CFLLI, 0.0)
!MODIFIED FROM:
! PARAMETER (LIVER1RMN = 1.0E-30)
! CFLLI= IMPLC(CLI-(CFLLIR*PLI+(LIBMAX*CFLLIR/(KDLI+CFLLIR & !
+LI VER_ 1 RMN))+((C YP1 A2_ 103* CFLLIR/(KDLI2+CFLLIR &
! +LIVER_1RMN)*PAS_INDUC)))-CFLLI,CFLLI0)
! CFLLIR=DIM(CFLLI, 0.0)
CBNDLI= LIBMAX*CFLLIR/(KDLI+CFLLIR) !CONC OF TCDD BOUDN TO AhR
!CBNDLI= LIBM AX* CFLLIR/(KDLI+CFLLIR+LIVER_1 RMN) !CONC BIND
IPO ST SIMULATION UNIT CONVERSION
CLITOTAL = (ALI + ALIB)/(WLI + WLIB) ! TOTAL CONCENTRATION IN NMOL/ML
PRCTLI = 100.0*CLITOTAL/(MSTT+1.0E-30)
rec_occ_AHR= 100.0*CFLLIR/(KDLI+CFLLIR+1.0) ! PERCENT BOUND TO AhR
OCCUPANCY
PROT_occ_lA2= 100.0*CFLLIR/(KDLI2+CFLLIR) ! PERCENT BOUND TO 1A2
OCCUPANCY
CLINGKG= CLITOTAL*MW ! [NG TCDD/KG]
CBNDLINGKG = CBNDLI*MW
! FRACTION INCREASE OF INDUCTION OF CYP1A2
fold_ind=CYP 1 A2_l OUT/CYP1 A2_l A2
V ARIATION OF AC =(CYP 1 A2_l OUT-CYP1 A2_l A2)/CYP 1 A2_l A2
! VARIABLE ELIMINATION BASED ON THE CYP1A2
KBILELIT = Kel v * V ARI ATION OF AC!
REXCLI = KBILE LI T*CFLLIR*WLI ! DOSE-DEPENDENT RATE OF BILLIARY
EXCRETION OF DIOXIN
EXCLI = INTEG(REXCLI,0.0) ! TOTAL AMOUNT OF DIOXIN EXCRETED
! CHEMICAL IN CYP450 (1A2) COMPARTMENT
!PARAMETER FOR INDUCTION OF CYP1A2
CYP1A21KINP = CYP 1 A2_ 1KOUT*CYP 1 A2_ 1 OUTZ ! BASAL RATE OF CYP1A2
PRODUCTION SET EQUAL TO BASAL RATE OF DEGRDATION AT STEADY STATE
! MODIFICATION OCTOBER 8 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
CYP1A210UT =INTEG(CYP1A2_1KINP * (1.0 + CYP1A21EMAX *(CBNDLI+1.0e-
30)**HILL&
/(CYP1 A2_ 1 EC 5 0 * *HILL + (CBNDLI+1.0e-30)**HILL)) &
- CYP1A2_1K0UT*CYP1A2_10UT, CYP1A210UTZ) ! LEVELS OF CYP1A2
! MODEIFIED FROM:
!PARAMETER (CYP1A21RMN = le-30)
ICYP1A210UT =INTEG(CYP 1 A2_ 1KINP * (1 + CYP1A21EMAX *(CBNDLI &
! +CYP 1 A2_ 1RMN)**HILL/(CYP 1 A2_ 1 EC50 + (CBNDLI + CYP1A2_1RMN)**HILL) &
! +CYP1A21RMN) - CYP1A2_1K0UT*CYP1A2_1&
! OUT, CYP 1 A2_ 1 OUTZ)
! EQUATIONS INCORPORATING DELAY OF CYP1A2 PRODUCTION (NOT USED IN
SIMULATIONS)
CYP1A21R02 = (CYP1A210UT - CYP1A2102)/ CYP1A21TAU
CYP1A2102 =INTEG(CYP 1 A2_ 1R02, CYP1A21A1)
CYP1A21R03 = (CYP1A2102 - CYP1A2103)/ CYP1A21TAU
CYP1A2103 =INTEG(CYP 1 A2_ 1R03, CYP1A21A2)
!CHECK MASS BALANCE
BDOSE= LYMLUM+LIMLUM+IVDO SE
BMASSE = EXCLI+AURI+AFB+AF+AREB+ARE+ALIB+ALI
BDIFF = BDO SE-BMAS SE
! BODY BURDEN IN TERMS OF CONCENTRATION (NG/KG)
BBNGKG = (AFB+AF+AREB+ARE+ALIB+ALI)*MW/WTO !
! COMMAND END OF THE SIMULATION
TERMT (T.GE. TIMELIMIT, 'Time limit has been reached.')
END ! END OF THE DERIVATIVE SECTION
END ! END OF THE DYTNAMIC SECTION
END ! END OF THE PROGRAM
C.2.1.2. Input File
% base file name = "TESTJULY2009.m"
%clear @variable
output @clear
prepare @clear year T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
CBNGKG
%output @all
% PARAMETERS FOR SIMULATION
CINT = 1 %0.5
EXP TIME ON = 0. % TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXP TIME OFF = 613200 %324120 % HOUR/YEAR !TIME AT WHICH EXPOSURE
ENDS (HOUR)
DAY_C YCLE = 24 % NUMBER OF HOURS BETWEEN DOSES (HOUR)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-20 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
BCK TIME ON = 613200 %324120 % TIME AT WHICH BACKGROUND
EXPOSURE BEGINS (HOUR)
BCKTIMEOFF = 613200 %324120 % TIME AT WHICH BACKGROUND
EXPOSURE ENDS (HOUR)
TIMELIMIT =613200 %324120 %324120 % SIMULATION TIME LIMIT (HOUR)
MSTOTBCKGR = 0. % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 9.97339283634997E-07 % ORAL DAILY EXPOSURE DOSE (NG/KG)
DOSEIV = 0 %NG/KG
% oral dose oral dose oral dose
MEANLIPID =730 %
PAS_INDUC= 1 % INDUCTION INCLUDED? (1=YES, 0=NO)
C.2.2. Human Gestational Model
C.2.2.1. Model Code
PROGRAM: 'Three Compartment PBPK Model for TCDD in Human (Gestation)'
! Parameters were change may 16, 2002
! Come from { 8MAI CHR PRE-EXP GD}
! Come from {12_Mouse_GD}file
! { {IMPORTANT-IMPORTANT-IMPORTANT-IMPORTANT} }
! REDUCTION OF MOTHER AND FETUS COMPARTMENT
! 2M_R_TCDDJUL Y2002 ////(JULY 18,2002)////
! TCDD_RED_4Species_2003_4 ////(APR 8 ,2003)////
! TCDD_RED_4Species_2003_9 ////(APR 17,2003)////
!TCDD_RED_4Species_2003_12 ////(APR 17,2003)////
! APRIL 18 2003
! TCDD_4C_4SP_2003 ////(APR 18,2003)////
! was "Gest 4 species 1 .csl" but update July 2009
! GEST_HUM_0_45 Y_4_ICF_afterKKfix_v3_humangestational. csl
! HUMGESTATIONALICFF083109. csl
! HUMGE ST ATION ALICFF100709. csl
ILegend/Legend/Legend/Legend/Legend/Legend/Legend/Legend/
!Legend for this PBPK model
IMating: control the tenure of exchange between fetus and
IMother and also control imitated tissue growth
!Control: WTFE, WPLA0, QPLAF
! (for rat, mouse, human, and monkey)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-21 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! Control transfer from mother to fetus and fetus to mother by TRANSTIMEON
! SWITCHtrans = 0 NO TRANSFER
! SWITCHtrans = 1 TRANSFER OCCURS
! These switches are also controlled by mating parameters
INITIAL !
! SIMULATION PARAMETERS
CONSTANT PARA ZERO = le-30
CONSTANT EXP TIME ON =0.0 !TIME AT WHICH EXPOSURE BEGINS
(HOURS)
CONSTANT EXP TIME OFF = 530.0 !TIME AT WHICH EXPOSURE ENDS (HOURS)
CONSTANT DAY CYCLE = 24.0 INUMBER OF HOURS BETWEEN DOSES
(HOURS)
CONSTANT BCK TIME ON =0.0 !TIME AT WHICH BACKGROUND EXPOSURE
BEGINS (HOURS)
CONSTANT BCK TIME OFF =0.0 !TIME AT WHICH BACKGROUND EXPOSURE
ENDS (HOURS)
CONSTANT TRANSTIME ON =0.0 ! CONTROL TRANSFER FROM MOTHER TO
FETUS AT 9 WEEKS OR 1512 HOURS OF GESTATION
! INTRAVENOUS SEQUENCY
CONSTANT IV LACK = 0.0
CONSTANT IV PERIOD = 0.0
! PREGNANCY PARAMETER
CONSTANT MATTING = 0.0 BEGINNING OF MATING (HOUR)
CONSTANT PFETUS = 4.0 IPARTITION COEFFICIENT
CONSTANT CLPLA FET = 1 0e-3 ! CLEARANCE TRANSFER FOR MOTHER TO
FETUS (L/HR)
!CONSTANT EXPOSURE CONTROL
! ACUTE, SUBCHRONIC, CHRONIC EXPOSURE =====
!OR BACKGROUND EXPOSURE (IN THIS CASE 3 TIMES A DAY)===
CONSTANT MSTOTBCKGR =0.0 ! ORAL BACKGROUND EXPOSURE DOSE
(NG/KG)
CONSTANT MSTOT =0.0 ! ORAL EXPOSURE DOSE (NG/KG)
!ORAL ABSORPTION
! MSTT= MSTOT/1000 *WT0 *1/322*1000 ! AMOUNT IN NMOL
MSTOT NM = MSTOT/MW ! CONVERTS THE DOSE TO NMOL/KG
!INTRAVENOUS ABSORPTION
CONSTANT DOSEIV = 0.0 ! INJECTED DOSE (NG/KG)
DOSEIV NM = DOSEIV/MW ! CONVERTS THE INJECTED DOSE TO NMOL/KG
CONSTANT DOSEIVLATE = 0.0 !INJECTED DOSE LATE (UG/KG)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
DOSEIVNMlate = DOSEIVLATE/MW ! AMOUNT IN NMOL/G
!INITIAL GUESS OF THE FREE CONCENTRATION IN THE LIGAND
(COMPARTMENT INDICATED BELOW)====
CONSTANT CFLLIO = 0.0 ILIVER (NMOL/L)
CONSTANT CFLPLA0 = 0.0 !PLACENTA (NMOL/L)
! BINDING CAPACITY (AhR) FOR NON LINEAR BINDING (COMPARTMENT
INDICATED BELOW) (NMOL/L) ===
CONSTANT LIBMAX =0.35 ! LIVER (NMOL/L)
CONSTANT PLABMAX =0.2 ! TEMPORARY PARAMETER
!PROTEIN AFFINITY CONSTANTS (1A2 OR AhR, COMPARTMENT INDICATED
BELOW) (NMOL/ML)===
CONSTANT KDLI =0.1 ILIVER (AhR) (NMOL/L), WANG ET AL. 1997
CONSTANT KDLI2 = 40.0 ILIVER (1A2) (NMOL/L), EMOND ET AL. 2004
CONSTANT KDPLA = 0.1 I ASSUME IDENTICAL TO KDLI (AhR)
I EXCRETION AND ABSORPTION CONSTANT
CONSTANT KST = 0.01 I GASTRIC RATE CONSTANT (HR-1), EMOND ET AL.
2005
CONSTANT KABS =0.06 I INTESTINAL ABSORPTION CONSTANT (HR-1),
EMOND ET AL. (2005)
I INTERSPECIES ELIMINATION CONSTANT
I TEST ELIMINATION VARIABLE, EMOND ET AL. 2005
CONSTANT KELV = l.le-3 I4.0D-3 I INTERSPECIES VARIABLE
ELIMINATION CONSTANT (1/HOUR)
I ELIMINATION CONSTANTS
CONSTANT CLURI = 4.17e-8 I URINARY CLEARANCE (L/HR), EMOND ET AL.
2005
I CONSTANT TO DIVIDE THE ABSORPTION INTO LYMPHATIC AND PORTAL
FRACTIONS
CONSTANT A =0.7 I LYMPHATIC FRACTION, WANG ET AL. 1997
I PARTITION COEFFICIENTS
CONSTANT PF = 1 0e2 I ADIPOSE TISSUE/BLOOD, WANG ET AL. 1997
CONSTANT PRE =1.5 I REST OF THE BODY/BLOOD, WANG ET AL. 1997
CONSTANT PLI = 6.0 I LIVER/BLOOD, WANG ET AL. 1997
CONSTANT PPLA =1.5 I TEMPORARY PARAMETER NOT CONFIGURED,
WANGETAL. 1997
I PARAMETER FOR INDUCTION OF CYP 1A2, WANG ET AL. 1997
CONSTANT PAS INDUC =1.0 I INCLUDE INDUCTION? (1 = YES, 0 = NO)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CONSTANT CYP1A210UTZ = 1.6e3 ! DEGRADATION CONCENTRATION
CONSTANT OF 1A2 (NMOL/L)
CONSTANT CYP1A21A1 = 1.6e3 ! BASAL CONCENTRATION OF 1A1 (NMOL/L)
CONSTANT CYP1A21EC50 = 1.3e2 ! DISSOCIATION CONSTANT TCDD-CYP1A2
(NMOL/L)
CONSTANT CYP1A21A2 = 1.6e3 IBASAL CONCENTRATION OF 1A2 (NMOL/ML)
CONSTANT CYP1A21KOUT =0.1 ! FIRST ORDER RATE OF DEGRADATION (H-l)
CONSTANT CYP1A21TAU =0.25 !HOLDING TIME (H)
CONSTANT CYP1A21EMAX = 9.3e3 ! MAXIMUM INDUCTION OVER BASAL
EFFECT (UNFILESS)
CONSTANT HILL = 0.6 !HILL CONSTANT; COOPERATIVELY LIGAND
BINDING EFFECT CONSTANT (UNITLESS)
!DIFFUSIONAL PERMEABILITY FRACTION, WANG ET AL (1997)
CONSTANT PAFF =0.12 ! ADIPOSE (UNITLESS)
CONSTANT PAREF = 0.03 ! REST OF THE BODY (UNITLESS)
CONSTANT PALIF = 0.35 ! LIVER (UNITLESS)
CONSTANT PAPLAF = 0.3 ! OPTIMIZED PARAMETER
!TISSUE BLOOD FLOW EXPRESSED AS A FRACTION OF CARDIAC OUTPUT,
KRISHNAN 2007
CONSTANT QFF = 0.05 ! ADIPOSE TISSUE BLOOD FLOW FRACTION
(UNITLESS), KRISHNAN 2008
CONSTANT QLIF = 0.26 ! LIVER (UNITLESS), KRISHNAN 2008
! ===FRACTION OF TISSUE BLOOD WEIGHT Wang et al . (1997)
CONSTANT WFB0 =0.050 ! ADIPOSE TISSUE, WANG ET AL. 1997
CONSTANT WREB0 = 0.030 !REST OF THE BODY, WANG ET AL. 1997
CONSTANT WLIB0 = 0.266 ILIVER, WANG ET AL. 1997
CONSTANT WPLAB0 =0.500 ! ASSUME HIGHLY VASCULARIZED
! EXPOSURE SCENARIO FOR UNIQUE OR REPETITIVE WEEKLY OR MONTHLY
EXPOSURE
! NUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK = 0.0 IDELAY BEFORE EXPOSURE ENDS (WEEK)
CONSTANT WEEK PERIOD =168.0 ! NUMBER OF HOURS IN THE WEEK
(HOURS)
CONSTANT WEEK FINISH = 168.0 ! TIME EXPOSURE ENDS (HOURS)
! NUMBER OF EXPOSURES PER MONTH
CONSTANT MONTH LACK =0.0 IDELAY BEFORE EXPOSURE BEGINS
(MONTHS)
!======= CONSTANT FOR BACKGROUND EXPOSURE===========
CONSTANT Day LACK BG = 0.0 ! DELAY BEFORE EXPOSURE BEGINS
(HOURS)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CONSTANT DayPERIODBG = 24.0 !LENGTH OF EXPOSURE (HOURS)
! NUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK BG =0.0 !DELAY BEFORE BACKGROUD EXPOSURE
BEGINS (WEEK)
CONSTANT WEEK PERIOD BG =168.0 ! NUMBER OF HOURS IN THE WEEK
(HOURS)
CONSTANT WEEK FINISH BG = 168.0 !TIME EXPOSURE ENDS (HOURS)
! CONSTANT USED IN CARDIAC OUTPUT EQUATION
CONSTANT QCC = 15.36 ![L/KG-H], EMOND ET AL. 2004
! COMPARTMENT LIPID EXPRESSED AS THE FRACTION OF TOTAL LIPID
!Data from Emonds Thesis 2001
CONSTANT F TOTLIP
CONSTANT B TOTLIP
CONSTANT RE TOTLIP
CONSTANT LI TOTLIP
CONSTANT PLA TOTLIP
0.8000
0.0057
= 0.0190
0.0670
= 0.019
CONSTANT FETUS TOTLIP = 0.019
! ADIPOSE TISSUE (UNITLESS)
! BLOOD (UNITLESS)
! REST OF THE BODY (UNITLESS)
! LIVER (UNITLESS)
! PLACENTA (UNITLESS)
! FETUS (UNITLESS)
CONSTANT MEANLIPID
= 974
END ! END OF THE INITIAL SECTION
DYNAMIC ! DYNAMIC SIMULATION SECTION
ALGORITHM IALG = 2
CINTERVAL CINT = 0.1
MAXTERVAL MAXT = 1.0e+10
MINTERVAL MINT = 1.0E-10
VARIABLE T = 0.0
CONSTANT TIMELIMIT = 100
CONSTANT Y0 = 0.0
SIMULATION
CONSTANT GROWON = 1.0
GROWTH? (1=YES, 0=NO)
! GEAR METHOD
COMMUNICATION INTERVAL
! MAXIMUM CALCULATION INTERVAL
MINIMUM CALCULATION INTERVAL
! SIMULATION LIMIT TIME (HOUR)
! AGE (YEARS) AT BEGINNING OF
INCLUDE BODY WEIGHT AND HEIGHT
CINTXY = CINT
PFUNC = CINT
!TIME TRANSFORMATION
DAY= T/24.0
WEEK =T/168.0
YEAR=Y0+T/8760.0
! TIME IN YEARS
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
GYR =Y0 + growon*T/8760.0
! TIME FOR USE IN GROWTH EQUATION
DERIVATIVE ! PORTION OF CODE THAT SOLVES DIFFERENTIAL EQUATIONS
!====== CHRONIC OR SUBCHRONIC EXPOSURE SCENARIO =======
! NUMBER OF EXPOSURES PER DAY
DAY LACK = EXP TIME ON ! DELAY BEFORE EXPOSURE BEGINS (HOURS)
DAY PERIOD = DAY CYCLE ! EXPOSURE PERIOD (HOURS)
DAY FINISH = CINTXY ! LENGTH OF EXPOSURE (HOURS)
MONTH PERIOD = TIMELIMIT ! EXPOSURE PERIOD (MONTHS)
MONTH FINISH = EXP TIME OFF ! LENGTH OF EXPOSURE (MONTHS)
! NUMBER OF EXPOSURES PER DAY AND MONTH
DAYFINISHBG = CINTXY
MONTH LACK BG = BCK TIME ON ! DELAY BEFORE BACKGROUND
EXPOSURE BEGINS (MONTHS)
MONTH PERIOD BG = TIMELIMIT BACKGROUND EXPOSURE PERIOD
(MONTHS)
MONTH FINISH BG = BCK TIME OFF !LENGTH OF BACKGROUND EXPOSURE
(MONTHS)
! INTRAVENOUS LATE
IVFINISH = CINTXY
B = 1-A ! FRACTION OF DIOXIN ABSORBED IN THE PORTAL FRACTION OF THE
LIVER
! MOTHER BODY WEIGHT GROWTH EQUATION
! MODIFICATION TO ADAPT THIS MODEL AT HUMAN MODEL
! BECAUSE LINEAR DESCRIPTION IS NOT GOOD ENOUGH FOR MOTHER GROWTH
! MOTHER BODY WEIGHT GROWTH
! HUMAN BODY WEIGHT (0 TO 45 YEARS)
! POLYNOMIAL REGRESSION EXPRESSION WRITTEN
! APRIL 10 2008, OPTIMIZED WITH DATA OF PELEKIS ET AL. 2001
! POLYNOMIAL REGRESSION EXPRESSION WRITTEN WITH
!HUH AND BOLCH 2003 FOR BMI CALCULATION
! BODY WEIGHT CALCULATION. UNIT IN KG FOR GESTATIONAL PORTION
WTO = (0.0006*GYR**3 - 0.0912*GYR**2 + 4.32*GYR + 3.652)
!BODY MASS INDEX CALCULATION
BH = -2D-5 *GYR**4+4.2D-3*GYR* *3.0-0.315 *GYR* *2.0+9.7465 *GYR+72.098
!HEIGHT EQUATION FORMULATED FOR USE FROM 0 TO 70 YEARS
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-26 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
BHM= (BH/100.0)! HUM AN HEIGHT IN METER (BHM)
HBMI= WT0/(BHM**2.0) ! HUMAN BODY MASS INDEX (BMI)
! MODIFICATION IN KG
RTESTGEST= T-MATTING ! STARTING TIME FOR FETAL GROWTH
TESTGEST=DIM(RTESTGEST,0.0)
! GROWTH OF FETAL TISSUE
GESTATTION_FE=((4d-15*TESTGEST**4 -3d-l 1*TESTGEST**3 + ld-7*TESTGEST**2 -
8d-5*TESTGEST +0.0608))
WTFER= DIM(GESTATTION_FE,0.0) ! FETAL COMPARTMENT WEIGHT
WTFE= WTFER
!///////////////////////////////////////////////////////////////////////
! FAT GROWTH EXPRESSION LINEAR DURING PREGNANCY
! FROM O'FLAHERTYl 992
!///////////////////////////////////////////////////////////////////////
WT0GR= WTO* 1 0e3 ! MOTHER BODY WEIGHT IN G
WF0 =(-6.36D-20*WT0GR**4.0 + 1.12D-14*WT0GR**3.0 &
-5.8D-10*WT0GR**2.0+1.2D-5*WT0GR+5.91D-2) ! MOTHER FAT
COMPARTMENT GROWTH
!///////////////////////////////////////////////////////////////////////
! WPLA PLACENTA GROWTH EXPRESSION, SINGLE EXPONENTIAL WITH OFFSET
! FROM O'FLAHERTY l 992 ! FOR EACH PUP
!///////////////////////////////////////////////////////////////////////
! SAME EQUATION THEN THE FORST MODEL. BODY WEIGHT KEPT IN G
! A CORRECTION FOR THE BODY WEIGHT (WTO(KG)*1000 = WTOGR)
WPL A0N_HUM AN= (8 5 0 * exp(-9.434* (exp(-5.23 d-4 * (TESTGEST)))))
WPLA0R = WPLA0NHUMAN/WT0GR
WPLA0W = DIM(WPLA0R, 0.0) ! PLACENTA WEIGHT
WPLA0=WPLA0W
!///////////////////////////////////////////////////////////////////////
! QPLA PLACENTA GROWTH EXPRESSION, DOUBLE EXPONENTIAL WITH OFFSET
! FROM O'FLAHERTY l 992
!///////////////////////////////////////////////////////////////////////
QPLAF_HUMAN= SWITCH_trans*((ld-10*TESTGEST**3.0-5D-7*TESTGEST**2.0
+0.0017*TESTGEST+1.1937)/QC)
GEST_QPLAF=DIM(QPLAF HUMAN,0.0) ! PLACENTA BLOOD FLOW RATE
QPLAF =GEST_QPLAF
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! LIVER,VOLUME (HUMAN 0 TO 70 YEARS)
! APPROACH BASED ON LUECKE (2007)
WLI0= (3.59D-2 -(4.76D-7*WT0GR)+(8.50D-12*WT0GR**2.0)-(5.45D-17*WT0GR**3.0)) !
LIVER VOLUME IN GROWING HUMAN
! VARIABILITY OF REST OF THE BODY DEPENDS ON OTHER ORGAN
WRE0 = (0.91-(WLIB0*WLI0+WFB0*WF0+ WPLAB0*WPLA0 + WLI0 + WF0 +
WPL AO))/(1+WREB 0)
QREF = 1 -(QFF+QLIF+QPLAF) !REST BODY BLOOD FLOW (ML/HR)
QTTQF = QFF+QREF+QLIF+QPLAF ! SUM MUST EQUAL 1
! COMPARTMENT TISSUE BLOOD VOLUME (L) =========
WF = WF0 * WTO ! ADIPOSE TISSUE
WRE = WRE0 * WTO ! REST OF THE BODY
WLI = WLI0 * WTO ! LIVER
WPLA= WPL AO* WTO ! PLACENTA
! COMPARTMENT TISSUE VOLUME (L) =========
WFB = WFB0 * WF ! ADIPOSE TISSUE
WREB = WREB0 * WRE ! REST OF THE BODY
WLIB = WLIB0 * WLI ! LIVER
WPLAB = WPLAB0* WPLA ! PLACANTA
! TOTAL VOLUME OF COMPARTMENT (L)======
WFT = WF ! TOTAL ADIPOSE TISSUE
WRET = WRE ! TOTAL REST OF THE BODY
WLIT = WLI ! TOTAL LIVER TISSUE
WPLAT= WPLAB ! TOTAL PLACENTA TISSUE
! CONSTANT USED IN CARDIAC OUTPUT EQUATION
! UNIT CHANGED ON JULY 14 2009 (L/HR)
QC= QCC*(WT0)**0.75
QF = QFF*QC ! ADIPOSE TISSUE BLOOD FLOW RATE (L/HR)
QLI = QLIF*QC ! LIVER TISSUE BLOOD FLOW RATE (L/HR)
QRE = QREF*QC !REST OF THE BODY BLOOD FLOW RATE (L/HR)
QPLA = QPLAF*QC !PLACENTA TISSUE BLOOD FLOW RATE (L/HR)
QTTQ = QF+QRE+QLI+QPLA ! TOTAL FLOW RATE (L/HR)
! ========= DIFFUSIONAL permeability factors fraction organ flow
PAF = PAFF*QF ! ADIPOSE TISSUE BLOOD FLOW RATE (L/HR)
PARE = PAREF*QRE ! REST OF THE BODY BLOOD FLOW RATE (L/HR)
PALI = PALIF*QLI ! LIVER TISSUE BLOOD FLOW RATE (L/HR)
PAPLA = PAPLAF*QPLA ! PLACENTA TISSUE BLOOD FLOW RATE (L/HR)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! ABSORPTION SECTION
! ORAL
! INTRAPERITONEAL
! SUBCUTANEOUS
! INTRAVENOUS
BACKGROUND EXPOSURE
! EXPO SURE FOR STEADY STATE CONSIDERATION
!REPETITIVE EXPOSURE SCENARIO
MSTOT NMBCKGR = MSTOTBCKGR/322 ! AMOUNT IN NMOL/G
MSTTBCKGR =MSTOT_NMBCKGR *WT0
DAYEXPOSUREBG = PULSE(DAY_LACK_BG,DAY_PERIOD_BG,DAY_FINISH_BG)
WEEKEXPOSUREBG =
PUL SE(WEEK_L ACKB G, WEEKPERIODB G, WEEKFINISHB G)
MONTHEXPO SUREB G =
PUL SE(MONTH_L ACKB G,MONTH_PERIOD_B G,MONTH_FINI SH_B G)
MSTTCHBG =
(DAY_EXPOSURE_BG*WEEK_EXPOSURE_BG*MONTH_EXPOSURE_BG)*MSTTBCK
GR
MSTTFRBG = MSTTBCKGR/CINT
C Y CLEBG =D AYEXPO SUREBG* WEEKEXPO SUREB G*MONTH_EXPO SUREB G
! CONDITIONAL ORAL EXPOSURE (BACKGROUND EXPOSURE)
IF (MSTTCH BG.EQ.MSTTBCKGR) THEN
ABSMSTT_GB= MSTTFRBG
ELSE
ABSMSTTGB = 0.0
END IF
C YCLETOTB G=INTEG(C Y CLEB G, 0.0)
!MULTIROUTE EXPOSURE
!REPETITIVE EXPOSURE SCENARIO
MSTT= MSTOT NM * WTO ! AMOUNT IN NMOL
DAYEXPOSURE = PULSE(DAY_LACK,DAY_PERIOD,DAY_FINISH)
WEEKEXPOSURE = PULSE(WEEK_LACK,WEEK_PERIOD,WEEK_FINISH)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
MONTHEXPOSURE = PUL SE(MONTH_L ACK,MONTH_PERIOD ,MONTH_FINI SH)
MSTTCH = (D AYEXPO SURE* WEEKEXPO SURE*MONTH_EXPO SURE) *MSTT
MSTTFR = MSTT/CINT
CYCLE = DAYEXPO SURE* WEEKEXPO SURE*MONTH_EXPO SURE
SUMEXPEVENT= INTEG (CYCLE,0.0) INUMBER OF CYCLES GENERATED DURING
SIMULATION
! CONDITIONAL ORAL EXPOSURE
IF (MSTTCH.EQ.MSTT) THEN
ABSMSTT= MSTTFR
ELSE
ABSMSTT = 0.0
END IF
C Y CLETOT=INTEG(C Y CLE, 0.0)
! MASS CHANGE IN THE LUMEN
RMSTT= -(KST+KABS)*MST + ABSMSTT +ABSMSTT GB ! RATE OF CHANGE
(NMOL/H)
MST = INTEG(RMSTT,0.0) ! AMOUNT REMAINING IN DUODENUM
(NMOL)
! ABSORPTION IN LYMPH CIRCULATION
LYRMLUM = KABS*MST*A
LYMLUM = INTEG(L YRMLUM, 0.0)
! ABSORPTION IN PORTAL CIRCULATION
LIRMLUM = KABS*MST*B
LIMLUM = INTEG(LIRMLUM, 0.0)
!IV ABSORPTION SCENARIO
IV= DOSEIV NM * WTO ! AMOUNT IN NMOL
IVR= IV/PFUNC ! RATE FOR IV INFUSION IN BLOOD
EXPIV= IVR * (1-STEP(PFUNC))
IVDOSE = integ(EXPIV,0.0)
!IV LATE IN THE CYCLE
! MODIFICATION JANUARY 13 2004
IVRlateR = DOSEIVNMlate*WT0
IV_EXPOSURE=PULSE(IV_LACK,IV_PERIOD,IV_FINISH)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
IV lateT = IVEXPOSURE *IV_RlateR
I V_1 ate = IVlateT/CINT
SUMEXPEVENTIV= integ(IV_EXPO SURE, 0.0) INUMBER OF CYCLE GENERATE
DURING SIMULATION
! SYSTEMIC BLOOD COMPARTMENT
! MODIFICATION OCT 8 2009
CB=(QF*CFB+QRE*CREB+QLI*CLIB+EXPIV+LYRMLUM+QPLA*CPLAB+IV_late)/(QC
+CLURI) !
CA = CB ! CONCENTRATION (NMOL/L)
!CB=(QF*CFB+QRE*CREB+QLI*CLIB+EXPIV+LYRMLUM+QPLA*CPLAB+IV_late-
RAURI)/QC ! (NMOL/L)
! URINARY EXCRETION BY KIDNEY
! MODIFICATION OCT 8 2009
RAURI = CLURI *CB
AURI = INTEG(R AURI, 0.0)
! RAURI = CLURI * CRE
! AURI = INTEG(RAURI, 0.0)
!UNIT CONVERSION POST SIMULATION
CONSTANT MW=322 IMOLECULAR WEIGHT (NG/NMOL)
CONSTANT SERBLO = 0.55
CONSTANT UNITCORR = 1.0e3
CBSNGKGLIADJ = CB*MW/(0.55*B_TOTLIP) !NG SERUM LIPID ADJUSTED/KG
AU CB S_N GKGLI AD J=integ(CB SN GKGLI AD J, 0.)
CBNGKG= CB*MW ING/KG
PRCT B = 100.0*CB/(MSTT+lE-30) IPERCENT OF ORAL DOSE IN BLOOD
PRCT BIV = 100.0*CB/(IV_R1 ateR+1E-30) ! PERCENT OF IV DOSE IN BLOOD
! ADIPOSE COMPARMTENT
! TISSUE BLOOD SUB COMPARTMENT
RAFB= QF * (C A-CFB)-PAF * (CFB-CF/PF) !(NMOL/H)
AFB = INTEG(RAFB ,0.0) ! (NMOL)
CFB = AFB/WFB ! (NMOL/L)
!TISSUE SUBCOMPARTMENT
RAF = PAF*(CFB-CF/PF) ! (NMOL/H)
AF = INTEG(RAF,0.0) !(NMOL)
CF = AF/WF ! (NMOL/L)
!UNIT CONVERSION POST SIMULATION
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CFTOTAL= (AF + AFB)/(WF + WFB) ! TOTAL CONCENTRATION IN NMOL/ML
PRCT F = 100.0*CFTOTAL/(MSTT+1E-30) IPERCENT OF ORAL DOSE IN FAT
PRCT FIV = 100.0*CFTOTAL/(IV_RlateR+lE-30) IPERCENT OF IV DOSE IN FAT
CFNGKG=CFTOTAL*MW ! FAT CONCENTRATION IN NG/KG
AU CF_N GKGH=integ(CFN GKG, 0.)
IREST OF THE BODY COMPARTMENT
! TISSUE BLOOD SUB COMPARTMENT
RAREB= QRE *(CA-CREB)-PARE*(CREB-CRE/PRE)
!(NMOL/H)
AREB = INTEG(RAREB ,0.0)
CREB = AREB/WREB
!TISSUE SUBCOMPARTMENT
RARE = PARE* (CREB - CRE/PRE)
ARE = INTEG(RARE,0.0)
CRE = A RE/W RE
ARETOT = ARE +AREB
!(NMOL)
! (NMOL/L)
! (NMOL/H)
! (NMOL)
! (NMOL/L)
IPO ST SIMULATION UNIT CONVERSION
CRETOTAL= (ARE + AREB)/(WRE + WREB) ! TOTAL CONCENTRATION
(NMOL/L)
PRCT RE = 100.0*CRETOTAL/(MSTT+1E-30) ! PERCENT OF ORAL DOSE IN REST OF
BODY
PRCT REIV = 100.0*CRETOTAL/(IV_RlateR+1E-30) ! [ PERCENT OF IV DOSE IN REST
OF BODY
CRENGKG=CRETOTAL*MW ! REST OF THE BODY CONCENTRATION
(NG/KG)
!LIVER COMPARTMENT
!TISSUE BLOOD SUBCOMPARTMENT
RALIB = QLI * (CA-CLIB )-PALI * (CLIB -CFLLIR)+LIRMLUM ! (NMOL/HR)
ALIB = INTEG(RALIB ,0.0) !(NMOL)
CLIB = ALIB/WLIB ! (NMOL/L)
!TISSUE SUBCOMPARMTENT
RALI = PALI* (CLIB - CFLLIR)-REXCLI ! (NMOL/HR)
ALI = INTEG(RALI, 0.0) ! (NMOL)
CLI = ALI/WLI ! (NMOL/L)
!FREE TCDD CONCENTRATION IN LIVER
! MODIFICATION OCTOBER 8 2009
CFLLI= IMPLC(CLI-(CFLLIR*PLI+(LIBMAX*CFLLIR/(KDLI+CFLLIR)) &
+((CYP1A2_103*CFLLIR/(KDLI2+CFLLIR)*PAS_INDUC)))-CFLLI,CFLLIO)
CFLLIR=DIM(CFLLI, 0.0) ! FREE TCDD CONCENTRATION IN LIVER
!MODIFIED FROM:
!PARAMETER (LIVER IRMN = 1.0E-30)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-32 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! CFLLI= IMPLC(CLI-(CFLLIR*PLI+(LIBMAX*CFLLIR/(KDLI+CFLLIR &
! +LIVER_1RMN))+((CYP 1 A2_l 03 *CFLLIR/(KDLI2 + CFLLIR &
! +LIVER1RMN) *P AS_INDUC)))-CFLLI, CFLLIO)
! CFLLIR=DIM(CFLLI, 0.0)
! MODIFICATION OCTOBER 8 2009
CBNDLI= LIBMAX*CFLLIR/(KDLI+CFLLIR) IBOUND CONCENTRATION (NMOL/L)
IPO ST SIMULATION UNIT CONVERSION
CLITOTAL= (ALI + ALIB)/(WLI + WLIB) ! TOTAL CONCENTRATION (NMOL/L)
PRCT LI = 100.0*CLITOTAL/(MSTT+1E-30) ! PERCENT OF ORAL DOSE IN LIVER
PRCT LIIV = 100.0*CLITOTAL/(IV_RlateR+1E-30) ! PERCENT OF IV DOSE IN LIVER
Rec_occ= CFLLIR/(KDLI+CFLLIR)
CLINGKG=CLITOTAL*MW ! LIVER CONCENTRATION IN NG/KG
AUCLI_NGKGH=integ(CLINGKG,0.0)
CBNDLINGKG = CBNDLI*MW ! BOUND CONCENTRATION IN NG/KG
AUCBNDLINGKGH =INTEG(CBNDLINGKG,0.0)
! FRACTION INCREASE OF INDUCTION OF CYP1A2
fold_ind=CYP 1 A2_l OUT/CYP1 A2_l A2
V ARIATION OF AC =(CYP 1 A2_l OUT-CYP1 A2_l A2)/CYP 1 A2_l A2
! VARIABLE ELIMINATION BASED ON THE CYP1A2
! MODIFICATION OCTOBER 8 2009
KBILELIT = Kel v * V ARI ATION OF AC! ! DOSE-DEPENDENT EXCRETION RATE
CONSTANT
REXCLI = KB ILELIT * CFLLIR* WLI ! DOSE-DEPENDENT BILLIARY EXCRETION
RATE
EXCLI = INTEG(REXCLI, 0.0)
!KBILE_LI_T =((CYP1 A2_l OUT-CYP 1 A2_l A2)/C YP1 A2_l A2)*Kelv !
! CHEMICAL IN CYP450 (1A2) COMPARTMENT
CYP1A21KINP = CYP1A21KOUT* CYP 1 A2_ 1 OUTZ ! BASAL PRODCUTION RATE OF
CYP1A2 SET EQUAL TO BASAL DEGREDATION RATE
! MODIFICATION OCTOBER 8 2009
CYP1A210UT =INTEG(CYP 1 A2_l KINP * (1.0 + CYP1A21EMAX *(CBNDLI+1.0e-
30)**HILL &
/(CYP 1 A2_ 1 EC5 0 * *HILL + (CBNDLI+1.0e-30)**HILL)) &
- CYP1A2_1K0UT*CYP1A2_10UT, CYP1A210UTZ)
!MODIFIED FROM:
!PARAMETER (CYP1A21RMN = 1E-30)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-33 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
ICYP1A210UT =INTEG(CYP1 A2_ 1KINP * (1 + CYP1A21EMAX *(CBND&
!LI +C YP 1 A2_ 1RMN) * *HILL/(C YP 1 A2_ 1 EC 5 0 + (CBNDLI + CYP1A21&
!RMN)**HILL) +CYP1A21RMN) - CYP1A2_1K0UT*CYP1A2_1&
!OUT, CYP1A210UTZ)
! EQUATIONS INCORPORATING DELAY OF CYP1A2 PRODUCTION (NOT USED IN
SIMULATIONS)
CYP1A21R02 = (CYP1A210UT - CYP1A2102)/ CYP1A21TAU
CYP1A2102 =INTEG(CYP 1 A2_ 1R02, CYP1A21A1)
CYP1A21R03 = (CYP1A2102 - CYP1A2103)/ CYP1A21TAU
CYP1A2103 =INTEG(CYP 1 A2_ 1R03, CYP1A21A2)
!PLACENTA COMPARTMENT
! TISSUE BLOOD SUB COMPARTMENT
RAPLAB= QPLA*(CA - CPLAB)-PAPLA*(CPLAB -CFLPLAR) ! NMOL/HR)
APLAB = INTEG(RAPLAB ,0.0) ! (NMOL)
CPLAB = APLAB/(WPLAB+1E-3 0) ! (NMOL/ML)
!TISSUE SUBCOMPARTMENT
RAPLA = PAPLA*(CPLAB-CFLPLAR)-RAMPF + RAFPM ! (NMOL/HR)
APLA = INTEG(RAPLA, 0.0) ! (NMOL)
CPLA = APLA/(WPLA+le-30) ! (NMOL/ML)
! NEW EQUATION AUGUST 28 2009
PARAMETER (PARA ZERO = 1.0E-30)
CFLPLA= IMPLC(CPLA-(CFLPLAR*PPLA +(PL ABM AX* CFLPL AR/(KDPL A&
+CFLPLAR+PARA_ZERO)))-CFLPLA,CFLPLA0)
CFLPL AR=DIM(CFLPL A, 0.0)
IPO ST SIMULATION UNIT CONVERSION
CPLATOTAL = ((APL AB+APL A)/(WPL AB +WPL A))
PRCTPLA = (CPLATOTAL/(MSTT +1 E-3 0)) * 100
PRCTPLAIV = (CPLATOTAL/(IV_RlateR+ 1E-30))* 100
! FETUS COMPARTMENT
RAFETUS= RAMPF-RAFPM
AFETU S=INTEG(RAFETU S, 0.0)
CFETU S=AFETU S/(WTFE+1. 0e-3 0)
CFETOTAL= CFETUS
CFETUSv = CFETU S/PFETU S
IPO ST SIMULATION UNIT CONVERSION
CFETUSNGKG = CFETUS*MW ! (NG/KG)
PRCTFE = 100.0*CFETOTAL/(MSTT+1E-30)
PRCTFEIV = 100.0*CFETOTAL/(IV_RlateR+lE-30)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-34 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
! TRANSFER OF DIOXIN FROM PLACENTA TO FETUS
!FETAL EXPOSURE ONLY DURING EXPOSURE
IF (T.LT. TRANSTIME ON) THEN
SWITCHtrans = 0.0
ELSE
SWITCHtrans = 1
END IF
! TRANSFER OF DIOXIN FROM PLACENTA TO FETUS
! MODIFICATION 26 SEPTEMBER 2003
RAMPF = (CLPLA_FET*CPLA)* SWITCHtrans
AMPF=INTEG(RAMPF ,0.0)
! TRANSFER OF DIOXIN FROM FETUS TO PLACENTA
RAFPM = (CLPL AFET * CFETU S_v) * S WIT CHtrans!
AFPM = INTEG(RAFPM, 0.0)
!CHECK MASS BALANCE
BDOSE= IVDOSE +LYMLUM+LIMLUM
BMASSE = EXCLI+AURI+AFB+AF+AREB+ARE+ALIB+ALI+APLA+APLAB+AFETUS !
BDIFF = BDO SE-BM AS SE
!BODY BURDEN (NMOL)
BODY BURDEN = AFB+AF+AREB+ARE+ALIB+ALI+APLA+APLAB
!BODY BURDEN CONCENTRATION (NG/KG)
BBNGKG =( AFB+AF+AREB+ARE+ALIB+ALI+APL A+APL AB ) * MW/WT 0
! END SIMULATION COMMAND
TERMT (T.GE. TimeLimit, 'Time limit has been reached.')
END ! END OF THE DERIVATIVE SECTION
END ! END OF THE DYNAMIC SECTION
END ! END OF THE PROGRAM
C.2.2.2. Input File
output @clear
prepare @clear T year CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
CBNGKG
CINT = 1 % 168% 100 %INTEGRATION TIME
%EXPOSURE SCENARIO
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-35 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
EXPTIMEON = 0 % TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXPTIMEOFF = 401190 %TIME AT WHICH EXPOSURE ENDS (HOUR)
DAY CYCLE = 24 %NUMBER OF HOURS BETWEEN DOSES (HOUR)
BCK TIME ON = 401190 %TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCK TIME OFF = 401190 %TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
IVLACK =401190
IVPERIOD =401190
%GESTATION CONTROL
MATTING =393120 % BEGINNING OF MATING (HOUR) AT 45 YEARS OLD
TIMELIMIT = 399840 %SIMULATION TIME LIMIT (HOUR)
TRANSTIME ON = 394632 % TRANSFER FROM MOTHER TO FETUS AT 1512
HOURS GESTATION
%EXPOSURE DOSE
MSTOT = 9.97339283634997E-07 % NG OF TCDD PER KG OF BW
MSTOTBCKGR =0. %0.1 % ORAL BACKGROUND EXPOSURE DOSE (NG/KG)
DOSEIV = 0. %10
DOSEIVLATE = 0. %10
% TRANFER MOTHER TO FETUS CLEARANCE
CLPLA FET = 0.001 % MOTHER TO FETUS TRANFER CLEARANCE (L/HR)
C.2.3. Rat Standard Model
C.2.3.1. Model Code
PROGRAM: 'Three Compartment PBPK Model in Rat: Standard Model (Non-
Gestation)'
!Rat_Dioxin_3C June09_2clean_icf_afterKKfix_v3_ratnongest.csl
!RAT_NON_GEST_ICF_F083109.CSL
! RATNONGESTICFF100609.CSL
INITIAL ! INITIALIZATION OF PARAMETERS
! SIMULATION PARAMETERS
CONSTANT PARA ZERO = ld-30
CONSTANT EXP TIME ON = 0.0
(HOURS)
CONSTANT EXP TIME OFF = 900.0
(HOURS)
CONSTANT DAY CYCLE = 900.0
(HOURS)
! TIME AT WHICH EXPOSURE BEGINS
! TIME AT WHICH EXPOSURE ENDS
! NUMBER OF HOURS BETWEEN DOSES
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-36 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
CONSTANT BCK TIME ON = 0.0 ! TIME AT WHICH BACKGROUND
EXPOSURE BEGINS (HOURS)
CONSTANT BCK TIME OFF = 0.0 ! TIME AT WHICH BACKGROUND
EXPOSURE ENDS (HOURS)
CONSTANT MW=322 IMOLECULAR WEIGHT (NG/NMOL)
CONSTANT SERBLO = 0.55
CONSTANT UNITCORR = 1000
!EXPO SURE DOSES
CONSTANT MSTOTBCKGR = 0.0
(UG/KG)
CONSTANT MSTOT = 10
CONSTANT MSTOTsc = 0.0
CONSTANT DOSEIV = 0.0
!ORAL BACKGROUND EXPOSURE DOSE
!ORAL EXPOSURE DOSE (UG/KG)
! SUBCUTANEOUS EXPOSURE DOSE (UG/KG)
! INJECTED DOSE (UG/KG)
!ORAL DOSE
MSTOT NM = MSTOT/MW ! AMOUNT IN NMOL/G
MSTOT NMBCKGR = MSTOTBCKGR/MW ! AMOUNT IN NMOL/G
!INTRAVENOUS DOSE
DOSEIV NM = DOSEIV/MW ! AMOUNT IN NMOL/G
!INITIAL GUESS OF THE FREE CONCENTRATION IN THE LIGAND
(COMPARTMENT INDICATED BELOW)====
CONSTANT CFLLI0 = 0.0 !LIVER (NMOL/ML)
IBINDING CAPACITY (AhR) FOR NON LINEAR BINDING (COMPARTMENT
INDICATED BELOW) (NMOL/ML) ===
CONSTANT LIBMAX = 3.5e-4 ! LIVER (NMOL/ML), WANG ET AL. 1997
! PROTEIN AFFINITY CONSTANTS (1A2 OR AhR, COMPARTMENT INDICATED
BELOW) (NMOL/ML)===
CONSTANT KDLI = 1 Oe-4 ! LIVER (AhR) (NMOL/ML), WANG ET AL. 1997
CONSTANT KDLI2 = 4.0e-2 ILIVER (1A2) (NMOL/ML), EMOND ET AL.
2004
!EXCRETION AND ABSORPTION CONSTANT [RAT]
CONSTANT KST = 0.36 ! GASTRIC RATE CONSTANT (HR-1), WANG ET
AL. (1997)
CONSTANT KABS = 0.48 !INTESTINAL ABSORPTION CONSTANT (HR-
1), WANGET AL. 1997
! URINARY ELIMINATION CLEARANCE (ML/HR)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-37 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CONSTANT CLURI = 0.01 !URINARY CLEARANCE (ML/HR), EMOND ET
AL. 2004
!INTERSPECIES VARIABLE ELIMINATION
CONSTANT KELV = 0.15 ! INTERSPECIES VARIABLE ELIMINATION
CONSTANT (1/HOUR) (OPTIMIZED), EMOND ET AL. 2004
! CONSTANT TO DIVIDE THE ABSORPTION INTO LYMPHATIC AND PORTAL
FRACTIONS
CONSTANT A =0.7 ! LYMPHATIC FRACTION, WANG ET AL. 1997
!PARTITION COEFFICIENTS
CONSTANT PF = 100
CONSTANT PRE = 1.5
1997
CONSTANT PLI = 6.0
! ADIPOSE TISSUE/BLOOD, WANGET AL. 1997
! REST OF THE BODY/BLOOD, WANG ET AL.
! LIVER/BLOOD, WANGET AL. 1997
!PARAMETER FOR INDUCTION OF CYP 1A2 [MOUSE] ===
CONSTANT PAS INDUC =1.0 ! INCLUDE INDUCTION? (1 = YES, 0 = NO)
CONSTANT CYP 1A210UTZ = 1.6 ! DEGRADATION CONCENTRATION
CONSTANT OF 1A2 (NMOL/ML), WANGET AL. 1997
CONSTANT CYP1A21A1 = 1.6 ! BASAL CONCENTRATION OF 1A1
(NMOL/ML), WANGET AL. 1997
CONSTANT CYP1A21EC50 = 0.13 ! DISSOCIATION CONSTANT TCDD-
CYP1A2 (NMOL/ML) , WANG ET AL. 1997
CONSTANT CYP 1A21A2 = 1.6 ! BASAL CONCENTRATION OF 1A2
(NMOL/ML) Wang et al (1997)
CONSTANT CYP1A21KOUT =0.1 ! FIRST ORDER RATE OF DEGRADATION
(H-l), WANG ET AL. 1997
CONSTANT CYP1A21TAU = 0.25 ! HOLDING TIME (H), WANG ET AL. 1997
CONSTANT CYP1A21EMAX = 600 ! MAXIMUM INDUCTION OVER BASAL
EFFECT (UNTITLESS), WANGET AL. 1997
CONSTANT HILL = 0.6 !HILL CONSTANT; COOPERATIVELY LIGAND
BINDING EFFECT CONSTANT (UNITLESS)
!TISSUE BLOOD FLOW EXPRESSED AS A FRACTION OF CARDIAC OUTPUT
CONSTANT QFF = 0.069 ! ADIPOSE TISSUE BLOOD FLOW FRACTION
(UNITLESS), WANGETAL. 1997
CONSTANT QLIF = 0.183 ! LIVER (UNITLESS), WANG ET AL. 1997
! DIFFUSION AL PERMEABILITY FRACTION
CONSTANT PAFF = 0.0910 ! ADIPOSE (UNITLESS), WANGET AL. 1997
CONSTANT PAREF = 0.0298 ! REST OF THE BODY (UNITLESS), WANG ET
AL. 1997
CONSTANT PALIF =0.35 ! LIVER (UNITLESS), WANG ET AL. 1997
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-38 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
!FRACTION OF TISSUE VOLUME (UNITLESS)
CONSTANT WLIO = 0.0360 ! LIVER, WANG ET AL. 1997
CONSTANT WF0 = 0.069 ! BLOOD, WANG ET AL. 1997
!COMPARTMENT TISSUE BLOOD EXPRESSED AS A FRACTION OF THE TOTAL
COMPARTMENT VOLUME =========
CONSTANT WFB0 = 0.050 ! ADIPOSE TISSUE, WANG ET AL. 1997
CONSTANT WREB0 = 0.030 ! REST OF THE BODY, WANG ET AL. 1997
CONSTANT WLIB0 = 0.266 ! LIVER , WANG ET AL. 1997
! EXPO SURE SCENARIO FOR UNIQUE OR REPETITIVE WEEKLY OR MONTHLY
EXPOSURE
! NUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK = 0.0 ! DELAY BEFORE EXPOSURE ENDS
(WEEK)
CONSTANT WEEK PERIOD =168.0 ! NUMBER OF HOURS IN THE WEEK
(HOURS)
CONSTANT WEEK FINISH = 168.0 ! TIME EXPOSURE ENDS (HOURS)
INUMBER OF EXPOSURES PER MONTH
CONSTANT MONTH LACK = 0.0 ! DELAY BEFORE EXPOSURE BEGINS
(MONTH)
! SET FOR BACKGROUND EXPOSURE===========
! CONSTANT FOR BACKGROUND EXPOSURE===========
CONSTANT Day LACK BG = 0.0 ! DELAY BEFORE EXPOSURE BEGINS
(HOURS)
CONSTANT Day PERIOD BG = 24.0 ! LENGTH OF EXPOSURE (HOURS)
INUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK BG = 0.0 ! DELAY BEFORE BACKGROUND
EXPOSURE (WEEK)
CONSTANT WEEK PERIOD BG =168.0 INUMBER OF HOURS IN THE WEEK
(HOURS)
CONSTANT WEEK FINISH BG = 168.0 ! TIME EXPOSURE ENDS (HOURS)
! GROWTH CONSTANT FOR RAT
!CONSTANT FOR MOTHER BODY WEIGHT GROWTH ======
CONSTANT BW T0 = 250.0 ICHANGED FOR SIMULATION
! CONSTANT USED IN CARDIAC OUTPUT EQUATION
CONSTANT QCCAR =311.4 ! CONSTANT (ML/MIN/KG), WANG ET AL.
! COMPARTMENT LIPID EXPRESSED AS THE FRACTION OF TOTAL LIPID
CONSTANT F TOTLIP = 0.855 IADIPOSE TISSUE (UNITLESS)
CONSTANT B TOTLIP = 0.0033 !BLOOD (UNITLESS)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CONSTANT RE TOTLIP =0.019 !REST OF THE BODY (UNITLESS)
CONSTANT LITOTLIP = 0.06 ILIVER (UNITLESS)
END !END OF THE INITIAL SECTION
DYNAMIC IDYNAMIC SIMULATION SECTION
ALGORITHM IALG = 2 ! GEAR METHOD
CINTERVAL CINT = 0.1 ! COMMUNICATION INTERVAL
MAXTERVAL MAXT = 1.0e+10 ! MAXIMUM CALCULATION INTERVAL
MINTERVAL MINT = 1.0E-10 ! MINIMUM CALCULATION INTERVAL
VARIABLE T = 0.0
CONSTANT TIMELIMIT = 900.0 ! SIMULATION TIME LIMIT (HOURS)
CINTXY = CINT
PFUNC = CINT
!TIME CONVERSION
DAY=T/24.0
WEEK =T/168.0
MONTH =T/730.0
YEAR=T/8760.0
! TIME IN DAYS
! TIME IN WEEKS
! TIME IN MONTHS
! TIME IN YEARS
DERIVATIVE ! PORTION OF CODE THAT SOLVES DIFFERENTIAL EQUATIONS
! CHRONIC OR SUBCHRONIC EXPOSURE SCENARIO =======
INUMBER OF EXPOSURES PER DAY
DAY LACK = EXP TIME ON ! DELAY BEFORE EXPOSURE BEGINS
(HOURS)
DAY PERIOD = DAY CYCLE ! EXPOSURE PERIOD (HOURS)
DAY FINISH = CINTXY ! LENGTH OF EXPOSURE (HOURS)
MONTH PERIOD = TIMELIMIT ! EXPOSURE PERIOD (MONTHS)
MONTH FINISH = EXP TIME OFF ! LENGTH OF EXPOSURE (MONTHS)
INUMBER OF EXPOSURES PER DAY AND MONTH
DAY FINISH BG = CINTXY ! LENGTH OF EXPOSURE (HOURS)
MONTH LACK BG = BCK TIME ON ! DELAY BEFORE BACKGROUND
EXPOSURE BEGINS (MONTHS)
MONTH PERIOD BG = TIMELIMIT ! BACKGROUND EXPOSURE PERIOD
(MONTHS)
MONTH FINISH BG = BCK TIME OFF ! LENGTH OF BACKGROUND
EXPOSURE (MONTHS)
B = 1-A ! FRACTION OF DIOXIN ABSORBED IN THE PORTAL
FRACTION OF THE LIVER
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-40 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! BODY WEIGHT GROWTH EQUATION========
PARAMETER (BW RMN = 1.0E-30)
WT0= (BW TO *(1,0+(0.41 *T)/(1402.5+T+BW_RMN)))
! VARIABILITY OF REST OF THE BODY DEPEND OTHERS ORGAN
WREO = (0.91 - (WLIB0*WLI0 + WFB0*WF0 + WLI0 + WF0))/(1.0+WREB0) !REST OF
THE BODY FRACTION; UPDATED FOR EPA ASSESSMENT
QREF = 1 0-(QFF+QLIF) !REST OF BODY BLOOD FLOW
QTTQF = QFF+QREF+QLIF ! SUM MUST EQUAL 1
! COMPARTMENT VOLUME (G) =========
WF = WF0 * WTO ! ADIPOSE
WRE = WREO * WTO ! REST OF THE BODY
WLI = WLI0 * WTO ! LIVER
!COMPARTMENT TISSUE BLOOD VOLUME (G) =========
WFB = WFB0 * WF ! ADIPOSE
WREB = WREB0 * WRE ! REST OF THE BODY
WLIB = WLIB0 * WLI ! LIVER
! CARDIAC OUTPUT FOR THE GIVEN BODY WEIGHT
QC= QCCAR*60.0*(WT0/UNITCORR)**0.75
! COMPARTMENT BLOOD FLOW (ML/HR)
QF = QFF*QC ! ADIPOSE TISSUE BLOOD FLOW RATE
QLI = QLIF*QC ! LIVER TISSUE BLOOD FLOW RATE
QRE = QREF*QC ! REST OF THE BODY BLOOD FLOW RATE
QTTQ = QF+QRE+QLI ! TOTAL FLOW RATE
! PERMEABILITY ORGAN FLOW (ML/HR)
PAF =PAFF*QF ! ADIPOSE
PARE = PAREF*QRE ! REST OF THE BODY
PALI = PALIF*QLI ! LIVER TISSUE
! CONDITIONAL ORAL EXPOSURE (BACKGROUND EXPOSURE)
! EXPO SURE + ! REPETITIVE EXPOSURE SCENARIO
IV= DOSEIV NM * WTO ! AMOUNT IN NMOL
MSTT= MSTOT NM * WTO ! AMOUNT IN NMOL
MSTTBCKGR =MSTOT_NMBCKGR *WT0
!REPETITIVE ORAL BACKGROUND EXPOSURE SCENARIOS
DAYEXPOSUREBG =
PUL SE(D A Y_L ACKB G,D A YPERIODB G,D A YFINI SH_B G)
WEEKEXPOSUREBG =
PULSE(WEEK_LACK_BG,WEEK_PERIOD_BG,WEEK_FINISH_BG)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-41 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
MONTHEXPOSUREBG =
PUL SE(MONTH_L ACKB G,MONTH_PERIOD_B G,MONTH_FINI SH_B G)
MSTTCHBG =
(DAY_EXPOSURE_BG*WEEK_EXPOSURE_BG*MONTH_EXPOSURE_BG)*MSTTBCK
GR
MSTTFRBG = MSTTBCKGR/CINT
C Y CLEBG =D AYEXPO SUREBG* WEEKEXPO SURE_BG*MONTH_EXPO SUREB G
IF (MSTTCH BG.EQ.MSTTBCKGR) THEN
ABSMSTT_GB= MSTTFRBG
ELSE
ABSMSTTGB = 0.0
END IF
!REPETITIVE ORAL MAIN EXPOSURE SCENARIO
DAYEXPOSURE = PULSE(DAY_LACK,DAY_PERIOD,DAY_FINISH)
WEEKEXPOSURE = PULSE(WEEK_LACK,WEEK_PERIOD,WEEK_FINISH)
MONTHEXPOSURE = PULSE(MONTH_LACK,MONTH_PERIOD,MONTH_FINISH)
MSTTCH = (DAY_EXPOSURE*WEEK_EXPOSURE*MONTH_EXPOSURE)*MSTT
CYCLE = DAYEXPO SURE* WEEKEXPO SURE*MONTH_EXPO SURE
MSTTFR = MSTT/CINT
SUMEXPEVENT= integ (CYCLE,0.0) INUMBER OF CYCLE GENERATE DURING
SIMULATION
! CONDITIONAL ORAL EXPOSURE
IF (MSTTCH.EQ.MSTT) THEN
ABSMSTT= MSTTFR
ELSE
ABSMSTT = 0.0
END IF
C Y CLETOT=INTEG(C Y CLE, 0.0)
!MASS CHANGE IN THE LUMEN
RMSTT = -(KST+KABS)*MST+ABSMSTT +ABSMSTT GB ! RATE OF CHANGE
(NMOL/H)
MST = INTEG(RMSTT,0.0) ! AMOUNT OF STAY IN DUODENUM (NMOL)
! ABSORPTION IN LYMPH CIRCULATION
LYRMLUM = KABS*MST*A
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-42 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
LYMLUM = INTEG(L YRMLUM, 0.0)
! ABSORPTION IN PORTAL CIRCULATION
LIRMLUM = KABS*MST*B
LIMLUM = INTEG(LIRMLUM, 0.0)
IPERCENT OF DOSE REMAINING IN THE GI TRACT
PRCTremainGIT = (MST/(MSTT+PARA_ZERO))* 100.0
! ABSORPTION of Dioxin by IV route
IVR= IV/PFUNC ! RATE FOR IV INFUSION IN BLOOD
EXPIV= IVR * (1 O-STEP(PFUNC))
IVDOSE = integ(EXPIV,0.0)
! SYSTEMIC BLOOD COMPARTMENT
! MODIFICATION ON OCTOBER 6, 2009
CB=(QF * CFB+QRE * CREB+QLI * CLIB+EXPIV+LYRMLUM)/(QC+CLURI) !
CA = CB
! URINARY EXCRETION BY KIDNEY
! MODIFICATION ON OCTOBER 6, 2009
RAURI = CLURI *CB
AURI = INTEG(R AURI, 0.0)
! CONVERSION EQUATION POST SIMULATION
PRCTB = (CB/(MSTT+PARA_ZERO))* 100.0
CBNGKG = CB*MW*UNITCORR ! [NG/KG]
CBSNGKGLIADJ= (CB*MW*UNITCORR*(1.0/B_TOTLIP)*(1.0/SERBLO))![NG of TCDD
Serum/Kg OF LIPIP]
!ADIPOSE TISSUE COMPARTMENT
! TISSUE BLOOD SUB COMPARTMENT
RAFB = QF * (C A-CFB)-P AF * (CFB -CF/PF) !(NMOL/HR)
AFB = INTEG(RAFB ,0.0) !(NMOL)
CFB = AFB/WFB ! (NMOL/ML)
!TISSUE SUBCOMPARTMENT
RAF = PAF*(CFB-CF/PF) !(NMOL/HR)
AF = INTEG(RAF,0.0) !(NMOL)
CF = AF/WF ! (NMOL/ML)
! CONVERSION EQUATION POST SIMULATION
CFTOTAL = (AF + AFB)/(WF + WFB) ! TOTAL CONCENTRATION IN NMOL/ML
PRCT F = (CFTOTAL/(MSTT+PARA_ZERO))* 100.0 ! PRCENT OF DOSE IN FAT
CFNGKG = CFTOTAL*MW*UNITCORR ! CONCENTRATION [NG/KG]
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-43 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
!REST OF THE BODY COMPARTMENT
! TISSUE BLOOD SUBCOMPARTMENT
RAREB= QRE*(CA-CREB)-PARE*(CREB-CRE/PRE) ! (NMOL/HR)
AREB = INTEG(RAREB,0.0) !(NMOL)
CREB = AREB/WREB ! (NMOL/ML)
! TISSUE COMPARTMENT
RARE = PARE*(CREB - CRE/PRE) ! (NMOL/HR)
ARE = INTEG(RARE,0.0) !(NMOL)
CRE = A RE/W RE ! (NMOL/ML)
!CONVERSION EQUATION POST SIMULATION
CRETOTAL= (ARE + AREB)/(WRE + WREB) ! TOTAL CONCENTRATION IN
NMOL/ML
PRCTRE = (CRETOTAL/(MSTT+PARA_ZERO))* 100.0
CTREPGG= CRETOTAL*MW*UNITCORR !(PG/ML)
AUCREPPG = integ(C TREPGG, 0.0)
! LIVER COMPARTMENT
!TISSUE BLOOD COMPARTMENT
RALIB = QLI*(CA-CLIB)-PALI*(CLIB-CFLLIR)+LIRMLUM !(NMOL/HR)
ALIB = INT eg(RALIB ,0.0) !(NMOL)
CLIB = ALIB/WLIB
!TISSUE COMPARTMENT
RALI = PALI * (CLIB -CFLLIR)-REXCLI ! (NMOL/HR)
ALI = integ(RALI,0.0) ! (NMOL)
CLI = ALI/WLI ! (NMOL/ML)
PARAMETER (LIVERIRMN = 1.0E-30)
CFLLI= IMPLC(CLI-(CFLLIR*PLI+(LIBMAX*CFLLIR/(KDLI+CFLLIR &
+LIVER1 RMN))+((C YP1 A2_ 103 * CFLLIR/(KDLI2+CFLLIR &
+LIVER_1RMN)*PAS_INDUC)))-CFLLIR,CFLLI0) ! FREE TCDD CONCENTRATION IN
LIVER
CFLLIR=DIM(CFLLI, 0.0)
CBNDLI= LIBM AX* CFLLIR/(KDLI+CFLLIR+LIVER_1 RMN) IBOUND
CONCENTRATION
! CONVERSION EQUATION POST SIMULATION
CLITOTAL= (ALI + ALIB)/(WLI + WLIB) ! TOTAL CONCENTRATION IN
NMOL/ML
PRCTLI = (CLITOTAL/(MSTT+PARA_ZERO))* 100.0
rec_occ_AHR= (CFLLIR/(KDLI+CFLLIR+1))* 100.0 ! PERCENT OF AhR
OCCUPANCY
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-44 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
PROT_occ_lA2= (CFLLIR/(KDLI2+CFLLIR))* 100.0 ! PERCENT OF 1A2
OCCUPANCY
CLINGKG =(CLITOTAL*MW*UNITCORR)
CBNDLINGKG = CBNDLI*MW*UNITCORR
AU CLIN GKGH=INTEG(CLIN GKG, 0.0)
CLINGG=CLITOTAL*MW
! VARIABLE ELIMINATION HALF-LIFE BASED ON THE CONCENTRATION OF
CYP1A2
KBILE LI T =((CYP 1 A2_l OUT-CYP1 A2_l A2)/CYP 1 A2_l A2)*Kelv ! INDUCED
BILIARY EXCRETION RATE CONSTANT
REXCLI= (KBILE_LI_T*CFLLIR*WLI) ! DOSE-DEPENDENT BILIARY EXCRETION
RATE
EXCLI = INTEG(REXCLI, 0.0)
! CHEMICAL IN CYP450 (1A2) COMPARTMENT
! ===PARAMETER FOR INDUCTION OF CYP1A2
CYP1A21KINP = CYP1A21KOUT* CYP1A210UTZ ! BASAL RATE OF CYP1A2
PRODUCTION SET EQUAL TO BASAL RATE OF DEGREDATION
! MODIFICATION ON OCTOBER 6, 2009
CYP1A210UT =INTEG(CYP1 A2_ 1KINP * (1.0 + CYP1A21EMAX *(CBNDLI+1.0e-
30)**HILL &
/(CYP 1 A2_ 1 EC50* *HILL + (CBNDLI+1.0e-30)**HILL)) &-
- CYP 1 A2_ 1K OUT * C YP 1 A2_ 1 OUT, CYP1A210UTZ)
! EQUATIONS INCORPORATING DELAY OF CYP1A2 PRODUCTION (NOT USED IN
SIMULATIONS)
CYP1A21R02 = (CYP1A210UT - CYP1A2102)/ CYP1A21TAU
CYP1A2102 =INTEG(CYP 1 A2_ 1R02, CYP1A21A1)
CYP1A21R03 = (CYP1A2102 - CYP1A2103)/ CYP1A21TAU
CYP1A2103 =INTEG(CYP 1 A2_ 1R03, CYP1A21A2)
I CHECK MASS BALANCE
BDOSE= LYMLUM+LIMLUM+IVDOSE
BMASSE = EXCLI+AURI+AFB+AF+AREB+ARE+ALIB+ALI
BDIFF = BDOSE-B MASSE
I BODY BURDEN
BBNGKG =(((AFB+AF+AREB+ARE+ALIB+ALI) *MW)/(WT0/UNITC ORR)) !
I END OF THE SIMULATION COMMAND
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-45 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
TERMT (T.GE. TimeLimit, 'Time limit has been reached.')
END ! END OF THE DERIVATIVE SECTION
END ! END OF THE DYNAMIC SIMULATION SECTION
END ! END OF THE PROGRAM.
C.2.3.2. Input Files
C.2.3.2.1. Cantoni et al. (1981).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
%Cantoni etal. 1981
%protocol: oral exposure 1 dose/week for 45 weeks; female CD-COBS rats
%dose levels: 0.01, 0.1, 1 ug/kg 1 dose/week for 45 weeks
%dose levels: 10, 100, 1000 ng/kg 1 dose/week for 45 weeks
%dose levels equivalent to: 1.43, 14.3 143 ng/kg 7 days/week for 45 weeks
MAXT =0.01
CINT =0.1
EXP TIME ON = 0. %TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXP TIME OFF = 7560 %TIME AT WHICH EXPOSURE ENDS (HOUR)
DAY CYCLE = 168
BCK TIME ON = 0. %TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCK TIME OFF =0. %TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
TIMELIMIT = 7584 %SIMULATION TIME LIMIT (HOUR)
BW T0 = 125 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.01 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.1 % EXPOSURE DOSE IN UG/KG
MSTOT =1 % EXPOSURE DOSE IN UG/KG
C.2.3.2.2. Chu et al. (2007).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Chu et al. 2007
%protocol: oral exposure daily for 28 days
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-46 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
%dose levels: 0.0025, 0.025, 0.250, 1.0 ug/kg every day for 28 days
% dose levels = 2.5, 25, 250, 1000 ng/kg every day for 28 days
MAXT = 0.01
CINT = 0.1
EXPTIMEON = 0.
experiment (age =12 weeks)
EXPTIMEOFF = 672.
weeks
DAY CYCLE = 24.
BCKTIMEON = 0.
BCKTIMEOFF = 0.
TIMELIMIT = 672.
BW TO = 200.
%delay before begin exposure (HOUR) 5 weeks after start of
%TIME EXPOSURE STOP (HOUR); 30 doses, 1 every two
% once every two weeks
%DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
%TIME OF BACKGROUND EXPOSURE STOP (HOUR)
% SIMULATION LIMIT TIME (HOUR)
% Body weight at the beginning of the simulation (g);
corresponds to 12 week old female
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.0025 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT =0.025 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT = 0.250 % ORAL EXPOSURE DOSE (UG/KG)
MSTOT = 1.0 % ORAL EXPOSURE DOSE (UG/KG)
C.2.3.2.3. Crofton et al (2005).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Crofton et al. 2005
%protocol: oral exposure daily for 4 days
%dose levels: 0.0001, 0.003, 0.01, 0.03, 0.1, 0.3, 1, 3, and 10 ug/kg every day for four days
%dose levels: 0.1, 3, 10, 30, 100, 300, 1000, 3000, and 10000 ng/kg every day for four days
MAXT = 0.01
CINT = 0.1
EXPTIMEON = 0.
experiment (age =12 weeks)
EXPTIMEOFF = 96.
weeks
DAYCYCLE = 24.
BCKTIMEON = 0.
BCKTIMEOFF = 0.
TIMELIMIT = 96.
BWT0 = 250
to 12 week old female
%delay before begin exposure (HOUR) 5 weeks after start of
%TIME EXPOSURE STOP (HOUR); 30 doses, 1 every two
% once every two weeks
%DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
%TIME OF BACKGROUND EXPOSURE STOP (HOUR)
% SIMULATION LIMIT TIME (HOUR)
% Body weight at the beginning of the simulation (g); corresponds
%EXPOSURE DOSE SCENARIOS (UG/KG)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-47 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
MSTOT
= 0.0001
%MSTOT
= 0.003
%MSTOT
= 0.01
%MSTOT
= 0.03
%MSTOT
= 0.1
%MSTOT
= 0.3
%MSTOT
= 1.
%MSTOT
= 3.
MSTOT
= 10.
% ORAL EXPOSURE DOSE (UG/KG)
% ORAL EXPOSURE DOSE (UG/KG)
% ORAL EXPOSURE DOSE (UG/KG)
% ORAL EXPOSURE DOSE (UG/KG)
% ORAL EXPOSURE DOSE (UG/KG)
% ORAL EXPOSURE DOSE (UG/KG)
% ORAL EXPOSURE DOSE (UG/KG)
% ORAL EXPOSURE DOSE (UG/KG)
% ORAL EXPOSURE DOSE (UG/KG)
C.2.3.2.4. Fattore et al. (2000).
output @ clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Fattore et al. 2000
%built and check in August 7 2009
%protocol: oral exposure in diet for 13 weeks; SD rats
%dose levels: 0.02, 0.1, 0.2, 2 ug/kg 7 days/week for 13 weeks
%dose levels equivalent to: 20, 100, 200, 2000 ng/kg 7 days/week for 13 weeks
MAXT = 0.01
CINT =0.1
EXPTIMEON = 0.
EXPTIMEOFF =2184
DAY CYCLE = 24
B CKTIMEON = 0.
(HOUR)
B CKTIMEOFF = 0.
(HOUR)
TIMELIMIT = 2184 %SIMULATION TIME LIMIT (HOUR)
BW T0 =150 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
%TIME AT WHICH EXPOSURE BEGINS (HOUR)
%TIME AT WHICH EXPOSURE ENDS (HOUR)
%TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.02 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.1 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.2 % EXPOSURE DOSE IN UG/KG
MSTOT = 2 % EXPOSURE DOSE IN UG/KG
C.2.3.2.5. Hassoun et al (2000).
output @ clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
% Hassoun et al. 2000
%protocol: oral exposure for 13 weeks; SD rats
%dose levels: 0.003, 0.010, 0.022, 0.046 0.1 ug/kg 5 days/weeks for 13 weeks
%dose levels equivalent to: 3, 10, 22, 46 100 ng/kg 5 days/weeks for 13 weeks
%dose levels equivalent to: 2.14, 7.14, 15.7, 32.9, 71.4 ng/kg 7 days/weeks for 13 weeks
MAXT
CI NT
= 0.01
= 0.1
EXPTIMEON
EXPTIMEOFF
DAY CYCLE
WEEKPERIOD
WEEKFINISH
BCKTIMEON
BCKTIMEOFF
TIMELIMIT
BW TO
= 0.
= 2184.
= 24.
= 168.
= 119.
= 0.
= 0.
= 2184.
= 215. %
%delay before begin exposure (HOUR)
%TIME EXPOSURE STOP (HOUR)
%DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
%TIME OF BACKGROUND EXPOSURE STOP (HOUR)
% SIMULATION LIMIT TIME (HOUR)
Body weight at the beginning of the simulation (g)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.003 % exposure dose ug/kg
%MSTOT = 0.010 % exposure dose ug/kg
%MSTOT = 0.022 % exposure dose ug/kg
%MSTOT = 0.046 % exposure dose ug/kg
MSTOT =0.1 % exposure dose ug/kg
C.2.3.2.6. Kitchin and Woods (1979).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Kitchin and Woods 1979
%dose levels: 0.0006, 0.002, 0.004, 0.020, 0.060, 0.200, 0.600, 2.000, 5.000, 20.000 ug/kg
single oral gavage
% dose levels = 0.6, 2, 4, 20, 60, 200, 600, 2000, 5000, 20000 ng/kg single oral gavage with
estimated 0.2 ng/kg/day background dose
MAXT = 0.01
CINT = 0.1
EXP TIME ON = 0. %delay before begin exposure (HOUR)
EXP TIME OFF =23. %TIME EXPOSURE STOP (HOUR)
DAY_CYCLE = 24. % once every two weeks
BCK TIME ON = 0. %DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
BCK TIME OFF = 72. %TIME OF BACKGROUND EXPOSURE STOP (HOUR)
TIMELIMIT = 72. % SIMULATION LIMIT TIME (HOUR)
BW_T0 = 225. % Body weight at the beginning of the simulation (g)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-49 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
%EXPOSURE
DOSE SCENARIOS (UG/KG)
%MSTOT
= 0.0006
% ORAL EXPOSURE DOSE (UG/KG)
%MSTOT
= 0.002
% ORAL EXPOSURE DOSE (UG/KG)
%MSTOT
= 0.004
% ORAL EXPOSURE DOSE (UG/KG)
%MSTOT
= 0.020
% ORAL EXPOSURE DOSE (UG/KG)
%MSTOT
= 0.060
% ORAL EXPOSURE DOSE (UG/KG)
%MSTOT
= 0.200
% ORAL EXPOSURE DOSE (UG/KG)
%MSTOT
= 0.600
% ORAL EXPOSURE DOSE (UG/KG)
%MSTOT
= 2.000
% ORAL EXPOSURE DOSE (UG/KG)
%MSTOT
= 5.000
% ORAL EXPOSURE DOSE (UG/KG)
MSTOT
= 20.000
% ORAL EXPOSURE DOSE (UG/KG)
C.2.3.2.7. Kocibaetal. (1976) (13 weeks).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIADJ BBNGKG CBNDLINGKG
% Kociba et al, 1976.
%built and check in August 7 2009
%protocol: 5 days/week exposure for 13 weeks; SD rats
%dose levels: 0.001, 0.01, 0.1, 1 ug/kg 5 days/week for 13 weeks
%dose levels: 1, 10, 100, 1000 ng/kg 5 days/week for 13 weeks
%dose levels equivalent to: 0.714, 7.14, 71.4, 714 ng/kg/d (adj) 7 days/week for 13 weeks
MAXT =0.01
CINT =0.1
EXPTIMEON = 0.
EXPTIMEOFF =2184
WEEKPERIOD = 168
WEEKFINISH =119
DAY CYCLE = 24
BCKTIMEON = 0.
(HOUR)
BCKTIMEOFF = 0.
(HOUR)
TIMELIMIT = 4368
BWT0 = 180
(G)
%TIME AT WHICH EXPOSURE BEGINS (HOUR)
%TIME AT WHICH EXPOSURE ENDS (HOUR)
% TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
% SIMULATION TIME LIMIT (HOUR)
% BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.001 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.01 % EXPOSURE DOSE IN UG/KG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-50 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
%MSTOT = 0.1 % EXPOSURE DOSE IN UG/KG
MSTOT =1 % EXPOSURE DOSE IN UG/KG
C.2.3.2.8. Kociba et al. (1978) (female) (104 weeks).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIADJ BBNGKG CBNDLINGKG
% Kociba et al, 1978.
%built and check in August 7 2009
%protocol: daily dietary exposure for 104 weeks; SD rats
%dose levels: 0.001, 0.01, 0.1 ug/kg 7 days/week for 104 weeks
%dose levels: 1, 10, 100 ng/kg 7 days/week for 104 weeks
MAXT
CI NT
= 0.01
= 0.1
EXPTIMEON = 0.
EXPTIMEOFF = 17472
DAY CYCLE = 24
B CKTIMEON = 0.
(HOUR)
B CKTIMEOFF = 0.
(HOUR)
TIMELIMIT = 17472
BWT0 = 180
SIMULATION (G)
%TIME AT WHICH EXPOSURE BEGINS (HOUR)
%TIME AT WHICH EXPOSURE ENDS (HOUR)
%TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
%SIMULATION TIME LIMIT (HOUR)
% BODY WEIGHT AT THE BEGINNING OF THE
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.001 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.01 % EXPOSURE DOSE IN UG/KG
MSTOT = 0.1 % EXPOSURE DOSE IN UG/KG
C.2.3.2.9. Kociba et al (1978) (male) (104 weeks).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIAD J BBNGKG CBNDLINGKG
% Kociba et al, 1978.
%built and check in August 7 2009
%protocol: daily dietary exposure for 104 weeks; SD rats
%dose levels: 0.001, 0.01, 0.1 ug/kg 7 days/week for 104 weeks
%dose levels: 1, 10, 100 ng/kg 7 days/week for 104 weeks
MAXT = 0.01
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-51 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
CINT = 0.1
EXPTIMEON = 0.
EXPTIMEOFF = 17472
DAY CYCLE = 24
BCKTIMEON = 0.
(HOUR)
BCKTIMEOFF = 0.
(HOUR)
TIMELIMIT = 17472
BWT0 = 250
SIMULATION (G)
%TIME AT WHICH EXPOSURE BEGINS (HOUR)
%TIME AT WHICH EXPOSURE ENDS (HOUR)
%TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
%SIMULATION TIME LIMIT (HOUR)
% BODY WEIGHT AT THE BEGINNING OF THE
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.001 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.01 % EXPOSURE DOSE IN UG/KG
MSTOT = 0.1 % EXPOSURE DOSE IN UG/KG
C.2.3.2.10. Latchoumycandane and Mathur. (2002).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Latchoumycandane and Mathur, 2002.
%built and check in August 7 2009
%protocol: 1 time per day for 45 days oral gavage
%dose levels: 0.001, 0.01, 0.1 ug/kg daily for 45 days
%dose levels: 1, 10, 100 ng/kg daily for 45 days
MAXT
CINT
= 0.01
= 0.1
EXPTIMEON = 0.
EXPTIMEOFF = 1080
DAYCYCLE = 24
BCKTIMEON = 0.
(HOUR)
BCKTIMEOFF = 0.
(HOUR)
TIMELIMIT =1104
BWT0 = 200
(G)
% TIME AT WHICH EXPOSURE BEGINS (HOUR)
% TIME AT WHICH EXPOSURE ENDS(HOUR)
% TIME AT WHICH BACKGROUND EXPOSURE BEGINS
% TIME AT WHICH BACKGROUND EXPOSURE ENDS
% SIMULATION TIME LIMIT (HOUR)
% BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.001 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.01 % EXPOSURE DOSE IN UG/KG
MSTOT =0.1 % EXPOSURE DOSE IN UG/KG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-52 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
C.2.3.2.11. Lietal. (1997).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIADJ BBNGKG CBNDLINGKG
% Li et al 1997
% created 1/10/10
% Non-gestational rat model
% dose levels: 3, 10, 30, 100, 300, 1000, 3000, 10000, 30000 nkd one dose via gavage, sacrificed
24 hrs later
MAXT = 0.1
CINT = 0.1
EXPTIMEON = 0.
EXPTIMEOFF = 24.
DAY CYCLE = 24.
BCKTIMEON = 0.
BCKTIMEOFF = 0.
TIMELIMIT = 24.
BW TO = 56.5
%delay before begin exposure (HOUR)
%TIME EXPOSURE STOP (HOUR)
%DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
%TIME OF BACKGROUND EXPOSURE STOP (HOUR)
%SIMULATION LIMIT TIME (HOUR)
% Body weight at the beginning of the simulation (g)
%EXPOSURE
MSTOT
%MSTOT
%MSTOT
%MSTOT
%MSTOT
%MSTOT
%MSTOT
%MSTOT
%MSTOT
DOSE SCENARIOS (UG/KG)
= 0.003 % ORAL EXPOSURE DOSE (UG/KG)
= 0.01 % ORAL EXPOSURE DOSE (UG/KG)
= 0.03 % ORAL EXPOSURE DOSE (UG/KG)
= 0.1 % ORAL EXPOSURE DOSE (UG/KG)
= 0.3 % ORAL EXPOSURE DOSE (UG/KG)
= 1. % ORAL EXPOSURE DOSE (UG/KG)
= 3. % ORAL EXPOSURE DOSE (UG/KG)
= 10. % ORAL EXPOSURE DOSE (UG/KG)
= 30. % ORAL EXPOSURE DOSE (UG/KG)
C.2.3.2.12. Murray et al. (1979).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIAD J BBNGKG CBNDLINGKG
% Murray et al 1979
%built and check in August 7 2009
%protocol: dietary exposure for 3 generations (assume 120 day exposure for each)
%dose levels: 0.001 0.01, 0.1 ug/kg/d
%dose levels: 1, 10, 100 ng/kg/d
MAXT = 0.01
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-53 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
CINT = 0.1
EXPTIMEON = 0. %TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXPTIMEOFF = 2880 %TIME AT WHICH EXPOSURE ENDS (HOUR);
CORRESPONDS TO 120 DAYS OF EXPOSURE
DAY CYCLE = 24.
BCKTIMEON = 0. %TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCKTIMEOFF = 0. %TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
TIMELIMIT = 2880 %SIMULATION TIME LIMIT (HOUR)
BW T0 = 4.5 % BODY WEIGHT AT THE BEGINNING OF THE
SIMULATION (G)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.001 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT =0.01 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT = 0.1 % ORAL EXPOSURE DOSE IN UG/KG
C.2.3.2.13. NTP (1982) (female) (chronic).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% NTP 1982
%built and check in August 7 2009
%protocol: twice weekly gavage for 104 weeks + 3 week observation period
%dose levels: 0.005, 0.025, 0.25 ug/kg biweekly for 104 weeks + 3 week observation period
%dose levels: 5, 25, 250 ng/kg biweekly for 104 weeks + 3 week observation period
%dose levels equivalent to: 1.43, 7.14, 71.4 ng/kg/d (adj)
MAXT
CINT
= 0.01
= 0.1
EXPTIMEON = 0.
EXPTIMEOFF = 17472
DAYCYCLE = 84
B CKTIMEON = 0.
(HOUR)
B CKTIMEOFF = 0.
(HOUR)
TIMELIMIT = 17976
BWT0 = 250
SIMULATION (G)
%TIME AT WHICH EXPOSURE BEGINS (HOUR)
%TIME AT WHICH EXPOSURE ENDS (HOUR)
%TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
% SIMULATION TIME LIMIT (HOUR)
Zo BODY WEIGHT AT THE BEGINNING OF THE
%EXPOSURE DOSE SCENARIOS (UG/KG)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-54 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
%MSTOT = 0.005 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.025 % EXPOSURE DOSE IN UG/KG
MSTOT = 0.25 % EXPOSURE DOSE IN UG/KG
C.2.3.2.14. NTP (1982) (male) (chronic).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIADJ BBNGKG CBNDLINGKG
% NTP 1982
%built and check in August 7 2009
%protocol: twice weekly gavage for 104 weeks + 3 week observation period
%dose levels: 0.005, 0.025, 0.25 ug/kg biweekly for 104 weeks + 3 week observation period
%dose levels: 5, 25, 250 ng/kg biweekly for 104 weeks + 3 week observation period
%dose levels equivalent to: 1.43, 7.14, 71.4 ng/kg/d (adj)
MAXT
CI NT
= 0.01
= 0.1
EXPTIMEON = 0.
EXPTIMEOFF = 17472
DAY CYCLE = 84
B CKTIMEON = 0.
(HOUR)
B CKTIMEOFF = 0.
(HOUR)
TIMELIMIT = 17976
BWT0 =350
SIMULATION (G)
%TIME AT WHICH EXPOSURE BEGINS (HOUR)
%TIME AT WHICH EXPOSURE ENDS (HOUR)
%TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
% SIMULATION TIME LIMIT (HOUR)
% BODY WEIGHT AT THE BEGINNING OF THE
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.005 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.025 % EXPOSURE DOSE IN UG/KG
MSTOT = 0.25 % EXPOSURE DOSE IN UG/KG
C.2.3.2.15. NTP (2006) 31 weeks.
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIAD J BBNGKG CBNDLINGKG
% NTP 2006
%built and check in August 7 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-55 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
%protocol: oral exposure for 31 weeks; SD rats
%Rat_Dioxin_3C June09_2clean.csl
%RAT_NON_GEST_ICF_F083109.CSL (now 09-11-09)
%dose levels: 0.003, 0.010, 0.022, 0.046, 0.1 ug/kg 5 days/week for 31 weeks
%dose levels equivalent to: 3, 10, 22, 46, 100 ng/kg 5 days/week for 31 weeks
%dose levels equivalent to: 2.14, 7.14, 15.7, 32.9, 71.4 ng/kg 7 days/week for 31 weeks
MAXT
CI NT
= 0.01
= 0.1
EXPTIMEON
EXPTIMEOFF
DAY CYCLE
WEEKPERIOD
WEEKFINISH
BCKTIMEON
BCKTIMEOFF
TIMELIMIT
BW TO
= 0. %delay before begin exposure (HOUR)
= 17640 %TIME EXPOSURE STOP (HOUR)
= 24
= 168
= 119
= 0.
= 0.
= 5208
= 215 °A
%DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
%TIME OF BACKGROUND EXPOSURE STOP (HOUR)
%SIMULATION LIMIT TIME (HOUR)
Body weight at the beginning of the simulation (g)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.003 % exposure dose ug/kg
%MSTOT = 0.010 % exposure dose ug/kg
%MSTOT = 0.022 % exposure dose ug/kg
%MSTOT = 0.046 % exposure dose ug/kg
MSTOT =0.1 % exposure dose ug/kg
C.2.3.2.16. NTP (2006) 53 weeks.
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% NTP 2006
%built and check in August 7 2009
%protocol: oral exposure for 53 weeks; SD rats
%Rat_Dioxin_3C June09_2clean.csl
%RAT_NON_GEST_ICF_F083109.CSL (now 09-11 -09)
%dose levels: 0.003, 0.010, 0.022, 0.046, 0.1 ug/kg 5 days/week for 53 weeks
%dose levels equivalent to: 3, 10, 22, 46, 100 ng/kg 5 days/week for 53 weeks
%dose levels equivalent to: 2.14, 7.14, 15.7, 32.9, 71.4 ng/kg 7 days/week for 53 weeks
MAXT = 0.01
CINT =0.1
EXP TIME ON = 0. %delay before begin exposure (HOUR)
EXP TIME OFF = 17640 %TIME EXPOSURE STOP (HOUR)
DAY CYCLE = 24
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-56 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
WEEKPERIOD = 168
WEEKFINISH =119
BCKTIMEON = 0.
BCKTIMEOFF = 0.
TIMELIMIT = 8904
BW TO
%DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
%TIME OF BACKGROUND EXPOSURE STOP (HOUR)
%SIMULATION LIMIT TIME (HOUR)
= 215 % Body weight at the beginning of the simulation (g)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.003 % exposure dose ug/kg
%MSTOT = 0.010 % exposure dose ug/kg
%MSTOT = 0.022 % exposure dose ug/kg
%MSTOT = 0.046 % exposure dose ug/kg
MSTOT =0.1 % exposure dose ug/kg
C.2.3.2.17. NTP (2006) 2 year.
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% NTP 2006
%built and check in August 7 2009
%protocol: oral exposure for 105 weeks; SD rats
%dose levels: 0.003, 0.010, 0.022, 0.046, 0.1 ug/kg 5 days/week for 105 weeks
%dose levels equivalent to: 3, 10, 22, 46, 100 ng/kg 5 days/week for 105 weeks
%dose levels equivalent to: 2.14, 7.14, 15.7, 32.9, 71.4 ng/kg 7 days/week for 105 weeks
MAXT = 0.01
CINT =0.1
EXP TIME ON = 0. %TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXP TIME OFF = 17640 %TIME AT WHICH EXPOSURE ENDS (HOUR)
DAY CYCLE = 24
WEEKPERIOD = 168
WEEKFINISH =119
BCK TIME ON = 0. %TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCK TIME OFF = 0. %TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
TIMELIMIT = 17640 %SIMULATION TIME LIMIT (HOUR)
BW T0 =215 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.003 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.010 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.022 % EXPOSURE DOSE IN UG/KG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-57 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
%MSTOT =0.046 % EXPOSURE DOSE IN UG/KG
MSTOT = 0.1 % EXPOSURE DOSE IN UG/KG
C.2.3.2.18. Sewalletal. (1995).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIADJ BBNGKG CBNDLINGKG
% Sewall et al. 1995
%Rat_Dioxin_3C June09_2clean.csl
%RAT_NON_GEST_ICF_F083109.CSL (now 09-11 -09)
%protocol: gavage every 2 weeks for 30 weeks
%dose levels: 0.049, 0.1498, 0.49, and 1.75 ug/kg every two weeks
%dose levels: 3.5, 10.7, 35, and 125 ng/kg/d or 49, 149.8, 490, and 1750 ng/kg every two weeks
MAXT = 0.01
CINT = 0.1
EXPTIMEON = 0.
experiment (age =12 weeks)
EXPTIMEOFF = 5030
weeks
DAY CYCLE = 336.
B CKTIMEON = 0.
B CKTIMEOFF = 0.
TIMELIMIT = 5040
BWT0 = 250
to 12 week old female
%delay before begin exposure (HOUR) 5 weeks after start of
%TIME EXPOSURE STOP (HOUR); 30 doses, 1 every two
% once every two weeks
%DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
%TIME OF BACKGROUND EXPOSURE STOP (HOUR)
%SIMULATION LIMIT TIME (HOUR)
% Body weight at the beginning of the simulation (g); corresponds
%EXPOSURE DOSE SCENARIOS (UG/KG)
MSTOT =0.049 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT = 0.1498 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT =0.49 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT = 1.75 % ORAL EXPOSURE DOSE (UG/KG)
C.2.3.2.19. Shi et al. (2007), adult portion.
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIAD J BBNGKG CBNDLINGKG
% Shi et al 2007
%built and check in August 7 2009
%protocol: gavage once per week for 322 days
%dose levels: 0.001, 0.005, 0.05 and 0.2 ug TCDD:kg body weight by gavage once per week
%dose levels: 1, 5, 50 and 200 ng/kg ng TCDD:kg body weight by gavage once per week
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-58 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
% dose equivalent adjusted 0.143, 0.714, 7.14 and 28.6 ng/kg/d
MAXT = 0.0001
CINT = 0.1
EXPTIMEON = 504. % TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXPTIMEOFF = 7728 %TIME AT WHICH EXPOSURE ENDS (HOUR);
CORRESPONDS TO 322 DAYS OF EXPOSURE
DAY CYCLE =168.
BCKTIMEON = 0. % TIME AT WHICH BACKGROUND EXPOSURE
BEGINS (HOUR)
BCKTIMEOFF = 0. % TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
TIMELIMIT = 7728 %SIMULATION TIME LIMIT (HOUR)
BW T0 = 4.5 % BODY WEIGHT AT THE BEGINNING OF THE
SIMULATION (G)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.001 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.005 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.05 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT = 0.2 % ORAL EXPOSURE DOSE IN UG/KG
C.2.3.2.20. Van Birgelen et al. (1995).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Van Birgelen et al. (1995)
%protocol: daily dietary exposure for 13 weeks
%dose levels: 0.0135, 0.0264, 0.0469, 0.320, 1.024 ug/kg every day for 13 weeks
% dose levels = 13.5, 26.4, 46.9, 320, 1024 ng/kg every day for 13 weeks
MAXT = 0.01
CINT = 0.1
EXP TIME ON = 0. %delay before begin exposure (HOUR)
EXP TIME OFF =2184. %TIME EXPOSURE STOP (HOUR)
DAY_CYCLE = 24. % once every two weeks
BCK TIME ON = 0. %DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
BCK TIME OFF = 0. %TIME OF BACKGROUND EXPOSURE STOP (HOUR)
TIMELIMIT =2184. % SIMULATION LIMIT TIME (HOUR)
BW T0 = 150. % Body weight at the beginning of the simulation (g)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.0135 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT =0.0264 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT =0.0469 % ORAL EXPOSURE DOSE (UG/KG)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-59 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
%MSTOT =0.320 % ORAL EXPOSURE DOSE (UG/KG)
MSTOT = 1.024 % ORAL EXPOSURE DOSE (UG/KG)
C.2.3.2.21. Vanden Heuvel et al (1994).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIADJ BBNGKG CBNDLINGKG
% Vanden Heuvel et al., 1994.
%built and check in August 7 2009
%protocol: single gavage
%Rat_Dioxin_3C June09_2clean.csl
%RAT_NON_GEST_ICF_F083109.CSL (now 09-11 -09)
%dose levels: 0.00005, 0.0001, 0.001, 0.010, 0.1, 1, 10 ug/kg/d + 4 days post treatment
%dose levels equivalent to: 0.05, 0.1, 1, 10, 100, 1000, 10000 ng/kg/d + 4 days post treatment
MAXT
CINT
= 0.01
= 0.01
EXPTIMEON = 0.
EXPTIMEOFF = 120
DAY CYCLE = 120
B CKTIMEON = 0.
B CKTIMEOFF = 0.
TIMELIMIT = 120
BW TO
= 250
%delay before begin exposure (HOUR)
%TIME EXPOSURE STOP (HOUR)
%DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
%TIME OF BACKGROUND EXPOSURE STOP (HOUR)
% SIMULATION LIMIT TIME (HOUR)
% Body weight at the beginning of the simulation (g)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.00005 % exposure dose ug/kg
%MSTOT =0.0001 % exposure dose ug/kg
%MSTOT =0.001 % exposure dose ug/kg
%MSTOT = 0.01 % exposure dose ug/kg
%MSTOT =0.1 % exposure dose ug/kg
MSTOT = 1 % exposure dose ug/kg
%MSTOT =10 % exposure dose ug/kg
C.2.4. Rat Gestational Model
C.2.4.1. Model Code
PROGRAM: 'Three Compartment PBPK Model for TCDD in Rat (Gestation)'
! Parameters were change May 16, 2002
! Come from { 8MAI CHR PRE-EXP GD}
! Come from {12_Mouse_GD}file
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-60 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! { {IMPORTANT-IMPORTANT-IMPORTANT-IMPORTANT} }
! REDUCTION OF MOTHER AND FETUS COMPARTMENT
! 2M R TCDD JULY2002 ////(JULY 18,2002)////
! TCDD_RED_4Species_2003_4 ////(APR 8 ,2003)////
! TCDD_RED_4Species_2003_9 ////(APR 17,2003)////
!TCDD_RED_4Species_2003_12 ////(APR 17,2003)////
! APRIL 18 2003
! TCDD_4C_4SP_2003 ////(APR 18,2003)////
! was "Gest 4 species 1 .csl" but update July 2009
!DevTCDD4Species_ICF_afterKKfix_v3_ratgest.csl
! RATGEST ATIONALICFF083109. csl
! RATGEST ATIONALICFF100609. csl
ILegend/Legend/Legend/Legend/Legend/Legend/Legend/Legend/
!Legend for this PBPK model
IMating: control the tenure of exchange between fetus and
IMother and also control imitated tissue growth
!Control: WTFE, WFO, WPLA0, QPLAF,WT0
! (for rat, mouse, human, and monkey)
! Control transfer from mother to fetus or fetus to mother by TRANSTIMEON
! SWITCHtrans = 0 NO TRANSFER
! SWITCHtrans = 1 TRANSFER OCCURS
IGestoff = 1
!Gest_on= 0.0
! These switches are also controlled by mating parameters
INITIAL !
! SIMULATION PARAMETERS ====
CONSTANT PARA ZERO = 1E-30
CONSTANT EXP TIME ON = 0.0
(HOURS)
CONSTANT EXP TIME OFF = 530
CONSTANT DAY CYCLE = 24.0
(HOURS)
CONSTANT BCK TIME ON = 0.0
BEGINS (HOURS)
CONSTANT BCK TIME OFF = 0.0
ENDS (HOURS)
CONSTANT TRANSTIME ON =144.0 !CONTROL TRANSFER FROM MOTHER TO
FETUS AT GESTATIONAL DAY 6
!UNIT CONVERSION
TIME AT WHICH EXPOSURE BEGINS
! TIME AT WHICH EXPOSURE ENDS (HOURS)
NUMBER OF HOURS BETWEEN DOSES
! TIME AT WHICH BACKGROUND EXPOSURE
! TIME AT WHICH BACKGROUND EXPOSURE
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-61 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CONSTANT MW=322 ! MOLECULAR WEIGHT (NG/NMOL)
CONSTANT SERBLO = 0.55
CONSTANT UNITCORR = 1000
! INTRAVENOUS SEQUENCE
constant IVLACK =0.0
constant IVPERIOD =0.0
! PREGNANCY PARAMETER ====
CONSTANT MATTING = 0.0 BEGINNING OF MATING (HOUR)
CONSTANT NFETUS = 10.0 INUMBER OF FETUS PRESENT
! CONSTANT EXPOSURE CONTROL ===========
! ACUTE, SUBCHRONIC, CHRONIC EXPOSURE =====
!OR BACKGROUND EXPOSURE (IN THIS CASE 3 TIMES A DAY)===
CONSTANT MSTOTBCKGR =0.0 ! ORAL BACKGROUND EXPOSURE DOSE
(UG/KG)
CONSTANT MSTOT =0.0 ! ORAL EXPOSURE DOSE (UG/KG)
!ORAL ABSORPTION
MSTOT NM = MSTOT/MW ! CONVERTS THE DOSE TO NMOL/G
!INTRAVENOUS ABSORPTION
CONSTANT DOSEIV = 0.0 ! INJECTED DOSE (UG/KG)
DOSEIV NM = DOSEIV/MW ! CONVERTS THE INJECTED DOSE TO NMOL/G
CONSTANT DOSEIVLATE = 0.0 ! INJECTED DOSE LATE (UG/KG)
DOSEIVNMlate = DOSEIVLATE/MW ! AMOUNT IN NMOL/G
!INITIAL GUESS OF THE FREE CONCENTRATION IN THE LIGAND
(COMPARTMENT INDICATED BELOW)====
CONSTANT CFLLI0 =0.0 ILIVER (NMOL/ML)
CONSTANT CFLPLA0 =0.0 !PLACENTA (NMOL/ML)
!BINDING CAPACITY (AhR) FOR NON LINEAR BINDING (COMPARTMENT
INDICATED BELOW) (NMOL/ML) ===
CONSTANT LIBMAX = 3.5E-4 ! LIVER (NMOL/ML), WANG ET AL. 1997
CONSTANT PLABMAX = 2.0E-4 ! TEMPORARY PARAMETER
! PROTEIN AFFINITY CONSTANTS (1A2 OR AhR, COMPARTMENT INDICATED
BELOW) (NMOL/ML)===
CONSTANT KDLI = 1.OE-4 !LIVER (AhR) (NMOL/ML), WANG ET AL. 1997
CONSTANT KDLI2 = 4.0E-2 ILIVER (1A2) (NMOL/ML), EMOND ET AL. 2004
CONSTANT KDPLA = 1.OE-4 ! TEMPORARY PARAMETER; ASSUME IDENTICAL
TO KDLI (AhR)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-62 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
!EXCRETION AND ABSORPTION CONSTANT
CONSTANT KST = 0.36 ! GASTRIC RATE CONSTANT (HR-1), WANG ET AL.
1997
CONSTANT KABS =0.48 !INTESTINAL ABSORPTION CONSTANT (HR-1) ),
WANGET AL. 1997
! ELIMINATION CONSTANTS
CONSTANT CLURI = 0.01 ! URINARY CLEARANCE (ML/HR), EMOND ET AL.
2004
! INTERSPECIES ELIMINATION VARIABLE
CONSTANT kelv =0.15 ! INTERSPECIES VARIABLE ELIMINATION
CONSTANT (1/HOUR)
! CONSTANT TO DIVIDE THE ABSORPTION INTO LYMPHATIC AND PORTAL
FRACTIONS
CONSTANT A =0.7 ! LYMPHATIC FRACTION WANG ET AL. 1997
!PARTITION COEFFICIENTS
CONSTANT PF =100
CONSTANT PRE =1.5
CONSTANT PLI = 6.0
CONSTANT PPLA =1.5
WANGET AL. 1997
ADIPOSE TISSUE/BLOOD, WANGET AL. 1997
! REST OF THE BODY/BLOOD, WANGET AL. 1997
! LIVER/BLOOD, WANGET AL. 1997
! TEMPORARY PARAMETER NOT CONFIGURED,
!PARAMETER FOR INDUCTION OF CYP 1A2, WANG ET AL. 1997
CONSTANT PAS INDUC =1.0 ! INCLUDE INDUCTION? (1 = YES, 0 = NO)
CONSTANT CYP 1 A2_l OUTZ =1.6 ! DEGRADATION CONCENTRATION
CONSTANT OF 1A2 (NMOL/ML)
CONSTANT CYP1A21A1 = 1.6
CONSTANT CYP1A21EC50 =0.13
(NMOL/ML)
CONSTANT CYP1A21A2 =1.6
CONSTANT CYP1A21KOUT = 0.1
CONSTANT CYP 1A21TAU =0.25
CONSTANT CYP 1 A2_ 1 EM AX = 600
EFFECT (UNTITLES S)
CONSTANT HILL = 0.6 !HILL CONSTANT; COOPERATIVELY LIGAND
BINDING EFFECT CONSTANT (UNITLESS)
! BASAL CONCENTRATION OF 1A1 (NMOL/ML)
! DISSOCIATION CONSTANT TCDD-CYP1A2
IBASAL CONCENTRATION OF 1A2 (NMOL/ML)
! FIRST ORDER RATE OF DEGRADATION (H-l)
! HOLDING TIME (H)
! MAXIMUM INDUCTION OVER BASAL
! DIFFUSIONAL PERMEABILITY FRACTION
CONSTANT PAFF
CONSTANT PAREF
1997
CONSTANT PALIF
CONSTANT PAPLAF
= 0.0910 ! ADIPOSE (UNITLESS), WANG ET AL. 1997
= 0.0298 !REST OF THE BODY (UNITLESS), WANG ET AL.
= 0.3500 !LIVER (UNITLESS), WANG ET AL. 1997
= 0.3 ITEMPORARY PARAMETER NOT CONFIGURED
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-63 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
!FRACTION OF TISSUE WEIGHT =========
CONSTANT WLIO = 0.0360 ILIVER, WANG ET AL. 1997
!TISSUE BLOOD FLOW EXPRESSED AS A FRACTION OF CARDIAC OUTPUT
CONSTANT QFF = 0.069 ! ADIPOSE TISSUE BLOOD FLOW FRACTION
(UNITLESS), WANGETAL. 1997
CONSTANT QLIF =0.183 ILIVER (UNITLESS), WANG ET AL. 1997
!COMPARTMENT TISSUE BLOOD EXPRESSED AS A FRACTION OF THE TOTAL
COMPARTMENT VOLUME
CONSTANT WFB0 =0.050 ! ADIPOSE TISSUE, WANG ET AL. 1997
CONSTANT WREB0 = 0.030 !REST OF THE BODY, WANG ET AL. 1997
CONSTANT WLIB0 = 0.266 ILIVER, WANG ET AL. 1997
CONSTANT WPLAB0 = 0.500 ! TEMPORARY PARAMETER NOT CONFIGURED
! EXPO SURE SCENARIO FOR UNIQUE OR REPETITIVE WEEKLY OR MONTHLY
EXPOSURE
INUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK = 0.0 IDELAY BEFORE EXPOSURE ENDS (WEEK)
CONSTANT WEEK PERIOD =168 ! NUMBER OF HOURS IN THE WEEK (HOURS)
CONSTANT WEEK FINISH =168 ! TIME EXPOSURE ENDS (HOURS)
INUMBER OF EXPOSURES PER MONTH
CONSTANT MONTH LACK =0.0 IDELAY BEFORE EXPOSURE BEGINS
(MONTHS)
I CONSTANT FOR BACKGROUND EXPOSURE===========
CONSTANT Day LACK BG = 0.0 IDELAY BEFORE EXPOSURE BEGINS (HOURS)
CONSTANT DayPERIODBG = 24 I LENGTH OF EXPOSURE (HOURS)
INUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK BG = 0.0
BEGINS (WEEKS)
CONSTANT WEEK PERIOD BG =168
(HOURS)
CONSTANT WEEK FINISH BG =168
IDELAY BEFORE BACKGROUD EXPOSURE
INUMBER OF HOURS IN THE WEEK
I TIME EXPOSURE ENDS (HOURS)
I INITIAL BODY WEIGHT
CONSTANT BW T0 = 250 I WANG ET AL. 1997
CONSTANT RATIORATFMOU SEF = 1.0 I RATIO OF FETUS MOUSE/RAT AT
GESTATIONAL DAY 22
I COMPARTMENT LIPID EXPRESSED AS THE FRACTION OF TOTAL LIPID, POULIN
ET AL 2002
CONSTANT F TOTLIP = 0.855 I ADIPOSE TISSUE (UNITLESS)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-64 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CONSTANT B TOTLIP = 0.0023
CONSTANT RE TOTLIP =0.019
CONSTANT LITOTLIP = 0.060
CONSTANT PLA TOTLIP =0.019
CONSTANT FETUSTOTLIP =0.019
END ! END OF THE INITIAL SECTION
! BLOOD (UNTITLES S)
! REST OF THE BODY (UNITLESS)
! LIVER (UNITLESS)
DYNAMIC ! DYNAMIC SIMULATION SECTION
ALGORITHM IALG = 2 ! GEAR METHOD
CINTERVAL CINT = 0.1 ! COMMUNICATION INTERVAL
MAXTERVAL MAXT = 1.0e+10 ! MAXIMUM CALCULATION INTERVAL
MINTERVAL MINT = 1.0E-10 ! MINIMUM CALCULATION INTERVAL
VARIABLE T = 0.0
CONSTANT TIMELIMIT = 100 ! SIMULATION LIMIT TIME (HOURS)
CINTXY = CINT
PFUNC = CINT
!TIME CONVERSION
DAY = T/24 ! TIME IN DAYS
WEEK = T/168 ! TIME IN WEEKS
MONTH = T/730 ! TIME IN MONTHS
YEAR = T/8760 ! TIME IN YEARS
DERIVATIVE ! PORTION OF CODE THAT SOLVES DIFFERENTIAL EQUATIONS
!CHRONIC OR SUBCHRONIC EXPOSURE SCENARIO =======
INUMBER OF EXPOSURES PER DAY
DAY LACK = EXP TIME ON ! DELAY BEFORE EXPOSURE BEGINS (HOURS)
DAY PERIOD = DAY CYCLE ! EXPOSURE PERIOD (HOURS)
DAY FINISH = CINTXY ! LENGTH OF EXPOSURE (HOURS)
MONTH PERIOD = TIMELIMIT ! EXPOSURE PERIOD (MONTHS)
MONTH FINISH = EXP TIME OFF ! LENGTH OF EXPOSURE (MONTHS)
INUMBER OF EXPOSURES PER DAY AND MONTH
DAYFINISHBG = CINTXY
MONTH LACK BG = BCK TIME ON IDELAY BEFORE BACKGROUD EXPOSURE
BEGINS (MONTHS)
MONTH PERIOD BG = TIMELIMIT BACKGROUND EXPOSURE (MONTHS)
MONTH FINISH BG = BCK TIME OFF !LENGTH OF BACKGROUND EXPOSURE
(MONTHS)
! INTRAVENOUS LATE
IVFINISH = CINTXY
B = 1-A ! FRACTION OF DIOXIN ABSORBED IN THE PORTAL FRACTION OF THE
LIVER
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-65 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
IFETUS,VOLUME,FETUS,VOLUME,FETUS,VOLUME,FETUS,VOLUME,FETUS,VOLUM
E,FETUS,VOLUME
! FROM OFLAHERTYl 992
RTESTGEST= T-MATTING
TESTGEST=DIM(RTESTGEST,0.0)
WTFER_RODENT= (2.3d-3*EXP(1.49d-2*(TESTGEST))+1.3d-2)*Gest_on
WTFER = (WTFER_RODENT*RATIO_RATF_MOU SEF *N_FETU S)
WTFE = DIM(WTFER,0.0)
FAT,VOLUME,FAT,VOLUME,FAT,VOLUME,FAT,VOLUME,FAT,VOLUME,F AT,VOLU
ME,FAT,VOLUME
! FAT GROWTH EXPRESSION LINEAR DURING PREGNANCY
! FROM O'FLAHERTYl 992
WF0= (((9.66d-5*(TESTGEST))*gest_on)+0.069)
! PLACENTA, VOLUME, PLACENTA, VOLUME, PLACENTA,VOLUME,
PLACENTA,VOLUME
! WPLA PLACENTA GROWTH EXPRESSION, SINGLE EXPONENTIAL WITH OFFSET
! FROM O'FLAHERTY l 992 ! FOR EACH PUP
WPLAONRODENT = (0.6/(l+(5d+3*EXP(-0.0225*(TESTGEST)))))*N_FETUS
WPLAOR = (WPL AONRODENT/WTO) * Geston
WPLAO = DIM(WPL AOR, 0.0)
! PLACENTA,FLOW RATE, PLACENTA,FLOW RATE, PLACENTA,FLOW RATE,
PLACENTA,FLOW RATE
! QPLA PLACENTA GROWTH EXPRESSION, DOUBLE EXPONENTIAL WITH OFFSET
! FROM O'FLAHERTY l 992
QPLARF = (1.67d-7 *exp(9.6d-3*(TESTGEST)) &
+1.6d-3*exp(7.9d-3*(TESTGEST))+0.0)*Gest_on*SWITCH_trans
QPLAF=DIM(QPLARF,0.0) ! FRACTION OF FLOW RATE IN PLACENTA
! GESTATION CONTROL
IF (T.LT.MATTING) THEN
Gestoff = 1.0
Gest_on= 0.0
ELSE
Gestoff = 0.0
Gest on =1.0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-66 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
END IF
! MOTHER BODY WEIGHT GROWTH EQUATION========
! MODIFICATION TO ADAPT THIS MODEL AT HUMAN MODEL
! BECAUSE LINEAR DESCRIPTION IS NOT GOOD ENOUGH FOR MOTHER GROWTH
! MOTHER BODY WEIGHT GROWTH
PARAMETER (BW RMN = 1.0E-30)
WT0= BW TO *(1+(0.41*T)/(1402.5+T+BW_RMN))
! VARIABILITY OF REST OF THE BODY DEPENDS ON OTHER ORGANS
WREO = (0.91 - (WLIB0*WLI0 + WFB0*WF0 +WPLAB0*WPLA0 + WLI0 + WF0 +
WPLAO))/(1 + WREB0) ! REST OF THE BODY FRACTION; UPDATED FOR EPA
ASSESSMENT
QREF = 1 -(QFF+QLIF+QPLAF) !REST OF BODY BLOOD FLOW RATE (ML/HR)
QTTQF = QFF+QREF+QLIF+QPLAF ! SUM MUST EQUAL 1
! COMPARTMENT VOLUME (ML OR G) =========
WF = WF0 * WTO ! ADIPOSE TISSUE
WRE = WREO * WTO ! REST OF THE BODY
WLI = WLI0 * WTO ! LIVER
WPLA= WPL AO* WTO ! PLACENTA
! COMPARTMENT TISSUE BLOOD (ML OR G) =======
WFB = WFB0 * WF ! ADIPOSE TISSUE
WREB = WREB0 * WRE ! REST OF THE BODY
WLIB = WLIB0 * WLI ! LIVER
WPLAB = WPLAB0* WPLA ! PLACANTA
! CARDIAC OUTPUT FOR THE GIVEN BODY WEIGHT (ML/H) =========
!QC= QCCAR*60*(WT0/1000.0)* *0.75
CONSTANT QCC=18684.0 ! EQUIVALENT TO 311.4 * 60
QC= QCC*(WT0/UNITCORR)**0.75
! COMPARTMENT BLOOD FLOW RATE (ML/HR)
QF = QFF*QC IADIPOSE TISSUE BLOOD FLOW RATE
QLI = QLIF*QC !LIVER TISSUE BLOOD FLOW RATE
QRE = QREF*QC !REST OF THE BODY BLOOD FLOW RATE
QPLA = QPLAF*QC !PLACENTA TISSUE BLOOD FLOW RATE
QTTQ = QF+QRE+QLI+QPLA ! TOTAL FLOW RATE
!PERMEABILITY ORGAN FLOW (ML/HR)=========
PAF =PAFF*QF ! ADIPOSE TISSUE
PARE = PAREF*QRE ! REST OF THE BODY
PALI = PALIF*QLI ! LIVER TISSUE
PAPLA = PAPLAF*QPLA ! PLACENTA
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-67 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
ABSORPTION SECTION
ORAL
INTRAPERITONEAL
INTRAVENOUS
!REPETITIVE ORAL BACKGROUND EXPOSURE SCENARIO
MSTOT NMBCKGR = MSTOTBCKGR/MW ! CONVERTS THE BACKGROUND DOSE
TO NMOL/G
MSTTBCKGR =MSTOT_NMBCKGR *WT0
DAYEXPOSUREBG = PULSE(DAY_LACK_BG,DAY_PERIOD_BG,DAY_FINISH_BG)
WEEKEXPOSUREBG =
PUL SE(WEEK_L ACKB G, WEEKPERIODB G, WEEKFINISHB G)
MONTHEXPO SUREB G =
PUL SE(MONTH_L ACKB G,MONTH_PERIOD_B G,MONTH_FINI SH_B G)
MSTTCHBG =
(DAY_EXPOSURE_BG*WEEK_EXPOSURE_BG*MONTH_EXPOSURE_BG)*MSTTBCK
GR
MSTTFRBG = MSTTBCKGR/CINT
C Y CLEBG =D AYEXPO SUREBG* WEEKEXPO SUREB G*MONTH_EXPO SUREB G
! CONDITIONAL ORAL EXPOSURE (BACKGROUND EXPOSURE)
IF (MSTTCH BG.EQ.MSTTBCKGR) THEN
ABSMSTT_GB= MSTTFRBG
ELSE
ABSMSTTGB = 0.0
END IF
C YCLETOTB G=INTEG(C Y CLEB G, 0.0)
!REPETITIVE ORAL EXPOSURE SCENARIO
MSTT= MSTOT NM * WTO ! AMOUNT IN NMOL
DAYEXPOSURE = PULSE(DAY_LACK,DAY_PERIOD,DAY_FINISH)
WEEKEXPOSURE = PULSE(WEEK_LACK,WEEK_PERIOD,WEEK_FINISH)
MONTHEXPOSURE = PUL SE(MONTH_L ACK,MONTH_PERIOD ,MONTH_FINI SH)
MSTTCH = (DAY EXPO SURE* WEEK EXPO SURE*MONTH EXPO SURE) *MSTT
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-68 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
MSTTFR = MSTT/CINT
CYCLE = DAYEXPO SURE* WEEKEXPO SURE*MONTH_EXPO SURE
SUMEXPEVENT= INTEG (CYCLE,0.0) INUMBER OF CYCLE GENERATE DURING
SIMULATION
! CONDITIONAL ORAL EXPOSURE
IF (MSTTCH.EQ.MSTT) THEN
ABSMSTT= MSTTFR
ELSE
ABSMSTT = 0.0
END IF
C Y CLETOT=INTEG(C Y CLE, 0.0)
! MASS CHANGE IN THE LUMEN
RMSTT= -(KST+KABS)*MST + ABSMSTT +ABSMSTT GB ! RATE OF CHANGE
(NMOL/H)
MST = INTEG(RMSTT,0.0) ! AMOUNT REMAINING IN DUODENUM
(NMOL)
! ABSORPTION IN LYMPH CIRCULATION
LYRMLUM = KABS*MST*A
LYMLUM = INTEG(L YRMLUM, 0.0)
! ABSORPTION IN PORTAL CIRCULATION
LIRMLUM = KABS*MST*B
LIMLUM = INTEG(LIRMLUM, 0.0)
! -—IV EXPOSURE
IV= DOSEIV NM * WTO ! AMOUNT IN NMOL
IVR= IV/PFUNC ! RATE FOR IV INFUSION IN BLOOD
EXPIV= IVR * (1 O-STEP(PFUNC))
IVDOSE = integ(EXPIV,0.0)
! IV LATE IN THE CYCLE
! MODIFICATION ON January 13 2004
IVRlateR = DOSEIVNMlate*WT0
IV_EXPOSURE=PULSE(IV_LACK,IV_PERIOD,IV_FINISH)
IV lateT = IV EXPOSURE *IV_RlateR
IV late = IV lateT/CINT
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-69 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
SUMEXPEVENTIV= integ (IV_EXPOSURE,0.0) ! NUMBER OF CYCLE GENERATE
DURING SIMULATION
! SYSTEMIC CONCENTRATION OF TCDD
! MODIFICATION ON OCTOBER 6, 2009
CB=
(QF * CFB+QRE* CREB+QLI* CLIB+EXPIV+L YRMLUM+QPL A* CPL AB+IV_late)/(QC+CL
URI) !
CA = CB ! CONCENTRATION (NMOL/ML)
! URINARY EXCRETION BY KIDNEY
! MODIFICATION ON OCTOBER 6, 2009
RAURI = CLURI *CB
AURI = INTEG(R AURI, 0.0)
!UNIT CONVERSION POST SIMULATION
CBSNGKGLIADJ=(CB*MW*UNITCORR*(l 0/B_TOTLIP)*(1 O/SERBLO))! [NG of TCDD
Serum/Kg OF LIPIP]
AU CB S_N GKGLI AD J=integ(CB SN GKGLI AD J, 0.0)
PRCT B = (CB/(MSTT+lE-30))* 100.0 IPERCENT OF ORAL DOSE IN BLOOD
PRCT BIV = (CB/(IV_RlateR+lE-30))* 100.0 ! PERCENT OF IV DOSE IN BLOOD
CBNGKG= CB*MW*UNITCORR
! ADIPOSE COMPARTMENT
!TISSUE BLOOD COMPARTMENT
RAFB= QF * (C A-CFB)-PAF * (CFB-CF/PF) !(NMOL/H)
AFB = INTEG(RAFB ,0.0) ! (NMOL)
CFB = AFB/WFB ! (NMOL/ML)
!TISSUE COMPARTMENT
RAF = PAF*(CFB-CF/PF) ! (NMOL/H)
AF = INTEG(RAF,0.0) !(NMOL)
CF = AF/WF !(NMZML)
!UNIT CONVERSION POST SIMULATION
CFTOTAL= (AF + AFB)/(WF + WFB) ! TOTAL CONCENTRATION IN NMOL/ML
CFTFREE = CFB + CF ! TOTAL FREE CONCENTRATION IN FAT (NM/ML)
PRCT F = (CFTOTAL/(MSTT+1E-30))* 100.0 ! PERCENT OF ORAL DOSE IN FAT
PRCT FIV = (CFTOTAL/(IV_RlateR+lE-30))* 100.0 ! PERCENT OF IV DOSE IN FAT
CFNGKG=CFTOTAL*MW*UNITCORR ! FAT CONCENTRATION NG/KG
AUCF_NGKGH=integ(CFNGKG,0.0)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-70 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
!REST OF THE BODY COMPARTMENT
RAREB= QRE *(CA-CREB)-PARE*(CREB-CRE/PRE) !(NMOL/H)
AREB = INTEG(RAREB ,0.0)
CREB = AREB/WREB
!TISSUE COMPARTMENT
RARE = PARE* (CREB - CRE/PRE)
ARE = INTEG(RARE,0.0)
CRE = A RE/W RE
! (NMOL)
! (NMOL/ML)
!(NMOL)
!(NMOL/H)
! (NMOL/H)
!UNIT CONVERSION POST SIMULATION
CRETOTAL= (ARE + AREB)/(WRE + WREB) ! TOTAL CONCENTRATION IN
NMOL/ML
PRCT RE = (CRETOTAL/(MSTT+1E-30))* 100.0 ! PERCENT OF ORAL DOSE IN REST
OF THE BODY
PRCT REIV = (CRETOTAL/(IV_RlateR+lE-30))* 100.0 IPERCENT OF IV DOSE IN
REST OF THE BODY
CRENGKG=CRETOTAL*MW*UNITCORR ! REST OF THE BODY CONCENTRATION
INNG/KG
! LIVER COMPARTMENT
!TISSUE BLOOD COMPARTMENT
RALIB = QLI * (CA-CLIB )-PALI * (CLIB -CFLLIR)+LIRMLUM !
ALIB = INTEG(RALIB ,0.0) !(NMOL)
CLIB = ALIB/WLIB ! (NMOL/ML)
!TISSUE COMPARTMENT
RALI = PALI* (CLIB - CFLLIR)-REXCLI ! (NMOL/HR)
ALI = INTEG(RALI, 0.0) ! (NMOL)
CLI = ALI/WLI ! (NMOL/ML)
!FREE TCDD CONCENTRATION IN LIVER COMPARTMENT
PARAMETER (LIVER IRMN = 1.0E-30)
CFLLI= IMPLC(CLI-(CFLLIR*PLI+(LIBMAX*CFLLIR/(KDLI+CFLLIR &
+LIVER_1RMN))+((CYP 1 A2_l 03 *CFLLIR/(KDLI2 + CFLLIR &
+LIVER_ 1RMN) *P A SINDU C)))-CFLLI, CFLLI0)
CFLLIR=DIM(CFLLI,0.0) ! FREE CONCENTRATION IN LIVER
CBNDLI= LIBM AX* CFLLIR/(KDLI+CFLLIR+LIVER_1 RMN) IBOUND
CONCENTRATION
! VARIABLE ELIMINATION BASED ON THE CYP1A2
KBILELIT =((CYP1 A2_l OUT-CYP1 A2_l A2)/CYP1 A2_l A2)*Kelv ! INDUCED
BILIARY EXCRETION RATE CONSTANT IN LIVER
REXCLI = KBILELIT * CFLLIR* WLI ! DOSE-DEPENDENT BILIARY EXCRETION
RATE
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-71 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
EXCLI = INTEG(REXCLI, 0.0)
!UNIT CONVERSION POST SIMULATION
CLITOTAL= (ALI + ALIB)/(WLI + WLIB) ! TOTAL CONCENTRATION IN NMOL/ML
PRCTLI = (CLITOTAL/(MSTT+1E-30))*100
PRCTLIIV = (CLITOT AL/(I V_R1 ateR+ 1E-30))*100.0
Rec_occ= CFLLIR/(KDLI+CFLLIR)
CLINGKG=CLITOTAL*MW*UNITCORR ! LIVER CONCENTRATION NG/KG
AU CLIN GKGH=INTEG(CLIN GKG, 0.0)
CBNDLINGKG = CBNDLI*MW*UNITCORR
AUCBNDLINGKGH =INTEG(CBNDLINGKG,0.0)
!CHEMICAL IN CYP450 (1A2) COMPARTMENT
CYP1A2 1KINP = CYP1A2 1KOUT* CYP1A2 10UTZ
! MODIFICATION ON OCTOBER 6, 2009
CYP1A210UT =INTEG(CYP1 A2_l KINP * (1.0 + CYP1A21EMAX *(CBNDLI+1.0e-
30)**HILL &
/(CYP 1 A2_ 1 EC5 0 * *HILL + (CBNDLI+1.0e-30)**HILL)) &
- CYP1A2_1K0UT*CYP1A2_10UT, CYP1A210UTZ)
! EQUATIONS INCORPORATING DELAY OF CYP1A2 PRODUCTION (NOT USED IN
SIMULATIONS)
CYP1A21R02 = (CYP1A210UT - CYP1A2102)/ CYP1A21TAU
CYP1A2102 =INTEG(CYP 1 A2_ 1R02, CYP1A21A1)
CYP1A21R03 = (CYP1A2102 - CYP1A2103)/ CYP1A21TAU
CYP1A2103 =INTEG(CYP 1 A2_ 1R03, CYP1A21A2)
! TRANSFER OF DIOXIN FROM PLACENTA TO FETUS
! FETAL EXPOSURE ONLY DURING EXPOSURE
IF (T.LT. TRANSTIME ON) THEN
SWITCHtrans = 0.0
ELSE
SWITCHtrans = 1.0
END IF
! TRANSFER OF DIOXIN FROM PLACENTA TO FETUS
! MODIFICATION 26 SEPTEMBER 2003
CONSTANT PFETUS= 4.0 !
CONSTANT CLPLA FET = 0.17 !
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-72 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
RAMPF = (CLPL AFET * CPLA) *SWITCH_trans
AMPF=INTEG(RAMPF ,0.0)
! TRANSFER OF DIOXIN FROM FETUS TO PLACENTA
RAFPM = (CLPL A FET * CFETUS_v) * S WITCH trans !
AFPM = INTEG(RAFPM, 0.0)
! TCDD IN PLACENTA (MOTHER) COMPARTMENT
RAPLAB= QPLA*(CA - CPLAB)-PAPLA*(CPLAB -CFLPLAR) ! NMOL/H)
APLAB = INTEG(RAPLAB ,0.0) ! (NMOL)
CPLAB = APLAB/(WPLAB+1E-3 0) ! (NMOL/ML)
RAPLA = PAPLA*(CPLAB-CFLPLAR)-RAMPF + RAFPM ! (NMOL/H)
APLA = INTEG(RAPLA, 0.0) ! (NMOL)
CPL A = APLA/(WPLA+le-30) ! (NMOL/ML)
PARAMETER (PARA ZERO = 1.0E-30)
CFLPLA= IMPLC(CPLA-(CFLPLAR*PPLA +(PL ABM AX* CFLPL AR/(KDPL A&
+CFLPLAR+PARA_ZERO)))-CFLPLA,CFLPLA0)
CFLPL AR=DIM(CFLPL A, 0.0)
!UNIT CONVERSION POST SIMULATION
CPLATOTAL= (APLA + APLAB)/((WPLA + WPLAB)+le-30)! TOTAL
CONCENTRATION IN NMOL/ML
PRCTPLA = (CPL ATOT AL/(MSTT +1 E-3 0)) * 100
PRCTPLAIV = (CPLATOTAL/(IV_RlateR+lE-30))*100
!FETUS COMPARTMENT
RAFETUS= RAMPF-RAFPM
AFETU S=INTEG(RAFETU S, 0.0)
CFETU S=AFETU S/(WTFE+1 E-3 0)
CFETOTAL= CFETUS
CFETUSv = CFETU S/PFETU S
! UNIT CONVERSION POST SIMULATION
CFETUSNGKG = CFETUS *MW*UNITCORR ! (NG/KG)
AUCFENGKGH = INTEG(CFETU SN GKG, 0.0)
PRCTFE = (CFETOTAL/(MSTT+1E-30))*100
PRCTFEIV = (CFETOTAL/(IV_RlateR+lE-30))*100
I CONTROL MASS BALANCE
BDOSE= IVDOSE +LYMLUM+LIMLUM
BMASSE = EXCLI+AURI+AFB+AF+AREB+ARE+ALIB+ALI+APL A+APL AB+AFETU S
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-73 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
BDIFF = BDO SE-BM AS SE
!BODY BURDEN (NG)
BODY BURDEN = AFB+AF+AREB+ARE+ALIB+ALI+APLA+APLAB !
BBFETUSNG = AFETUS*MW*UNITCORR ! UNIT (NG)
! BODY BURDEN IN TERMS OF CONCENTRATION (NG/KG)
BBNGKG
=(((AFB+AF+AREB+ARE+ALIB+ALI+APLA+APLAB)/WTO)*MW*UNITCORR) !
AU CBBN GKGH=INTEG(BBN GKG, 0.0)
I COMMAND OF THE END OF SIMULATION
TERMT (T.GE. TimeLimit, 'Time limit has been reached.')
END ! END OF THE DERIVATIVE SECTION
END ! END OF THE DYNAMIC SECTION
END ! END OF THE PROGRAM
C.2.4.2. Input Files
C.2.4.2.1. Bell et al. (2007).
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AU CLIN GKGH AUCFNGKGH AUCB SNGKGLIAD J AUCBBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLI NGKGH
% output @nciout=l T BBFETUSNG %AJS turned off 9/21/09
%Bell et al.2007 (rat species)
%protocol: exposure daily dose in diet for 12 weeks followed by a two week mating time and 21
day gestation period
%DevTCDD4Species.csl
%RAT_GESTATIONAL_ICF_F083109.csl (now 09-11-09)
%dose levels: 0.0024, 0.008, 0.046 ug/kg/d with 0.00003 ug/kg/d background
%dose levels: 2.4, 8, 46 ng/kg/d with 0.03 ng/kg/day background
%EXPOSURES SCENARIOS
MAXT =1
CINT =0.1 %
EXP TIME ON = 0 % delay before begin exposure (HOUR)
EXP TIME OFF = 2856 % TIME EXPOSURE STOP (HOUR) 12 weeks exposure + 2
weeks for mating + 21 days gestation with exposure
DAY CYCLE = 24
BCK TIME ON = 0. % DELAY BEFORE BACKGROUND EXPOSURE (HOUR)
BCK TIME OFF = 2856. % TIME OF BACKGROUND EXPOSURE STOP (HOUR)
IV LACK = 505.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-74 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
IVPERIOD = 505.
TIMELIMIT = 2856 % SIMULATION LIMIT TIME (HOUR)
BWTO = 85
MATTING =2352 % BEGINNING MATTING (HOUR)
TRANSTIME ON = 2496 % SHOULD BE MATTING TIME + 6 DAYS( 144 HOURS)
N FETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.00243 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT = 0.008 % ORAL EXPOSURE DOSE (UG/KG)
MSTOT = 0.0461 % ORAL EXPOSURE DOSE (UG/KG)
C.2.4.2.2. Hojo et al (2002).
%TO BE USED AFTER THE
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCF NGKGH AUCB S NGKGLIAD J AUC BBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLI NGKGH
%Hojo et al. 2002
%protocol: Single oral dose at GD8
%dose levels: 0.02 0.06, and 0.18 ug/kg at GD8
%dose levels: 20, 60, 180 ng/kg at GD8
% author provided the body weight for each group at the beginning of gestation (g)
%20 ng/kg BW = 275g
%60 ng/kg BW = 262g
% 180 ng/kg BW = 278g
%EXPOSURES SCENARIOS
MAXT=0.1
CINT =0.1 %
EXPTIMEON = 192
EXPTIMEOFF = 505
DAY CYCLE = 505
B CKTIMEON = 0.
(HOUR)
B CKTIMEOFF = 0.
(HOUR)
IVLACK
IV PERIOD
TIMELIMIT
% BW TO
MATTING
= 505
= 505
= 504
= 190
= 0.
This document is a draft for
1/15/10
% TIME AT WHICH EXPOSURE BEGINS (HOUR)
% TIME AT WHICH EXPOSURE ENDS (HOUR)
% TIME AT WHICH BACKGROUND EXPOSURE BEGINS
% TIME AT WHICH BACKGROUND EXPOSURE ENDS
Vo SIMULATION TIME LIMIT (HOUR)
BEGINNING OF MATING (HOUR)
review purposes only and does not constitute Agency policy.
C-75 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
TRANSTIME ON = 144. % SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
NFETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.02 % ORAL EXPOSURE DOSE IN UG/KG
%BW_T0 =275 % AT 20 NG/KG, BW = 271 g
%MSTOT =0.06 % ORAL EXPOSURE DOSE IN UG/KG
%BW_T0 = 262 % AT 60 NG/KG, BW = 275g
MSTOT =0.18 % ORAL EXPOSURE DOSE IN UG/KG
BW T0 = 278 % AT 180 NG/KG, BW = 262g
C.2.4.2.3. Ikeda et al (2005).
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCFNGKGH AUCB S NGKGLIAD J AUC BBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLI NGKGH
%Ikeda et al. 2005 (rat species)
%protocol: loading dose of 400 ng/kg followed by weekly maintenance doses of 80 ng/kg for 6
weeks,
%dose levels: 0.4 ug/kg/day followed by weekly 0.08 ug/kg/day
%dose levels: 400 ng/kg/day followed by weekly 80 ng/kg/day
%EXPOSURES SCENARIOS
MAXT =1
CINT =0.1 %
EXP TIME ON = 0 % TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXP TIME OFF = 1008 % TIME AT WHICH EXPOSURE ENDS (HOUR); PRE-
MATING (2 WEEKS) + MATING (1 WEEK) + GESTATION (3 WEEKS)
DAY CYCLE = 168 % WEEKLY CYCLE
BCK TIME ON = 0. % TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCK TIME OFF = 167. % TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
IVLACK = 505.
IVPERIOD = 505.
TIMELIMIT = 1008 % SIMULATION TIME LIMIT (HOUR)
BWT0 = 250
MATTING = 504 % BEGINNING OF MATING (HOUR)
TRANSTIME ON =648 % SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
N FETUS = 10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-76 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
%EXPOSURE DOSE SCENARIOS (UG/KG)
MSTOT =0.08 % ORAL EXPOSURE DOSE IN UG/KG
MSTOTBCKGR = 0.32 % BACKGROUND EXPOSURE IN UG/KG
C.2.4.2.4. Kattainen et al. (2001).
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCFNGKGH AUCB S NGKGLIAD J AUC BBNGKGH
AUCFENGKGH CBNDLINGKG AUCBNDLI NGKGH
%Kattainen et al. 2001
%protocol: Single gavage at GD15
%dose levels: 0.03 0.1, 0.3, 1 ug/kg at GD15
%dose levels: 30, 100 300, 1000 ng/kg at GD15
MAXT=0.1
CINT =0.1
%EXPOSURES SCENARIOS
EXPTIMEON =336
EXPTIMEOFF = 340
DAY CYCLE = 505
B CKTIMEON = 0.
BEGINS (HOUR)
B CKTIMEOFF = 0.
(HOUR)
IVLACK = 505
IVPERIOD = 505
TIMELIMIT = 504
BWT0 = 190
MATTING =0. °/
TRAN STIMEON = 144.
HOURS)
N FETUS = 10
% TIME AT WHICH EXPOSURE BEGINS (HOUR)
% TIME AT WHICH EXPOSURE ENDS (HOUR)
% TIME AT WHICH BACKGROUND EXPOSURE
% TIME AT WHICH BACKGROUND EXPOSURE ENDS
Vo SIMULATION TIME LIMIT (HOUR)
BEGINNING OF MATING (HOUR)
% SHOULD BE MATING TIME + 6 DAYS (144
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.03 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT =0.1 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT =0.3 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT = 1 % ORAL EXPOSURE DOSE IN UG/KG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-77 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
C.2.4.2.5. Markowski et al. (2001).
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCFNGKGH AUCB SNGKGLIAD J AUCBBNGKGH
AUCFENGKGH CBNDLINGKG AUCBNDLINGKGH
%Markowski et al.2001
%protocol: Single gavage at GD18
%dose levels: 0.02 0.06, 0.18, 1 ug/kg at GDI8
%dose levels: 20, 60, 180 ng/kg at GD18
%EXPOSURES SCENARIOS
MAXT=0.1
CINT =0.1 %
EXPTIMEON = 408
EXPTIMEOFF =415
DAY CYCLE = 505
BCKTIMEON = 0.
(HOUR)
B CKTIMEOFF = 0.
(HOUR)
IVLACK = 505
IVPERIOD = 505
TIMELIMIT = 504
BWT0 = 190
MATTING = 0. % BEGINNING OF MATING (HOUR)
TRANSTIME ON = 144. % SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
NFETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.02 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT =0.06 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT =0.18 % ORAL EXPOSURE DOSE IN UG/KG
C.2.4.2.6. Miettinen et al. (2006).
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AU CLIN GKGH AUCFNGKGH AUCB SNGKGLIAD J AUCBBNGKGH
AUCFENGKGH CBNDLINGKG AU CBNDLIN GKGH
%Miettinnen et al 2006
% TIME AT WHICH EXPOSURE BEGINS (HOUR)
% TIME AT WHICH EXPOSURE ENDS (HOUR)
% TIME AT WHICH BACKGROUND EXPOSURE BEGINS
% TIME AT WHICH BACKGROUND EXPOSURE ENDS
% SIMULATION TIME LIMIT (HOUR)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-78 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
%protocol: Single oral dose at GDI5
%dose levels: 0.03 0.1, 0.3, 1 ug/kg at GD15
%dose levels: 30, 100, 300, 1000 ng/kg at GD15
MAXT=0.1
CINT =0.1 %
%EXPOSURES SCENARIOS
EXP TIME ON = 336 % TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXPTIMEOFF = 340 % TIME AT WHICH EXPOSURE ENDS (HOUR)
DAY CYCLE = 505
BCK TIME ON = 0. % TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCKTIMEOFF = 0. % TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
IVLACK = 505
IVPERIOD = 505
TIMELIMIT = 504 % SIMULATION TIME LIMIT (HOUR)
BWT0 = 180
MATTING = 0. % BEGINNING OF MATING (HOUR)
TRANSTIME ON = 144. % SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
NFETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.03 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.1 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.3 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT = 1 % ORAL EXPOSURE DOSE IN UG/KG
C.2.4.2.7. Murray et al. (1979).
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCF NGKGH AUCB S NGKGLIAD J AUC BBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLI NGKGH
% output @nciout=l T BBFETUSNG %AJS turned off 9/21/09
%Murray et al.1979 (rat species)
%protocol: dietary exposure for 90 days followed by gestation (21 days)
%dose levels: 0.001 0.01, 0.1 ug/kg/d
%dose levels: 1, 10, 100 ng/kg/d
%EXPOSURES SCENARIOS
MAXT =1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-79 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
CINT =0.1 %
EXPTIMEON = 0
EXPTIMEOFF = 2660
DAY CYCLE = 24
BCKTIMEON = 0.
(HOUR)
BCKTIMEOFF = 0.
(HOUR)
IVLACK = 2664
IVPERIOD = 2664
TIMELIMIT = 2664
BWT0 = 85
MATTING =2160
TRAN STIMEON = 2304
N FETUS = 10
% TIME AT WHICH EXPOSURE BEGINS (HOUR)
% TIME AT WHICH EXPOSURE ENDS (HOUR)
% TIME AT WHICH BACKGROUND EXPOSURE BEGINS
% TIME AT WHICH BACKGROUND EXPOSURE ENDS
% SIMULATION TIME LIMIT (HOUR)
% BEGINNING OF MATING (HOUR)
% SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.001 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT =0.01 % ORAL EXPOSURE DOSE N UG/KG
MSTOT = 0.1 % ORAL EXPOSURE DOSE N UG/KG
C.2.4.2.8. Nohara et al (2000).
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCF NGKGH AUCB S NGKGLIAD J AUC BBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLI NGKGH
%Nohara et al 2000
%protocol: exposure daily dose in diet
%dose levels: 0.0125, 0.050, 0.2 or 0.8 ug TCDD:kg body weight by gavage on GD15.
%dose levels: 12.5, 50, 200 or 800 ng TCDD:kg body weight by gavage on GD15.
MAXT=0.1
CINT =0.1 %
%EXPOSURES SCENARIOS
EXP TIME ON = 336 % TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXP TIME OFF = 340 % TIME AT WHICH EXPOSURE ENDS (HOUR)
DAY CYCLE = 505 TIME AT WHICH BACKGROUND EXPOSURE BEGINS (HOUR)
BCK TIME OFF = 0. % TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
IVLACK = 505
IVPERIOD = 505
TIMELIMIT = 504 % SIMULATION TIME LIMIT (HOUR)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-80 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
BWTO = 180
MATTING = 0. % BEGINNING OF MATING (HOUR)
TRANSTIMEON = 144. % SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
NFETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.0125 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.050 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.2 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT = 0.8 % ORAL EXPOSURE DOSE IN UG/KG
C.2.4.2.9. Ohsako et al (2001).
%TO BE USED AFTER THE
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCF NGKGH AUCB S NGKGLIAD J AUC BBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLINGKGH
%Ohsako et al. 2001
%protocol: exposure SINGLE DOSE AT GDI5
%dose levels: 0.0125, 0.05, and 0.2 and 0.8 ug/kg AT GD15
%dose levels: 12.5, 50, 200 and 800 ng/kg AT GD15
%EXPOSURES SCENARIOS
MAXT=0.001
CINT =0.1 %
EXPTIMEON =360
EXPTIMEOFF = 505
DAY CYCLE = 505
B CKTIMEON = 0.
(HOUR)
B CKTIMEOFF = 0.
(HOUR)
IVLACK = 505
IVPERIOD = 505
TIMELIMIT = 504
BWTO = 200
MATTING = 0. % BEGINNING OF MATING (HOUR)
TRAN STIMEON = 144. % SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
NFETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.0125 % ORAL EXPOSURE DOSE IN UG/KG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-81 DRAFT—DO NOT CITE OR QUOTE
% TIME AT WHICH EXPOSURE BEGINS (HOUR)
% TIME AT WHICH EXPOSURE ENDS (HOUR)
% TIME AT WHICH BACKGROUND EXPOSURE BEGINS
% TIME AT WHICH BACKGROUND EXPOSURE ENDS
% SIMULATION TIME LIMIT (HOUR)
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
%MSTOT =0.05 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.20 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT =0.80 % ORAL EXPOSURE DOSE IN UG/KG
C.2.4.2.10. Schantz et al. (1996) andAmin et al. (2000).
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCFNGKGH AUCB S NGKGLIAD J AUC BBNGKGH
AUCFENGKGH CBNDLINGKG AUCBNDLI NGKGH
%Amin et al 2000 (rat species) and Schantz et al 1995
%protocol: Daily doses during gestation day 10 to 16
%DevTCDD4Species.csl
%RAT_GESTATIONAL_ICF_F083109.csl (now 09-11-09)
%dose levels: 25 and 100 ug/kg/day
%dose levels: 0.25 and 0.100 ng/kg/day
%EXPOSURES SCENARIOS
MAXT =1
CINT =0.1 %
EXP TIME ON = 240. % delay before begin exposure (HOUR)
EXP TIME OFF = 384. % TIME EXPOSURE STOP (HOUR) 12 weeks exposure + 2
weeks for mating + 21 days gestation with exposure
DAY_CYCLE =24 % weekly cycle
BCK TIME ON = 1000. % DELAY BEFORE BACKGROUND EXPOSURE
(HOUR)
BCK TIME OFF = 1000. % TIME OF BACKGROUND EXPOSURE STOP (HOUR)
IVLACK = 505.
IVPERIOD = 505.
TIMELIMIT = 384. % SIMULATION LIMIT TIME (HOUR)
BWT0 = 250.
MATTING = 0 % BEGINNING MATTING (HOUR)
TRANSTIME ON = 144. % SHOULD BE MATTING TIME + 6 DAYSQ44 HOURS)
NFETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 025 % ORAL EXPOSURE DOSE (UG/KG)
MSTOT =.100
MSTOTBCKGR = 0 % Background Exposure (UG/KG)
C.2.4.2.11. Seo et al. (1995).
%clear variable
output @clear
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-82 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCFNGKGH AUCB SNGKGLIAD J AUCBBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLI NGKGH
%Seo et al. 1995
%protocol: exposure GD 10-16
%DevTCDD4Species.csl
%RAT_GESTATIONAL_ICF_F083109.csl (now 09-11-09)
%dose levels: 0.025 and 0.1 ug/kg GD 10-16
%dose levels: 25 and 100 ng/kg GD 10-16
MAXT=0.1
CINT =0.1
%EXPOSURES SCENARIOS
EXPTIMEON = 240
EXPTIMEOFF =385
DAY CYCLE = 24
B CKTIMEON = 0.
(HOUR)
B CKTIMEOFF = 0.
IVLACK = 505
IVPERIOD = 505
TIMELIMIT = 504
BWT0 = 190
MATTING = 0.
TRAN STIMEON = 144.
HOURS)
N FETUS = 10
% delay before begin exposure (HOUR)
% TIME EXPOSURE STOP (HOUR)
% DELAY BEFORE BACKGROUND EXPOSURE
% TIME OF BACKGROUND EXPOSURE STOP (HOUR)
% SIMULATION LIMIT TIME (HOUR)
% BEGINNING MATING (HOUR)
% SHOULD BE MATING TIME + 6 DAYS (144
%EXPOSURE DOSE SCENARIOS (UG/KG)
MSTOT = 0.025 % ORAL EXPOSURE DOSE (UG/KG)
%MSTOT = 0.1 % ORAL EXPOSURE DOSE (UG/KG)
C.2.4.2.12. Shi et al (2007).
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AU CLIN GKGH AUCFNGKGH AUCB SNGKGLIAD J AUCBBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLI NGKGH
% output @nciout=l T BBFETUSNG %AJS turned off 9/21/09
%Shi et al 2007
%protocol: exposure at GDI4 and GD21 orl exposure
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-83 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
%dose levels: 0.001, 0.005, 0.05 and 0.2 ug TCDD:kg body weight by gavage on GD14 and
GD21.
%dose levels: 1, 5, 50 and 200 ng/kg ng TCDD:kg body weight by gavage on GD14 and GD21.
% dose equivalent adjusted 0.143, 0.714, 7.14 and 28.6 ng/kg/d
MAXT=0.001
CINT =0.1 %
CFLI0 = 0
CFPLA0 = 0
%EXPOSURES SCENARIOS
EXP TIME ON =312 % TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXPTIMEOFF = 485 % TIME AT WHICH EXPOSURE ENDS (HOUR)
DAY CYCLE = 168
BCK TIME ON = 0. % TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCK TIME OFF = 0. % TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
IVLACK = 505
IVPERIOD = 505
TIMELIMIT = 504 % SIMULATION TIME LIMIT (HOUR)
BW T0 = 190 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
MATTING = 0. % BEGINNING OF MATING (HOUR)
TRANSTIME ON = 144. % SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
NFETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.001 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT = 0.005 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.05 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT = 0.2 % ORAL EXPOSURE DOSE IN UG/KG
C.2.5. Mouse Standard Model
C.2.5.1. Model Code
PROGRAM: 'Three Compartment PBPK Model for TCDD in Mice: Standard Model
(Non-Gestation)'
!Mice_Dioxin_3C_June09_l_icf_afterKKfix_v3_mousenongest.csl
! MICENONGEST ATICFF083109. csl
! MICENONGEST ATICFF093 009. csl
! MICENONGEST ATICFF100609. csl
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-84 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
INITIAL ! INITIALIZATION OF PARAMETERS
! SIMULATION PARAMETERS ====
CONSTANT PARA ZERO = 1D-30
CONSTANT EXP TIME ON = 0.0
(HOURS)
CONSTANT EXPTIMEOFF = 2832
(HOURS)
CONSTANT DAY CYCLE = 24
(HOURS)
CONSTANT BCK TIME ON = 0.0
EXPOSURE BEGINS (HOURS)
CONSTANT BCK TIME OFF = 0.0
EXPOSURE ENDS (HOURS)
CONSTANT MW=322 ! MOLECULAR WEIGHT (NG/NMOL)
CONSTANT SERBLO = 0.55
CONSTANT UNITCORR = 1000
!CONSTANT EXPOSURE CONTROL ===========
! ACUTE, SUBCHRONIC, CHRONIC EXPOSURE =====
!OR BACKGROUND EXPOSURE (IN THIS CASE 3 TIMES A DAY)===
CONSTANT MSTOTBCKGR = 0.0 ! ORAL BACKGROUND EXPOSURE DOSE
(UG/KG)
CONSTANT MSTOT = 0.15 ! ORAL EXPOSURE DOSE (UG/KG)
CONSTANT MSTOTsc = 0.0 ! SUBCUTANEOUS EXPOSURE DOSE (UG/KG)
!ORAL ABSORPTION
MSTOT NM = MSTOT/MW ! AMOUNT IN NMOL/G
! INTRAVENOUS ABSORPTION
CONSTANT DOSEIV = 0.0 !INJECTED DOSE (UG/KG)
DOSEIV NM = DOSEIV/MW ! CONVERTS THE INJECTED DOSE TO NMOL/G
!INITIAL GUESS OF THE FREE CONCENTRATION IN THE LIGAND (COMPARTMENT
INDICATED BELOW)====
CONSTANT CFLLI0 = 0.0 !LIVER (NMOL/ML)
!BINDING CAPACITY (AhR) FOR NON LINEAR BINDING (COMPARTMENT
INDICATED BELOW) (NMOL/ML)
CONSTANT LIBMAX = 3.5e-4 ! LIVER (NMOL/ML), WANG ET AL. 1997
! PROTEIN AFFINITY CONSTANTS (1A2 OR AhR, COMPARTMENT INDICATED
BELOW) (NMOL/ML)===
CONSTANT KDLI = 1 Oe-4 ILIVER (AhR)(NMOL/ML), WANG ET AL. 1997
CONSTANT KDLI2 = 2.0e-2 ILIVER (1A2)(NM0L/ML), EMOND ET AL. 2004
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-85 DRAFT—DO NOT CITE OR QUOTE
! TIME AT WHICH EXPOSURE BEGINS
! TIME AT WHICH EXPOSURE ENDS
! NUMBER OF HOURS BETWEEN DOSES
! TIME AT WHICH BACKGROUND
! TIME AT WHICH BACKGROUND
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! ===EXCRETION AND ABSORPTION CONSTANT (OPTIMIZED)
CONSTANT KST = 0.3 ! GASTRIC RATE CONSTANT (HR-1),
CONSTANT KABS = 0.48 !INTESTINAL ABSORPTION CONSTANT (HR-1) ),
WANGET AL. 1997
! ELIMINATION CONSTANTS
CONSTANT CLURI = 0.09 ! URINARY CLEARANCE (ML/HR)
! ==test elimination variable
constant kelv = 0.4 ! INTERSPECIES VARIABLE ELIMINATION CONSTANT
(1/HOUR)
! CONSTANT TO DIVIDE THE ABSORPTION INTO LYMPHATIC AND PORTAL
FRACTIONS
CONSTANT A = 0.7 ! LYMPHATIC FRACTION, WANG ET AL. 1997
!PARTITION COEFFICIENTS OPTIMIZED
CONSTANT PF = 400 ! ADIPOSE TISSUE/BLOOD
CONSTANT PRE = 3 ! REST OF THE BODY/BLOOD, WANG ET AL. 2000
CONSTANT PLI = 6 ! LIVER/BLOOD, WANGET AL. 1997
! ===PARAMETER FOR INDUCTION OF CYP 1A2
CONSTANT PAS_INDUC= 1.0 ! INCLUDE INDUCTION? (1 = YES, 0 = NO)
CONSTANT CYP1A210UTZ =1.6 ! DEGRADATION CONCENTRATION CONSTANT
OF 1A2 (NMOL/ML)
CONSTANT CYP1A21A1 =1.5 ! BASAL CONCENTRATION OF 1A1 (NMOL/ML)
CONSTANT CYP1A21EC50 = 0.13 ! DISSOCIATION CONSTANT TCDD-CYP1A2
(NMOL/ML)
CONSTANT CYP1A21A2 = 1.5 ! BASAL CONCENTRATION OF 1A2 (NMOL/ML)
CONSTANT CYP1A21KOUT = 0.1 ! FIRST ORDER RATE OF DEGRADATION (H-l)
CONSTANT CYP1A21TAU = 1.5 ! HOLDING TIME (H)
CONSTANT CYP1A21EMAX = 600 ! MAXIMUM INDUCTION OVER BASAL EFFECT
(UNTITLES S)
CONSTANT HILL =0.6 !HILL CONSTANT; COOPERATIVELY LIGAND
BINDING EFFECT CONSTANT (UNITLESS)
! DIFFUSIONAL PERMEABILITY FRACTION
CONSTANT PAFF =0.12 ! ADIPOSE (UNITLESS), WANG ET AL. 2000
CONSTANT PAREF = 0.03 ! REST OF THE BODY (UNITLESS)
CONSTANT PALIF = 0.35 ! LIVER (UNITLESS)
!COMPARTMENT TISSUE BLOOD VOLUME =========
CONSTANT WLI0 = 0.0549 ! LIVER, ILSI 1994
CONSTANT WF0 = 0.069 ! ADIPOSE
!TISSUE BLOOD FLOW EXPRESSED AS A FRACTION OF CARDIAC OUTPUT
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-86 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CONSTANT QFF = 0.070 ! ADIPOSE TISSUE BLOOD FLOW FRACTION
(UNITLESS), LEUNG ET AL. 1990
CONSTANT QLIF = 0.161 ! LIVER (UNITLESS) ILSI ET AL. 1994
!COMPARTMENT TISSUE BLOOD EXPRESSED AS A FRACTION OF THE TOTAL
COMPARTMENT VOLUME
CONSTANT WFB0 = 0.050 ! ADIPOSE TISSUE, WANG ET AL. 1997
CONSTANT WREB0 = 0.030 ! REST OF THE BODY, WANG ET AL. 1997
CONSTANT WLIB0 = 0.266 ! LIVER, WANG ET AL. 1997
! EXPOSURE SCENARIO FOR UNIQUE OR REPETITIVE WEEKLY OR MONTHLY
EXPOSURE
! NUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK = 0.0 ! DELAY BEFORE EXPOSURE ENDS (WEEK)
CONSTANT WEEK PERIOD =168 ! NUMBER OF HOURS IN THE WEEK (HOURS)
CONSTANT WEEK FINISH = 120 ! TIME EXPOSURE ENDS (HOURS)
! NUMBER OF EXPOSURES PER MONTH
CONSTANT MONTH LACK = 0.0 ! DELAY BEFORE EXPOSURE (MONTH)
! SET FOR BACKGROUND EXPOSURE===========
! CONSTANT FOR BACKGROUND EXPOSURE===========
CONSTANT Day LACK BG = 0.0 ! DELAY BEFORE EXPOSURE BEGINS (HOURS)
CONSTANT Day PERIOD BG = 24 ! LENGTH OF EXPOSURE (HOURS)
! NUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK BG = 0.0 ! DELAY BEFORE BACKGROUD EXPOSURE
(WEEK)
CONSTANT WEEK PERIOD BG = 168 INUMBER OF HOURS IN THE WEEK (HOURS)
CONSTANT WEEK FINISH BG = 168 ! TIME EXPOSURE ENDS (HOURS)
! GROWTH CONSTANT FOR RAT AND MOUSE
! CONSTANT FOR MOTHER BODY WEIGHT GROWTH ======
CONSTANT BW T0 = 20 ! CHANGED FOR SIMULATION
! CONSTANT USED IN CARDIAC OUTPUT EQUATION, KRISHNAN 2001
CONSTANT QCCAR =275 ! CONSTANT (ML/MIN/KG)
! COMPARTMENT LIPID EXPRESSED AS THE FRACTION OF TOTAL LIPID
CONSTANT F TOTLIP = 0.855 IADIPOSE TISSUE (UNITLESS)
CONSTANT B TOTLIP = 0.0033 IBLOOD (UNITLESS)
CONSTANT RE TOTLIP = 0.019 !REST OF THE BODY (UNITLESS)
CONSTANT LI TOTLIP = 0.06 ILIVER (UNITLESS)
END ! END OF THE INITIAL SECTION
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-87 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
DYNAMIC ! DYNAMIC SIMULATION SECTION
ALGORITHM IALG = 2 ! GEAR METHOD
CINTERVAL CINT = 1.0 ! COMMUNICATION INTERVAL
MAXTERVAL MAXT = 1.0e+10 ! MAXIMUM CALCULATION INTERVAL
MINTERVAL MINT = 1.0E-10 ! MINIMUM CALCULATION INTERVAL
VARIABLE T = 0.0 !HOUR
CONSTANT TIMELIMIT = 2904.0 ! SIMULATION TIME LIMIT (HOURS)
CINTXY = CINT
PFUNC = CINT
!TIME CONVERSION
DAY = T/24.0
WEEK = T/168.0
MONTH = T/730.0
YEAR = T/8760.0
! TIME IN DAYS
! TIME IN WEEKS
! TIME IN MONTHS
! TIME IN YEARS
!NMAX =MAX(T,CTFNGKG)
nmax =max(T,CFNGKG)
DERIVATIVE ! PORTION OF CODE THAT SOLVES DIFFERENTIAL EQUATIONS
!CHRONIC OR SUBCHRONIC EXPOSURE SCENARIO =======
INUMBER OF EXPOSURES PER DAY
DAY LACK = EXP TIME ON ! DELAY BEFORE EXPOSURE BEGINS (HOURS)
DAY PERIOD = DAY CYCLE ! EXPOSURE PERIOD (HOURS)
DAY FINISH = CINTXY ! LENGTH OF EXPOSURE (HOURS)
MONTH PERIOD = TIMELIMIT ! EXPOSURE PERIOD (MONTHS)
MONTH FINISH = EXP TIME OFF ! LENGTH OF EXPOSURE (MONTHS)
INUMBER OF EXPOSURES PER DAY AND MONTH
DAYFINISHBG = CINTXY
MONTH LACK BG = BCK TIME ON ! DELAY BEFORE BACKGROUD EXPOSURE
BEGINS (MONTHS)
MONTH PERIOD BG = TIMELIMIT ! BACKGROUND EXPOSURE PERIOD
(MONTHS)
MONTH FINISH BG = BCK TIME OFF ! LENGTH OF BACKGROUND EXPOSURE
(MONTHS)
! FRACTION OF DIOXIN ABSORBED IN THE PORTAL FRACTION OF THE LIVER
B = 1.0-A
! GROWTH UP EQUATION (G)
PARAMETER (BWRMN = 1.0E-30)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-88 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
WT0= (BW TO *(1.0+(0.41*T)/(1402.5+T+BW_RMN)))
! VARIABILITY OF REST OF THE BODY DEPENDS ON OTHER ORGANS
!REST OF THE BODY FRACTION; UPDATED FOR EPA ASSESSMENT
WREO = (0.91 - (WLIB0*WLI0 + WFB0*WF0 + WLI0 + WF0))/(1+WREB0)
! REST OF THE BODY BLOOD FLOW FRACTION
QREF = 1.0-(QFF+QLIF) !REST OF BODY BLOOD FLOW (ML/HR)
! SUMMATION OF BLOOD FLOW FRACTION (SHOULD BE EQUAL TO 1)
QTTQF = QFF+QREF+QLIF ! SUM MUST EQUAL 1
! COMPARTMENT VOLUME (G)
WF = WF0 * WTO ! ADIPOSE
WRE = WREO * WTO ! REST OF THE BODY
WLI = WLI0 * WTO ! LIVER
!COMPARTMENT TISSUE BLOOD (G)
WFB = WFB0 * WF ! ADIPOSE
WREB = WREB0 * WRE ! REST OF THE BODY
WLIB = WLIB0 * WLI ! LIVER
! CARDIAC OUTPUT FOR THE GIVEN BODY WEIGHT
QC= QCCAR*60*(WT0/1000.0)* *0.75
QF = QFF*QC ! ADIPOSE TISSUE BLOOD FLOW RATE (ML/HR)
QLI = QLIF*QC ! LIVER TISSUE BLOOD FLOW RATE (ML/HR)
QRE = QREF*QC ! REST OF THE BODY BLOOD FLOW RATE (ML/HR)
QTTQ = QF+QRE+QLI ! TOTAL FLOW RATE (ML/HR)
! PERMEABILITY ORGAN FLOW (ML/HR) =======
PAF = PAFF*QF ! ADIPOSE TISSUE
PARE = PAREF*QRE ! REST OF THE BODY
PALI = PALIF*QLI ! LIVER TISSUE
! ABSORPTION SECTION
!ORAL
BACKGROUND EXPOSURE
! EXPO SURE FOR STEADY STATE CONSIDERATION
!REPETITIVE EXPOSURE SCENARIO
MSTOT NMBCKGR = MSTOTBCKGR/322 ! AMOUNT IN NMOL/G
MSTTBCKGR =MSTOT_NMBCKGR *WT0
!REPETITIVE ORAL BACKGROUND EXPOSURE SCENARIOS
DAYEXPOSUREBG = PULSE(DAY_LACK_BG,DAY_PERIOD_BG,DAY_FINISH_BG)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-89 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
WEEKEXPOSUREBG =
PUL SE(WEEK_L ACKB G, WEEKPERIODB G, WEEKFINISHB G)
MONTHEXPO SUREB G =
PUL SE(MONTH_L ACKB G,MONTH_PERIOD_B G,MONTH_FINI SH_B G)
MSTTCHBG =
(DAY_EXPOSURE_BG*WEEK_EXPOSURE_BG*MONTH_EXPOSURE_BG)*MSTTBCK
GR
MSTTFRBG = MSTTBCKGR/CINT
totalBG= integ (MSTTCH_BG,0.0)
CYCLE BG =DAY EXPOSURE BG*WEEK EXPOSURE BG*MONTH EXPOSURE BG
!CONDITIONAL ORAL EXPOSURE (BACKGROUND EXPOSURE)
IF (MSTTCH BG.EQ.MSTTBCKGR) THEN
ABSMSTT_GB= MSTTFRBG
ELSE
ABSMSTTGB = 0.0
END IF
! EXPO SURE + !REPETITIVE EXPOSURE SCENARIO
IV= DOSEIV NM * WTO ! AMOUNT IN NMOL
MSTT= MSTOT NM * WTO ! AMOUNT IN NMOL
DAYEXPOSURE = PULSE(DAY_LACK,DAY_PERIOD,DAY_FINISH)
WEEKEXPOSURE = PULSE(WEEK_LACK,WEEK_PERIOD,WEEK_FINISH)
MONTHEXPOSURE = PULSE(MONTH_LACK,MONTH_PERIOD,MONTH_FINISH)
MSTTCH = (DAY EXPO SURE* WEEK EXPO SURE*MONTH EXPO SURE) *MSTT
CYCLE = DAYEXPO SURE* WEEKEXPO SURE*MONTH_EXPO SURE
SUMEXPEVENT= integ (CYCLE,0.0)*cint INUMBER OF CYCLE GENERATE DURING
SIMULATION
MSTTFR = MSTT/CINT
! CONDITIONAL ORAL EXPOSURE
IF (MSTTCH.EQ.MSTT) THEN
ABSMSTT= MSTTFR
ELSE
ABSMSTT = 0.0
END IF
C Y CLETOT=INTEG(C Y CLE, 0.0)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-90 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
!MASS CHANGE IN THE LUMEN
RMSTT= -(KST+KABS)*MST+ABSMSTT +ABSMSTTGB ! RATE OF CHANGE
(NMOL/H)
MST = INTEG(RMSTT,0.0) ! AMOUNT OF STAY IN DUODENUM (NMOL)
! ABSORPTION IN LYMPH CIRCULATION
LYRMLUM = KABS*MST*A
LYMLUM = INTEG(L YRMLUM, 0.0)
! ABSORPTION IN PORTAL CIRCULATION
LIRMLUM = KABS*MST*B
LIMLUM = INTEG(LIRMLUM, 0.0)
IPERCENT OF DOSE REMAINING IN THE GI TRACT
PRCTremainGIT = (MST/(MSTT+1E-30))*100
RFECES = KST*MST + REXCLI
FECES = INTEG(RFECES,0.0)
prctFECES = (FECES/(BDOSE_TOTAL+1E-30))*100
! ABSORPTION OF DIOXIN BY IV ROUTE
IVR= IV/PFUNC ! RATE FOR IV INFUSION IN BLOOD
EXPIV= IVR * (1 O-STEP(PFUNC))
IVDOSE = integ(EXPIV,0.0)
! SYSTEMIC BLOOD CONCENTRATION (NMOL/ML)
! MODIFICATION ON OCTOBER 6, 2009
CB=(QF * CFB+QRE * CREB+QLI * CLIB+EXPIV+LYRMLUM)/(QC+CLURI) !
CA = CB
! URINARY EXCRETION BY KIDNEY
! MODIFICATION ON OCTOBER 6, 2009
RAURI = CLURI *CB
AURI = INTEG(R AURI, 0.0)
prctAURI = (AURI/(BDOSE_TOTAL+1E-30))*100
!UNIT CONVERSION POST SIMULATION
PRCT B = (CB/(MSTT+1E-30))*100 ! PERCENT OF DOSE/G TISSUE
CBNGKG=CB *MW*UNITCORR
CBSNGKGLIADJ= (CB*MW*UNITCORR*(1.0/B_TOTLIP)*(1.0/SERBLO))![NG of TCDD
Serum/Kg OF LIPIP]
CBPMOL KG= CB*UNITCORR*UNITCORR !CONCENTRATION IN PMOL/KG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-91 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CBNGG = CB*MW
! ADIPOSE TISSUE COMPARTMENT
! TISSUE BLOOD SUB COMPARTMENT
RAFB = QF * (C A-CFB)-P AF * (CFB -CF/PF) !(NMOL/HR)
AFB = INTEG(RAFB ,0.0) ! (NMOL)
CFB = AFB/WFB ! (NMOL/ML)
!TISSUE SUBCOMPARTMENT
RAF = PAF*(CFB-CF/PF) ! (NMOL/HR)
AF = INTEG(RAF,0.0) !(NMOL)
CF = AF/WF ! (NMOL/ML)
IPO ST SIMULATION UNIT CONVERSION
CFTOTAL = (AF + AFB)/(WF + WFB) ! TOTAL CONCENTRATION IN FAT(NM/ML)
PRCT F = (CFTOTAL/(MSTT+1E-30))*100 ! PERCENT OF DOSE IN FAT
CFNGKG = CFTOTAL*MW*UNITCORR
CFUGG=(CFTOTAL*MW)/UNITCORR
CFPMOL KG= CFTOTAL*UNITCORR*UNITCORR !CONCENTRATION IN
PMOL/KG
CFNGG = CFTOTAL*MW
!REST OF THE BODY COMPARTMENT
!TISSUE BLOOD SUBCOMPARTMENT
RAREB= QRE * (C A-CREB )-P ARE * (CREB -CRE/PRE)
! (NMOL/HR)
AREB = INTEG(RAREB ,0.0)
CREB = AREB/WREB
!TISSUE SUBCOMPARTMENT
RARE = PARE* (CREB - CRE/PRE)
ARE = INTEG(RARE,0.0)
CRE = ARE/WRE
!(NMOL)
! (NMOL/ML)
! (NMOL/HR)
!(NMOL)
! (NMOL/ML)
IPO ST SIMULATION UNIT CONVERSION
CRETOTAL= (ARE + AREB)/(WRE + WREB)
STATE
PRCTRE = (CRETOT AL/(MSTT +1E-3 0)) * 100
! CONCENTRATION AT STEADY
!LIVER COMPARTMENT
!TISSUE BLOOD SUBCOMPARTMENT
RALIB = QLI*(CA-CLIB)-PALI*(CLIB-CFLLIR)+LIRMLUM ! (NMOL/HR)
ALIB = INT eg(RALIB ,0.0) ! (NMOL)
CLIB = ALIB/WLIB
!TISSUE SUBCOMPARTMENT
RALI = PALI * (CLIB -CFLLIR)-REXCLI ! (NMOL/HR)
ALI = integ(RALI,0.0) ! (NMOL)
CLI = ALI/WLI ! (NMOL/ML)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-92 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
!FREE TCCD CONCENTRATION IN LIVER (NMOL/ML)
PARAMETER (LIVER1RMN = 1.0E-30)
CFLLI= IMPLC(CLI-(CFLLIR*PLI+(LIBMAX*CFLLIR/(KDLI+CFLLI &
+LI VER_ 1 RMN))+((C YP1 A2_ 103 * CFLLIR/(KDLI2+CFLLIR &
+LI VER_ 1RMN) * P AS_INDUC)))-CFLLI, CFLLIO)
CFLLIR=DIM(CFLLI,0.0) ! FREE CONCENTRATION IN LIVER
CBNDLI= LIBM AX* CFLLIR/(KDLI+CFLLIR+LIVER_1 RMN) IBOUND
CONCENTRATION
!POST SIMULATION UNIT CONVERSION
CLITOTAL= (ALI + ALIB)/(WLI + WLIB)!
PRCT LI = (CLITOTAL/(MSTT+1E-30))*100 ! PERCENT OF DOSE IN LIVER
rec_occ_AHR= (CFLLIR/(KDLI+CFLLIR+lE-30))* 100.0 ! PERCENT OF AhR
OCCUPANCY
PROT_occ_lA2= (CFLLIR/(KDLI2+CFLLIR))* 100.0 ! PERCENT OF 1A2 OCCUPANCY
CLINGKG =(CLITOTAL*MW*UNITCORR)
CBNDLINGKG = CBNDLI*MW*UNITCORR
CLIU GG=(CLITOT AL *MW)/UNIT C ORR
CLIPMOL KG= CLITOTAL*UNITCORR*UNITCORR ! CONCENTRATION IN
PMOL/KG
CLINGG = CLITOTAL*MW
[Fraction increase of induction of CYP1A2
fold_ind=(CYP 1 A2_l OUT/C YP 1 A2_l A2)
V ARIATION Of AC =(CYP 1 A2_l OUT-CYP1 A2_l A2)/CYP 1 A2_l A2
! VARIABLE ELIMINATION BASED ON THE CYP1A2
KBILELIT =((CYP 1 A2_lOUT-CYP 1 A2_l A2)/CYP 1 A2_l A2) *Kelv IINDUCED BILIARY
EXCRETION RATE CONSTANT
REXCLI= (KBILE LI T*CFLLIR*WLI) !DOSE-DEPENDENT EXCRETION RATE
EXCLI = INTEG(REXCLI, 0.0)
!CHEMICAL IN CYP450 (1A2) COMPARTMENT
!EQUATION FOR INDUCTION OF CYP1A2
CYP1A21KINP = CYP1A21KOUT* CYP 1 A2_ 1 OUTZ
! MODIFICATION ON OCTOBER 6, 2009
CYP1A210UT =INTEG(CYP 1 A2_l KINP * (1.0 + CYP1A21EMAX *(CBNDLI+1.0e-
30)**HILL &
/(CYP 1 A2_ 1 EC5 0 * *HILL + (CBNDLI+1.0e-30)**HILL)) &
- CYP1A2_1K0UT*CYP1A2_10UT, CYP1A210UTZ)
! EQUATIONS INCORPORATING DELAY OF CYP1A2 PRODUCTION (NOT USED IN
SIMULATIONS)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-93 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
CYP1A21R02 = (CYP1A210UT - CYP1A2102)/ CYP1A21TAU
CYP1A2102 =INTEG(CYP1 A2_ 1R02, CYP1A21A1)
CYP1A21R03 = (CYP1A2102 - CYP1A2103)/ CYP1A21TAU
CYP1A2103 =INTEG(CYP 1 A2_ 1R03, CYP1A21A2)
! MASS BALANCE CONTROL
BDOSE= LYMLUM+LIMLUM+IVDOSE
B MASSE = EXCLI+AURI+AFB+AF+AREB+ARE+ALIB+ALI
BDIFF = BDOSE-BMASSE
! AMOUNT TOTAL PRESENT IN THE GI TRACT
BDOSETOTAL =L YMLUM+LIMLUM+FECE S
!BODY BURDEN IN NG
Bodyburden =(AFB+AF+AREB+ARE+ALIB+ALI) *MW
!BODY BURDEN CONCENTRATION (NG/KG)
BBNGKG =(((AFB+AF+AREB+ARE+ALIB+ALI) *MW)/(WTO/UNITCORR)) !
! COMMAND FOR END OF SIMULATION
TERMT (T.GE. TimeLimit, 'Time limit has been reached.')
END ! END OF THE DERIVATIVE SECTION
END ! END OF THE DYNAMIC SECTION
END ! END OF PROGRAM
C.2.5.2. Input Files
C.2.5.2.1. Hassoun etal. (1998) (13 weeks).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Hassoun et al 1998
%built and check in August 7 2009
%protocol: oral exposure single dose
%dose levels: 0.00045, 0.0015, 0.015, 0.15 ug/kg single dose + 7 days post exposure
%dose levels: 0.45, 1.5, 15, 150 ng/kg single dose + 7 days post exposure
%dose levels equivalent 0.321, 1.07, 10.7, 107 ng/kg/day
MAXT = 0.01
CINT =0.1
EXP TIME ON = 0. %TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXP TIME OFF = 2184 %2208 %TIME AT WHICH EXPOSURE ENDS (HOUR)
DAY CYCLE = 24
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-94 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
WEEKPERIOD = 168
WEEKFINISH =119
BCKTIMEON = 0. % TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCKTIMEOFF = 0. % TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
TIMELIMIT = 2208 %SIMULATION TIME LIMIT (HOUR)
BW T0 = 23 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION (G)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.00045 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.0015 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.015 % EXPOSURE DOSE IN UG/KG
MSTOT = 0.150 % EXPOSURE DOSE IN UG/KG
NTP (1982) (female) (chronic)
%RAT2.m
%clear variable
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
%output @nciout=168 T SUMEXPEVENT
% NTP subchronic Mice exposure 1982.
%built and check in September 20, 2009
%protocol: repetitive doses
%MICE_NON_GESTAT_ICF_F092009.csl (now 09-20-09)
%dose levels: 0.02, 0.1,1 ug/kg/biweekly, ug/kg for 104 weeks + 3 weeks post treatment
%dose levels: 20, 100 and 1000 ng/kg/Biweekly,ng/kg for 104 weeks + 3 weeks post treatment
%dose levels equivalent to: 5.71, 28.57, 285.1 ng/kg/d
= 0. %TIME AT WHICH EXPOSURE BEGINS (HOUR)
= 17472 %TIME AT WHICH EXPOSURE ENDS (HOUR)
= 84
= 0. %TIME AT WHICH BACKGROUND EXPOSURE BEGINS
MAXT = 0.01
CINT =0.1
EXPTIMEON
EXPTIMEOFF
DAY CYCLE
BCKTIMEON
(HOUR)
B CKTIMEOFF
(HOUR)
TIMELIMIT = 17976 %SIMULATION TIME LIMIT (HOUR)
BW T0 = 23 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
= 0.
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-95 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
%EXPOSURE DOSE SCENARIOS (UG/KG)
MSTOT =0.02 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.1 % EXPOSURE DOSE IN UG/KG
%MSTOT =1.0 % EXPOSURE DOSE IN UG/KG
C.2.5.2.2. NTP (1982) (male) (chronic).
%RAT2.m
%clear variable
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIADJ BBNGKG CBNDLINGKG
%output @nciout=168 T SUMEXPEVENT
% NTP subchronic Mice exposure 1982.
%built and check in September 20, 2009
%protocol: repetitive doses
%dose levels: 0.005, 0.025, 0.25 ug/kg/biweekly, ug/kg for 104 weeks + 3 weeks post treatment
%dose levels: 5, 25 and 250 ng/kg/Biweekly,ng/kg for 104 weeks + 3 weeks post treatment
%dose levels equivalent to: 1.4, 7.1, 71 ng/kg/d
= 0. %TIME AT WHICH EXPOSURE BEGINS (HOUR)
= 17472 %TIME AT WHICH EXPOSURE ENDS (HOUR)
= 84
= 0. % TIME AT WHICH BACKGROUND EXPOSURE BEGINS
MAXT = 0.01
CINT =0.1
EXPTIMEON
EXPTIMEOFF
DAY CYCLE
B CKTIMEON
(HOUR)
B CKTIMEOFF
(HOUR)
TIMELIMIT = 17976 %SIMULATION TIME LIMIT (HOUR)
BW T0 = 25 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
= 0.
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.005 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.025 % EXPOSURE DOSE IN UG/KG
MSTOT = 0.25 % EXPOSURE DOSE IN UG/KG
C.2.5.2.3. Smialowicz et al. (2008).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIAD J BBNGKG CBNDLINGKG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-96 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
% Smialowicz et al, 2008.
%built and check in August 7 2009
%protocol: oral exposure single dose
%protocol: 5/7 gavage for 13 wk; Female B6C3F1 mice
%dose levels: 0, 0.0015, 0.015, 0.15, 0.45 ug/kg
%dose levels: 0, 1.5, 15, 150, 450 nkd (0, 1.07, 10.7, 107, 321 nkd adj)
MAXT = 0.01
CINT =0.1
TIMELIMIT =2184
EXPTIMEON =0.
EXPTIMEOFF =2180
DAY CYCLE = 24
WEEKPERIOD = 168
WEEKFINISH =119
BCKTIMEON =0.
(HOUR)
BCKTIMEOFF =0.
(HOUR)
BW T0 = 28 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT =0.0015 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.015 % EXPOSURE DOSE IN UG/KG
%MSTOT =0.150 % EXPOSURE DOSE IN UG/KG
MSTOT = 0.450 % EXPOSURE DOSE IN UG/KG
% SIMULATION TIME LIMIT (HOUR)
%TIME AT WHICH EXPOSURE BEGINS (HOUR)
%TIME AT WHICH EXPOSURE ENDS (HOUR)
%TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
C.2.5.2.4. Toth et al (1979) (1 year).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Toth et al 1979
%built and check in August 7 2009
%protocol: oral exposure single dose
%dose levels: 7, 700, 7000 ng/kg 1/week for 52 weeks (1 year)
%dose levels: 0.007, 0.7 and 7 ug/kg 1/week for 52 weeks (1 year)
%dose equivalent: 1, 100, 1000 ng/kg/day
MAXT = 0.01
CINT =0.1
TIMELIMIT = 8736
EXP TIME ON = 0. %TIME AT WHICH EXPOSURE BEGINS (HOUR)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-97 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
EXPTIMEOFF = 8736
DAY CYCLE = 168
WEEKPERIOD = 8760
WEEKFINISH = 8760
BCKTIMEON = 0.
(HOUR)
BCKTIMEOFF = 0.
(HOUR)
BW T0 = 27 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.007 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.7 % EXPOSURE DOSE IN UG/KG
MSTOT = 7 % EXPOSURE DOSE IN UG/KG
C.2.5.2.5. Toth et al. (1979) (2 year).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CBNDLINGKG
% Toth et al 1979
%built and check in August 7 2009
%protocol: oral exposure single dose
%dose levels: 7, 700, 7000 ng/kg 1/week for 52 weeks (1 year)
%dose levels: 0.007, 0.7 and 7 ug/kg 1/week for 52 weeks (1 year)
%dose levels equivalent: 1, 100, 1000 ng/kg/day
MAXT = 0.01
CINT =0.1
TIMELIMIT = 15576 % WEEKLY GAVAGE FOR 1 YEAR; LIFETIME FOLLOW-UP
(AVG 424-649 DAYS); USED MAXIMUM OF 649 DAYS
EXP TIME ON = 0. %TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXP TIME OFF = 8736 %2208 %tIME AT WHICH EXPOSURE ENDS (HOUR)
DAYCYCLE = 168
WEEKPERIOD = 8760
WEEKFINISH = 8760
BCK TIME ON = 0. %TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
BCK TIME OFF = 0. %TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
BW T0 = 27 % BODY WEIGHT AT THE BEGINNING OF THE SIMULATION
(G)
%2208 %TIME AT WHICH EXPOSURE ENDS (HOUR)
%TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-98 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.007 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.7 % EXPOSURE DOSE IN UG/KG
MSTOT = 7 % EXPOSURE DOSE IN UG/KG
C.2.5.2.6. White et al (1986).
output @clear
prepare @clear
prepare T CLINGKG CFNGKG CB SNGKGLIADJ BBNGKG CBNDLINGKG
% White et al 1986
%built and check in August 7 2009
%protocol: oral exposure single dose
%dose levels: 0.714, 3.57, 7.14, 35.71, 71.43, 142.86 ng/kg/d ug/kg 1/day for 14 consecutive
days
%dose have been modified following Jeff email on Friday August 21 2009
%dose levels: 10, 50, 100, 500, 1000, 2000 ng /kg/d ug/kg 1/day for 14 consecutive days
%dose levels: 0.010, 0.050, 0.100, 0.500, 1.0, 2.0 ug /kg/d ug/kg 1/day for 14 consecutive days
MAXT = 0.01
CINT =0.1
TIMELIMIT =336
EXPTIMEON = 0.
EXPTIMEOFF = 336
DAY CYCLE = 24
WEEKPERIOD =336
WEEKFINISH =336
B CKTIMEON = 0.
(HOUR)
BCK TIME OFF = 0.
%TIME AT WHICH EXPOSURE BEGINS (HOUR)
%TIME AT WHICH EXPOSURE ENDS (HOUR)
%TIME AT WHICH BACKGROUND EXPOSURE BEGINS
%TIME AT WHICH BACKGROUND EXPOSURE ENDS (HOUR)
BW TO
= 23
% BODY WEIGHT AT THE BEGINNING OF THE SIMULATION (G)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.010 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.050 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.100 % EXPOSURE DOSE IN UG/KG
%MSTOT = 0.500 % EXPOSURE DOSE IN UG/KG
%MSTOT = 1 % EXPOSURE DOSE IN UG/KG
MSTOT = 2 % EXPOSURE DOSE IN UG/KG
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-99 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
C.2.6. Mouse Gestational Model
C.2.6.1. Model Code
PROGRAM: 'Three Compartment PBPK Model for TCDD in Mice (Gestation)'
! Parameters were change may 16, 2002
! Come from { 8M AICHRPRE-EXPGD}
! Come from {12_Mouse_GD}file
! { {IMPORTANT-IMPORTANT-IMPORTANT-IMPORTANT} }
! REDUCTION OF MOTHER AND FETUS COMPARTMENT
! 2M_R_TCDDJUL Y2002 ////(JULY 18,2002)////
! TCDD_RED_4Species_2003_4 ////(APR 8 ,2003)////
! TCDD_RED_4Species_2003_9 ////(APR 17,2003)////
!TCDD_RED_4Species_2003_12 ////(APR 17,2003)////
! APRIL 18 2003
! TCDD_4C_4SP_2003 ////(APR 18,2003)////
! was "Gest 4 species 1 .csl" but update July 2009
!DevTCDD4Species_ICF_afterKKfix_v3_ratgest.csl
!MICE_GESTATIONAL_ICF_F092309.csl
! MICEGEST ATIONALICFF100609.csl
ILegend/Legend/Legend/Legend/Legend/Legend/Legend/Legend/
!Legend for this PBPK model
IMating: control the tenure of exchange between fetus and
IMother and also control imitated tissue growth
! Ctrl: WTFE, WFO, WPLA0, QPLAF,WT0
!(for rat, mouse, human, and monkey)
! Control transfer from mother to fetus and fetus to mother by TRANSTIME ON
! SWITCHtrans = 0 NO TRANSFER
! SWITCHtrans = 1 TRANSFER OCCURS
IGestoff = 1
!Gest_on= 0.
! These switches are also controlled by mating parameters
INITIAL !
! SIMULATION PARAMETERS ====
CONSTANT PARA ZERO = 1E-30
CONSTANT EXP TIME ON =288. ! TIME AT WHICH EXPOSURE BEGINS
(HOURS)
CONSTANT EXP TIME OFF = 504 ! TIME AT WHICH EXPOSURE ENDS (HOURS)
CONSTANT DAY CYCLE = 504. ! NUMBER OF HOURS BETWEEN DOSES
(HOURS)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-100 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
CONSTANT BCK TIME ON = 0.0
BEGINS (HOURS)
CONSTANT BCKTIMEOFF = 0.0
ENDS (HOURS)
CONSTANT TRANSTIME ON = 144
FETUS AT GESTATIONAL DAY 6
!UNIT CONVERSION
CONSTANT MW=322 ! MOLECULAR WEIGHT (NG/NMOL)
CONSTANT SERBLO = 0.55
CONSTANT UNITCORR = 1000
! INTRAVENOUS SEQUENCY
constant IVLACK =0.0
constant IVPERIOD =0.0
! PREGNANCY PARAMETER ====
CONSTANT MATTING = 0.0 BEGINNING OF MATING (HOUR)
CONSTANT N FETUS =10 INUMBER OF FETUS PRESENT
! CONSTANT EXPOSURE CONTROL ===========
! ACUTE, SUBCHRONIC, CHRONIC EXPOSURE =====
!OR BACKGROUND EXPOSURE (IN THIS CASE 3 TIMES A DAY)===
CONSTANT MSTOTBCKGR =0.0 ! ORAL BACKGROUND EXPOSURE DOSE
(UG/KG)
CONSTANT MSTOT = 0.0 ! ORAL EXPOSURE DOSE (UG/KG)
!ORAL ABSORPTION
MSTOT NM = MSTOT/MW ! CONVERTS THE DOSE TO NMOL/G
! INTRAVENOUS ABSORPTION
CONSTANT DOSEIV = 0.0 ! INJECTED DOSE (UG/KG)
DOSEIV NM = DOSEIV/MW ! CONVERTS THE INJECTED DOSE TO NMOL/G
CONSTANT DOSEIVLATE = 0.0 ! INJECTED DOSE LATE (UG/KG)
DOSEIVNMlate = DOSEIVLATE/MW ! AMOUNT IN NMOL/G
!INITIAL GUESS OF THE FREE CONCENTRATION IN THE LIGAND
(COMPARTMENT INDICATED BELOW)====
CONSTANT CFLLI0 =0.0 ILIVER (NMOL/ML)
CONSTANT CFLPLA0 =0.0 !PLACENTA (NMOL/ML)
!BINDING CAPACITY (AhR) FOR NON LINEAR BINDING (COMPARTMENT
INDICATED BELOW) (NMOL/ML) ===
CONSTANT LIBMAX = 3.5E-4 ! LIVER (NMOL/ML), WANG ET AL. 1997
CONSTANT PLABMAX = 2.0E-4 ! TEMPORARY PARAMETER
! TIME AT WHICH BACKGROUND EXPOSURE
! TIME AT WHICH BACKGROUND EXPOSURE
! CONTROL TRANSFER FROM MOTHER TO
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-101 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! PROTEIN AFFINITY CONSTANTS (1A2 OR AhR, COMPARTMENT INDICATED
BELOW) (NMOL/ML)===
CONSTANT KDLI = 1 .OE-4 ILIVER (AhR) (NMOL/ML), WANG ET AL. 1997
CONSTANT KDLI2 = 4.0E-2 ILIVER (1A2) (NMOL/ML), EMOND ET AL. 2004
CONSTANT KDPLA = 1.OE-4 ! TEMPORARY PARAMETER (AhR)
!EXCRETION AND ABSORPTION CONSTANT
CONSTANT KST = 0.3 ! GASTRIC RATE CONSTANT (HR-1)
CONSTANT KABS = 0.48 !INTESTINAL ABSORPTION CONSTANT (HR-1) ),
WANGET AL. 1997
! ELIMINATION CONSTANTS
CONSTANT CLURI = 0.09 ! URINARY CLEARANCE (ML/HR)
!TEST ELIMINATION VARIABLE
constant kelv = 0.4 ! INTERSPECIES VARIABLE ELIMINATION CONSTANT
(1/HOUR)
! CONSTANT TO DIVIDE THE ABSORPTION INTO LYMPHATIC AND PORTAL
FRACTIONS
CONSTANT A =0.7 ! LYMPHATIC FRACTION, WANG ET AL. 1997
!PARTITION COEFFICIENTS
CONSTANT PF = 400 ! ADIPOSE TISSUE/BLOOD
CONSTANT PRE = 3 ! REST OF THE BODY/BLOOD, WANG ET AL. 2000
CONSTANT PLI = 6 ! LIVER/BLOOD, WANG ET AL. 1997
CONSTANT PPLA = 3 ! TEMPORARY PARAMETER NOT CONFIGURED
!PARAMETER FOR INDUCTION OF CYP 1A2, WANG ET AL. 1997 OR OPTIMIZED
CONSTANT PAS INDUC = 1 ! INCLUDE INDUCTION? (1 = YES, 0 = NO)
CONSTANT CYP 1 A2_l OUTZ =1.6 ! DEGRADATION CONCENTRATION
CONSTANT OF 1A2 (NMOL/ML) (OPTIMIZED)
CONSTANT CYP1A21A1 =1.5 ! BASAL CONCENTRATION OF 1A1 (NMOL/ML),
WANG ET AL . (2000)
CONSTANT CYP1A21EC50 =0.13 ! DISSOCIATION CONSTANT TCDD-CYP1A2
(NMOL/ML)
CONSTANT CYP1A21A2 =1.5 !BASAL CONCENTRATION OF 1A2
(NMOL/ML),WANG ET AL. (2000)
CONSTANT CYP1A21KOUT =0.1 ! FIRST ORDER RATE OF DEGRADATION (H-l)
CONSTANT CYP 1 A2_l TAU =1.5 !HOLDING TIME (H) (OPTIMIZED), WANG ET
AL . (2000)
CONSTANT CYP 1 A2_ 1 EM AX = 600 ! MAXIMUM INDUCTION OVER BASAL
EFFECT (UNITLESS)
CONSTANT HILL = 0.6 !HILL CONSTANT; COOPERATIVELY LIGAND
BINDING EFFECT CONSTANT (UNITLESS)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-102 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
!DIFFUSIONAL PERMEABILITY FRACTION, WANG ET AL. 1997
CONSTANT PAFF
2000
CONSTANT PAREF
CONSTANT PALIF
CONSTANT PAPLAF
= 0.12 ! ADIPOSE (UNTITLESS) OPTIMIZED, WANG ET AL.
= 0.03 !REST OF THE BODY (UNITLESS)
= 0.35 ILIVER (UNITLESS)
= 0.03 ! TEMPORARY PARAMETER NOT CONFIGURED
!FRACTION OF TISSUE WEIGHT =========
CONSTANT WLI0 =0.0549 ILIVER ILSI (1994)
!TISSUE BLOOD FLOW EXPRESSED AS A FRACTION OF CARDIAC OUTPUT
CONSTANT QFF = 0.070 ! ADIPOSE TISSUE BLOOD FLOW FRACTION
(UNITLESS), LEUNG ET AL. 1990
CONSTANT QLIF = 0.161 ILIVER (UNITLESS), ILSI 1994
!COMPARTMENT TISSUE BLOOD EXPRESSED AS A FRACTION OF THE TOTAL
COMPARTMENT VOLUME
CONSTANT WFB0 =0.050 ! ADIPOSE TISSUE, WANG ET AL. 1997
CONSTANT WREB0 = 0.030 !REST OF THE BODY, WANG ET AL. 1997
CONSTANT WLIB0 = 0.266 ILIVER, WANG ET AL. 1997
CONSTANT WPLAB0 = 0.500 ! TEMPORARY PARAMETER NOT CONFIGURED
! EXPO SURE SCENARIO FOR UNIQUE OR REPETITIVE WEEKLY OR MONTHLY
EXPOSURE
INUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK = 0.0 IDELAY BEFORE EXPOSURE ENDS (WEEK)
CONSTANT WEEK PERIOD =168 ! NUMBER OF HOURS IN THE WEEK (HOURS)
CONSTANT WEEK FINISH =168 ! TIME EXPOSURE ENDS (HOURS)
INUMBER OF EXPOSURES PER MONTH
CONSTANT MONTH LACK =0.0 IDELAY BEFORE EXPOSURE BEGINS
(MONTH)
I CONSTANT FOR BACKGROUND EXPOSURE===========
CONSTANT Day LACK BG = 0.0 I DELAY BEFORE EXPOSURE BEGINS (HOUR)
CONSTANT DayPERIODBG = 24 I LENGTH OF EXPOSURE (HOUR)
INUMBER OF EXPOSURES PER WEEK
CONSTANT WEEK LACK BG = 0.0
(WEEK)
CONSTANT WEEK PERIOD BG =168
(HOURS)
CONSTANT WEEK FINISH BG =168
I INITIAL BODY WEIGHT
CONSTANT BW TO =30
IDELAY BEFORE BACKGROUD EXPOSURE
I NUMBER OF HOURS IN THE WEEK
I TIME EXPOSURE ENDS (HOURS)
I WANGET AL. 1997
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-103 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
CONSTANT RATIORATFMOUSEF = 0.2
GESTATIONAL DAY 22
! FOR RAT (1) AND FOR MOUSE (0.2)
! RATIO OF FETUS MOUSE/RAT AT
!COMPARTMENT LIPID EXPRESSED AS THE FRACTION OF TOTAL LIPID, POULIN
ET AL. 2002
CONSTANT F TOTLIP = 0.855
CONSTANT B TOTLIP = 0.0033
CONSTANT RE TOTLIP =0.019
CONSTANT LI TOTLIP = 0.060
CONSTANT PLA TOTLIP =0.019
CONSTANT FETUS TOTLIP =0.019
! ADIPOSE TISSUE (UNITLESS)
! BLOOD (UNITLESS)
! REST OF THE BODY (UNITLESS)
! LIVER (UNITLESS)
! PLACENTA (UNITLESS)
! FETUS (UNITLESS)
END ! END OF THE INITIAL SECTION
DYNAMIC ! DYNAMIC SIMULATION SECTION
ALGORITHM IALG = 2 ! GEAR METHOD
CINTERVAL CINT = 0.1 ! COMMUNICATION INTERVAL
MAXTERVAL MAXT = 1.0e+10 ! MAXIMUM CALCULATION INTERVAL
MINTERVAL MINT
VARIABLE T = 0.0
CONSTANT TIMELIMIT =
CINTXY = CINT
PFUNC = CINT
1 0E-10 ! MINIMUM CALCULATION INTERVAL
313 ! SIMULATION LIMIT TIME (HOUR)
!TIME CONVERSION
DAY = T/24 ! TIME IN DAYS
WEEK = T/168 ! TIME IN WEEKS
MONTH = T/730 ! TIME IN MONTHS
YEAR = T/8760 ! TIME IN YEARS
DERIVATIVE ! PORTION OF CODE THAT SOLVES DIFFERENTIAL EQUATIONS
!CHRONIC OR SUBCHRONIC EXPOSURE SCENARIO =======
INUMBER OF EXPOSURES PER DAY
DAY LACK = EXP TIME ON ! DELAY BEFORE EXPOSURE BEGINS (HOURS)
DAY PERIOD = DAY CYCLE ! EXPOSURE PERIOD (HOURS)
DAY FINISH = CINTXY ! LENGTH OF EXPOSURE (HOURS)
MONTH PERIOD = TIMELIMIT ! EXPOSURE PERIOD (MONTHS)
MONTH FINISH = EXP TIME OFF ! LENGTH OF EXPOSURE (MONTHS)
INUMBER OF EXPOSURES PER DAY AND MONTH
DAYFINISHBG = CINTXY
MONTH LACK BG = BCK TIME ON IDELAY BEFORE BACKGROUD EXPOSURE
BEGINS (MONTHS)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-104 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
MONTH PERIOD BG = TIMELIMIT BACKGROUND EXPOSURE PERIOD
(MONTHS)
MONTHFINISHBG = BCKTIMEOFF !LENGTH OF BACKGROUND EXPOSURE
(MONTHS)
! INTRAVENOUS LATE
IVFINISH = CINTXY
B = 1-A ! FRACTION OF DIOXIN ABSORBED IN THE PORTAL FRACTION OF THE
LIVER
IFETUS,VOLUME,FETUS,VOLUME,FETUS,VOLUME,FETUS,VOLUME,FETUS,VOLUM
E,FETUS,VOLUME
! FROM OFLAHERTY l 992
RTESTGEST= T-MATTING
TESTGEST=DIM(RTESTGEST,0.0)
WTFER_RODENT= (2.3d-3*EXP(1.49d-2*(TESTGEST))+1.3d-2)*Gest_on
WTFER = (WTFER_RODENT*RATIO_RATF_MOU SEF *N_FETU S)
WTFE = DIM(WTFER, 0.0)
FAT,VOLUME,FAT,VOLUME,FAT,VOLUME,FAT,VOLUME,FAT,VOLUME,F AT,VOLU
ME,FAT,VOLUME
! FAT GROWTH EXPRESSION LINEAR DURING PREGNANCY
! FROM O'FLAHERTYl 992
WF0= (((9.66d-5*(TESTGEST))*gest_on)+0.069)
! PLACENTA, VOLUME, PLACENTA, VOLUME, PLACENTA,VOLUME,
PLACENTA,VOLUME
! WPLA PLACENTA GROWTH EXPRESSION, SINGLE EXPONENTIAL WITH OFFSET
! FROM O'FLAHERTY l 992 ! FOR EACH PUP
WPLAONRODENT = (0.6/(l+(5d+3*EXP(-0.0225*(TESTGEST)))))*N_FETUS
WPLA0R = (WPL A0NRODENT/WT0) * Geston
WPLA0 = DIM(WPL A0R, 0.0)
! PLACENTA,FLOW RATE, PLACENTA,FLOW RATE, PLACENTA,FLOW RATE,
PLACENTA,FLOW RATE
! QPLA PLACENTA GROWTH EXPRESSION, DOUBLE EXPONENTIAL WITH OFFSET
! FROM O'FLAHERTY l 992
QPLARF = (1.67d-7 *exp(9.6d-3*(TESTGEST)) &
+1.6d-3*exp(7.9d-3*(TESTGEST))+0.0)*Gest_on*SWITCH_trans
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-105 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
QPL AF=DIM(QPL ARF ,0.0)
! FRACTION OF FLOW RATE IN PLACENTA
! GESTATION CONTROL
IF (T.LT.MATTING) THEN
Gestoff = 1
Gest_on= 0.0
ELSE
Gestoff = 0.0
Geston = 1
END IF
! MOTHER BODY WEIGHT GROWTH EQUATION========
! MODIFICATION TO ADAPT THIS MODEL AT HUMAN MODEL
! BECAUSE LINEAR DESCRIPTION IS NOT GOOD ENOUGH FOR MOTHER GROWTH
! MOTHER BODY WEIGHT GROWTH
PARAMETER (BW RMN = 1.0E-30)
WT0= BW T0 *(1.0+(0.41*T)/(1402.5+T+BW_RMN))
! VARIABILITY OF REST OF THE BODY DEPENDS ON OTHER ORGANS
WRE0 = (0.91 - (WLIB0*WLI0 + WFB0*WF0 +WPLAB0*WPLA0 + WLI0 + WF0 +
WPLAO))/(1.0+WREB0) ! REST OF THE BODY FRACTION; UPDATED FOR EPA
ASSESSMENT
QREF = 1.0-(QFF+QLIF+QPLAF) !REST OF BODY BLOOD FLOW RATE (ML/HR)
QTTQF = QFF+QREF+QLIF+QPLAF ! SUM MUST EQUAL 1
! COMPARTMENT VOLUME (ML OR G) =========
WF = WF0 * WTO ! ADIPOSE TISSUE
WRE = WRE0 * WTO ! REST OF THE BODY
WLI = WLI0 * WTO ! LIVER
WPLA= WPL AO* WTO ! PLACENTA
! COMPARTMENT TISSUE BLOOD (ML OR G) =======
WFB = WFB0 * WF ! ADIPOSE TISSUE
WREB = WREB0 * WRE ! REST OF THE BODY
WLIB = WLIB0 * WLI ! LIVER
WPLAB = WPLAB0* WPLA ! PLACANTA
! CARDIAC OUTPUT FOR THE GIVEN BODY WEIGHT
!QC= QCCAR*60*(WT0/1000.0)* *0.75
CONSTANT QCC=16500 ! EQUIVALENT TO 275 * 60
QC= QCC*(WT0/UNITCORR)**0.75
! COMPARTMENT BLOOD FLOW RATE (ML/HR)
QF = QFF*QC IADIPOSE TISSUE BLOOD FLOW RATE
QLI = QLIF*QC !LIVER TISSUE BLOOD FLOW RATE
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-106 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
QRE = QREF*QC
QPLA = QPLAF*QC
!REST OF THE BODY BLOOD FLOW RATE
!PLACENTA TISSUE BLOOD FLOW RATE
QTTQ = QF+QRE+QLI+QPLA ! TOTAL FLOW RATE
!PERMEABILITY ORGAN FLOW (ML/HR)
PAF =PAFF*QF
PARE = PAREF*QRE
PALI = PALIF*QLI
PAPLA = PAPLAF*QPLA
! ADIPOSE TISSUE
! REST OF THE BODY
! LIVER TISSUE
! PLACENTA
ABSORPTION SECTION
ORAL,
INTRAPERITONEAL,
INTRAVENOUS
!REPETITIVE ORAL BACKGROUND EXPOSURE SCENARIO
MSTOT NMBCKGR = MSTOTBCKGR/322 ! AMOUNT IN NMOL/G
MSTTBCKGR =MSTOT_NMBCKGR *WT0
DAYEXPOSUREBG = PULSE(DAY_LACK_BG,DAY_PERIOD_BG,DAY_FINISH_BG)
WEEKEXPOSUREBG =
PUL SE(WEEK_L ACKB G, WEEKPERIODB G, WEEKFINISHB G)
MONTHEXPO SUREB G =
PUL SE(MONTH_L ACKB G,MONTH_PERIOD_B G,MONTH_FINI SH_B G)
MSTTCHBG =
(D AY EXPO SURE B G* WEEKEXPO SUREB G*MONTH_EXPO SUREB G)*MSTTBCK
GR
MSTTFRBG = MSTTBCKGR/CINT
C YCLE BG =D AY EXPO SURE BG* WEEK EXPO SURE B G*MONTH_EXPO SURE B G
! CONDITIONAL ORAL EXPOSURE (BACKGROUND EXPOSURE)
IF (MSTTCH BG.EQ.MSTTBCKGR) THEN
ABSMSTT_GB= MSTTFRBG
ELSE
ABSMSTTGB = 0.0
END IF
C YCLETOTB G=INTEG(C Y CLEB G, 0.0)
!REPETITIVE ORAL EXPOSURE SCENARIO
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-107 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
MSTT= MSTOT NM * WTO ! AMOUNT IN NMOL
DAYEXPOSURE = PULSE(DAY_LACK,DAY_PERIOD,DAY_FINISH)
WEEKEXPOSURE = PULSE(WEEK_LACK,WEEK_PERIOD,WEEK_FINISH)
MONTHEXPOSURE = PUL SE(MONTH_L ACK,MONTH_PERIOD ,MONTH_FINI SH)
MSTTCH = (D AYEXPO SURE* WEEKEXPO SURE*MONTH_EXPO SURE) *MSTT
MSTTFR = MSTT/CINT
CYCLE = DAYEXPO SURE* WEEKEXPO SURE*MONTH_EXPO SURE
SUMEXPEVENT= INTEG (CYCLE,0.0)/cint INUMBER OF CYCLES GENERATED
DURING SIMULATION
! CONDITIONAL ORAL EXPOSURE
IF (MSTTCH.EQ.MSTT) THEN
ABSMSTT= MSTTFR
ELSE
ABSMSTT = 0.0
END IF
C Y CLETOT=INTEG(C Y CLE, 0.0)
! MASS CHANGE IN THE LUMEN
RMSTT= -(KST+KABS)*MST + ABSMSTT +ABSMSTT GB ! RATE OF CHANGE
(NMOL/H)
MST = INTEG(RMSTT,0.0) ! AMOUNT REMAINING IN DUODENUM
(NMOL)
! ABSORPTION IN LYMPH CIRCULATION
LYRMLUM = KABS*MST*A
LYMLUM = INTEG(L YRMLUM, 0.0)
! ABSORPTION IN PORTAL CIRCULATION
LIRMLUM = KABS*MST*B
LIMLUM = INTEG(LIRMLUM, 0.0)
! -—IV EXPOSURE
IV= DOSEIV NM * WTO ! AMOUNT IN NMOL
IVR= IV/PFUNC ! RATE FOR IV INFUSION IN BLOOD
EXPIV= IVR * (1 O-STEP(PFUNC))
IVDOSE = integ(EXPIV,0.0)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-108 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
I iy late in the cycle
! MODIFICATION ON January 13 2004
IVRlateR = DOSEIVNMlate*WTO
IV_EXPOSURE=PULSE(IV_LACK,IV_PERIOD,IV_FINISH)
IV lateT = IVEXPOSURE *IV_RlateR
I V_1 ate = IVlateT/CINT
SUMEXPEVENTIV= integ (IV_EXPOSURE,0.0) ! NUMBER OF CYCLE GENERATE
DURING SIMULATION
! SYSTEMIC CONCENTRATION OF TCDD
! MODIFICATION ON OCTOBER 6, 2009
CB=(QF*CFB+QRE*CREB+QLI*CLIB+EXPIV+LYRMLUM+QPLA*CPLAB+IV_late)/(QC
+CLURI) !
CA = CB ! CONCENTRATION (NMOL/ML)
! URINARY EXCRETION BY KIDNEY
!MODIFICATION ON OCTOBER 6, 2009
RAURI = CLURI *CB
AURI = INTEG(R AURI, 0.0)
!UNIT CONVERSION POST SIMULATION
CB SNGKGLIAD J=(CB *MW*UNITCORR* (1 /B TOTLIP) * (1 /SERBLO))! [NG of TCDD
Serum/Kg OF LIPIP]
AU CB S_N GKGLI AD J=integ(CB SN GKGLI AD J, 0.0)
PRCT B = (CB/(MSTT+1E-30))*100 ! PERCENT OF ORAL DOSE IN BLOOD
PRCT BIV = (CB/(IV_RlateR+1E-30))* 100 ! PERCENT OF IV DOSE IN BLOOD
CBNGKG= CB*MW*UNITCORR
CBNGG = CB*MW
! ADIPOSE COMPARTMENT
!TISSUE BLOOD COMPARTMENT
RAFB= QF * (C A-CFB)-PAF * (CFB-CF/PF) !(NMOL/H)
AFB = INTEG(RAFB ,0.0) ! (NMOL)
CFB = AFB/WFB ! (NMOL/ML)
!TISSUE COMPARTMENT
RAF = PAF*(CFB-CF/PF) !(NMOL/H)
AF = INTEG(RAF,0.0) !(NMOL)
CF = AF/WF ! (NMOL/ML)
!UNIT CONVERSION POST SIMULATION
CFTOTAL= (AF + AFB)/(WF + WFB) ! TOTAL CONCENTRATION IN NMOL/ML
CFTFREE = CFB + CF ! TOTAL FREE CONCENTRATION IN FAT (NM/ML)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-109 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
PRCTF = (CFTOTAL/(MSTT+1E-30))*100 ! PERCENT OF ORAL DOSE IN FAT
PRCT FIV = (CFTOTAL/(IV_RlateR+lE-30))*100 ! PERCENT OF IV DOSE IN FAT
CFNGKG=CFTOTAL*MW*UNITCORR ! FAT CONCENTRATION IN NG/KG
AUCF_NGKGH=integ(CFNGKG,0.0)
CFNGG = CFTOTAL*MW
!REST OF THE BODY COMPARTMENT
RAREB= QRE * (C A-CREB)-P ARE* (CREB-CRE/PRE) !(NMOL/H)
AREB = INTEG(RAREB ,0.0) ! (NMOL)
CREB = AREB/WREB !(NMOL/H)
!TISSUE COMPARTMENT
RARE = PARE* (CREB - CRE/PRE) ! (NMOL/H)
ARE = INTEG(RARE,0.0) ! (NMOL)
CRE = A RE/W RE ! (NMOL/ML)
!UNIT CONVERSION POST SIMULATION
CRETOTAL= (ARE + AREB)/(WRE + WREB) ! TOTAL CONCENTRATION IN
NMOL/ML
PRCT RE = (CRETOTAL/(MSTT +1E-30))* 100 ! PERCENT OF ORAL DOSE IN REST OF
BODY
PRCT REIV = (CRETOTAL/(IV_R1 ateR+ 1E-30))*100 ! [ PERCENT OF IV DOSE IN REST
OF THE BODY ]
CRENGKG=CRETOTAL*MW*UNITCORR ! REST OF THE BODY CONCENTRATION
IN NG/KG
! LIVER COMPARTMENT
!TISSUE BLOOD COMPARTMENT
RALIB = QLI * (CA-CLIB )-PALI * (CLIB -CFLLIR)+LIRMLUM !
ALIB = INTEG(RALIB ,0.0) !(NMOL)
CLIB = ALIB/WLIB ! (NMOL/ML)
!TISSUE COMPARTMENT
RALI = PALI*(CLIB - CFLLIR)-REXCLI ! (NMOL/HR)
ALI = INTEG(RALI, 0.0) ! (NMOL)
CLI = ALI/WLI ! (NMOL/ML)
!FREE TCDD IN LIVER COMPARTMENT
PARAMETER (LIVER IRMN = 1.0E-30)
CFLLI= IMPLC(CLI-(CFLLIR*PLI+(LIBMAX*CFLLIR/(KDLI+CFLLIR &
+LIVER_1RMN))+((CYP 1 A2_l 03 *CFLLIR/(KDLI2 + CFLLIR &
+LIVER1RMN) *P ASINDU C)))-CFLLI,CFLLI0)
CFLLIR=DIM(CFLLI,0.0) ! FREE CONCENTRATION IN LIVER
CBNDLI= LIBM AX* CFLLIR/(KDLI+CFLLIR+LIVER_1 RMN) IBOUND
CONCENTRATION
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! VARIABLE ELIMINATION BASED ON THE CYP1A2
KBILELIT =((CYP1 A2_l OUT-CYP1 A2_l A2)/CYP1 A2_l A2)*Kelv ! INDUCED
BILIARY EXCRETION RATE CONSTANT
REXCLI = KBILELIT * CFLLIR* WLI ! DOSE-DEPENDENT EXCRETION RATE
EXCLI = INTEG(REXCLI, 0.0)
!UNIT CONVERSION POST SIMULATION
CLITOTAL= (ALI + ALIB)/(WLI + WLIB) ! TOTAL CONCENTRATION IN NMOL/ML
PRCT LI = (CLITOTAL/(MSTT+1E-30))* 100 ! PERCENT ORAL DOSE IN LIVER
PRCT LIIV = (CLITOTAL/(IV_RlateR+lE-30))*100 ! PERCENT IV DOSE IN LIVER
Rec_occ= CFLLIR/(KDLI+CFLLIR)
CLINGKG=CLITOTAL*MW*UNITCORR ! LIVER CONCENTRATION IN NG/KG
AU CLIN GKGH=INTEG(CLIN GKG, 0.0)
CBNDLINGKG = CBNDLI*MW*UMTCORR
AU CBNDLIN GKGH =INTEG(CBNDLIN GKG, 0.0)
CLINGG = CLITOTAL*MW
! CHEMICAL IN CYP450 (1A2) COMPARTMENT
CYP1A21KINP = CYP 1 A2_ 1KOUT* CYP 1 A2_ 1 OUTZ ! BASAL RATE OF CYP1A2
PRODUCTION SET EQUAL TO BASAL RATE OF DEGREDATION
! MODIFICATION ON OCTOBER 6, 2009
CYP1A210UT =INTEG(CYP1A2_1KINP * (1.0 + CYP1A21EMAX *(CBNDLI+1.0e-
30)**HILL &
/(CYP 1 A2_ 1 EC5 0 * *HILL + (CBNDLI+1.0e-30)**HILL)) &
- C YP 1 A2_ 1K OUT * C YP 1 A2_ 1 OUT, CYP1A210UTZ)
! EQUATIONS INCORPORATING DELAY OF CYP1A2 PRODUCTION (NOT USED IN
SIMULATIONS)
CYP1A21R02 = (CYP1A210UT - CYP1A2102)/ CYP1A21TAU
CYP1A2102 =INTEG(CYP 1 A2_ 1R02, CYP1A21A1)
CYP1A21R03 = (CYP1A2102 - CYP1A2103)/ CYP1A21TAU
CYP1A2103 =INTEG(CYP 1 A2_ 1R03, CYP1A21A2)
! TRANSFER OF DIOXIN FROM PLACENTA TO FETUS
! FETAL EXPOSURE ONLY DURING EXPOSURE
IF (T.LT. TRANSTIME ON) THEN
SWITCHtrans = 0.0
ELSE
SWITCHtrans = 1
END IF
! TRANSFER OF DIOXIN FROM PLACENTA TO FETUS
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 11 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
! MODIFICATION 26 SEPTEMBER 2003
CONSTANT PFETUS= 4 !
CONSTANT CLPLAFET = 0.17 !
RAMPF = (CLPL AFET * CPLA) *SWITCH_trans
AMPF=INTEG(RAMPF ,0.0)
! TRANSFER OF DIOXIN FROM FETUS TO PLACENTA
RAFPM = (CLPL A FET * CFETUS_v) * S WITCH trans !
AFPM = INTEG(RAFPM, 0.0)
! TCDD IN PLACENTA MOTHER COMPARTMENT
RAPLAB= QPLA*(CA - CPLAB)-PAPLA*(CPLAB -CFLPLAR) ! NMOL/H)
APLAB = INTEG(RAPLAB ,0.0) ! (NMOL)
CPLAB = APLAB/(WPLAB+1E-3 0) ! (NMOL/ML)
RAPLA = PAPLA*(CPLAB-CFLPLAR)-RAMPF + RAFPM ! (NMOL/H)
APLA = INTEG(RAPLA, 0.0) ! (NMOL)
CPL A = APLA/(WPLA+le-30) ! (NMOL/ML)
PARAMETER (PARA ZERO = 1.0E-30)
CFLPLA= IMPLC(CPLA-(CFLPLAR*PPLA +(PL ABM AX* CFLPL AR/(KDPL A&
+CFLPLAR+PARA_ZERO)))-CFLPLA,CFLPLA0)
CFLPL AR=DIM(CFLPL A, 0.0)
!UNIT CONVERSION POST SIMULATION
CPLATOTAL= (APLA + APLAB)/((WPLA + WPLAB)+le-30)! TOTAL
CONCENTRATION IN NMOL/ML
PRCTPLA = (CPL ATOT AL/(MSTT +1 E-3 0)) * 100
PRCTPLAIV = (CPLATOTAL/(IV_RlateR+ 1E-30))* 100
CPLANGG = CPLATOTAL*MW
!FETUS COMPARTMENT
RAFETUS= RAMPF -RAFPM
AFETU S=INTEG(RAFETU S, 0.0)
CFETU S=AFETU S/(WTFE+1 E-3 0)
CFETOTAL= CFETUS
CFETUSv = CFETU S/PFETU S
! UNIT CONVERSION POST SIMULATION
CFETUSNGKG = CFETUS *MW*UNITCORR ! (NG/KG)
AUCFENGKGH = INTEG(CFETU SN GKG, 0.0)
PRCTFE = (CFETOTAL/(MSTT+1E-30))*100
PRCTFEIV = (CFETOT AL/(I V_R1 ateR+ 1E-30))*100
CFETUSNGG = CFETOTAL*MW
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
I CONTROL MASS BALANCE
BDOSE= IVDOSE +LYMLUM+LIMLUM
BMASSE = EXCLI+AURI+AFB+AF+AREB+ARE+ALIB+ALI+APL A+APL AB+AFETU S
BDIFF = BDO SE-BM AS SE
!BODY BURDEN (NG)
BODY BURDEN = AFB+AF+AREB+ARE+ALIB+ALI+APLA+APLAB !
BBFETUSNG = AFETUS*MW*UNITCORR ! NG
! BODY BURDEN IN TERMS OF CONCENTRATION (NG/KG)
BBNGKG
=(((AFB+AF+AREB+ARE+ALIB+ALI+APLA+APLAB)/WTO)*MW*UNITCORR) !
AU CBBN GKGH=INTEG(BBN GKG, 0.0)
I COMMAND OF THE END OF SIMULATION
TERMT (T.GE. TimeLimit, 'Time limit has been reached.')
END ! END OF THE DERIVATIVE SECTION
END ! END OF THE DYNAMIC SECTION
END ! END OF THE PROGRAM
C.2.6.2. Input Files
C.2.6.2.1. Keller et al. (2007).
%TO BE USED AFTER THE
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AU CLIN GKGH AUCFNGKGH AUCB SNGKGLIAD J AUCBBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLI NGKGH
%output @nciout=10 T SUMEXPEVENT wtO
%kELLER ET AL 2007
%protocol: SINGLE DO SE from GD13
%dose levels: 0.01, 0.100 1 ug/kg at GDI3
%dose levels: 10, 100 lOOOng/kg atGD13
%EXPOSURES SCENARIOS
MAXT=0.01
CINT =0.1
% TIME AT WHICH EXPOSURE BEGINS(HOUR)
% TIME AT WHICH EXPOSURE ENDS (HOUR)
% TIME AT WHICH BACKGROUND EXPOSURE BEGINS
% TIME AT WHICH BACKGROUND EXPOSURE ENDS
(HOUR)
EXPTIMEON =312.
EXPTIMEOFF =330
DAY CYCLE = 505
B CKTIMEON = 0.
(HOUR)
BCK TIME OFF =0.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
IVLACK
IV PERIOD
TIMELIMIT
BWTO
MATTING
= 505
= 505
= 504
= 24
= 0.
TRAN STIMEON = 144.
N FETUS = 10
% SIMULATION TIME LIMIT (HOUR)
% BEGINNING OF MATING (HOUR)
% SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
%EXPOSURE DOSE SCENARIOS (UG/KG)
%MSTOT = 0.01 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT =0.1 % ORAL EXPOSURE DOSE IN UG/KG
MSTOT = 1 % ORAL EXPOSURE DOSE IN UG/KG
C.2.6.2.2. Li et al. (2005).
%TO BE USED AFTER THE
%clear variable
output @clear
prepare @clear T CLINGKG CFNGKG CBSNGKGLIADJ BBNGKG CFETUSNGKG
AUCLI NGKGH AUCF NGKGH AUCB S NGKGLIAD J AUC BBNGKGH
AUC FENGKGH CBNDLINGKG AUCBNDLINGKGH
%output @nciout=10 T SUMEXPEVENT
%LI ET AL 2006
%protocol: exposure repetitive DOSE from GDI to GD3
%dose levels: 0.002, 0.050 AND 0.10 ug/kg/day at GDI TO GD8
%dose levels: 2, 50 and 100 ng/kg/day from GDI to GD8
%EXPOSURES SCENARIOS
MAXT=0.001
CINT =0.1
EXP TIME ON = 0. % TIME AT WHICH EXPOSURE BEGINS (HOUR)
EXP TIME OFF = 70 % TIME AT WHICH EXPOSURE ENDS (HOUR); 2 HOURS
LESS THAN GD8; SET EQUAL TO 70 TO BE SURE ONLY 3 DOSES ADMINISTERED
% BECAUSE i STARTED TIME 0 FOR GDI
DAY CYCLE = 24
BCK TIME ON = 0. % TIME AT WHICH BACKGROUND EXPOSURE BEGINS
(HOUR)
B CKTIMEOFF = 0.
(HOUR)
IVLACK = 505
IVPERIOD = 505
TIMELIMIT =216
BWTO = 27
MATTING = 0.
TRANSTIME ON = 144.
% TIME AT WHICH BACKGROUND EXPOSURE ENDS
% SIMULATION TIME LIMIT (HOUR)
% BEGINNING OF MATING (HOUR)
% SHOULD BE MATING TIME + 6 DAYS (144 HOURS)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
NFETUS = 10
%EXPOSURE DOSE SCENARIOS (UG/KG)
MSTOT = 0.002 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT =0.05 % ORAL EXPOSURE DOSE IN UG/KG
%MSTOT =0.10 % ORAL EXPOSURE DOSE IN UG/KG
C.3. TOXICOKINETIC MODELING RESULTS FOR KEY ANIMAL BIOASSAY
STUDIES
The simulated TCDD serum-adjusted lipid concentrations reported in this appendix for
the rodent bioassays were converted to TCDD concentrations in rodent whole blood. Initially,
EPA multiplied the serum-adjusted lipid concentrations by 0.0033, the ratio of lipid content to
total serum volume, then by 0.55, the value of the hematocrit. This product yields the TCDD
concentration in whole rodent blood as predicted by the PBPK model. EPA assumed that the
same whole blood TCDD concentration would result in the same effects in humans and rodents.
This conversion accomplishes the following:
1. Allows the human equivalent dose (HED) to be based on equivalent blood concentration
(that represents serum plus erythrocyte TCDD), which is proportional to tissue exposure;
2. Avoids criticism that the total blood concentration is normalized to serum lipid alone in
an unbalanced way (thus EPA does not contradict Centers for Disease Control and
Prevention (CDC) data or methods);
3. Factors out any impact of the lipid content used in the PBPK model; and
4. TCDD concentration in whole blood is encouraged for use in the assessments by the NAS
(NAS, 2006, p. 43); see additional information in Section 3.3.
C.3.1. Nongestational Studies
C.3.1.1. Cantoni et al. (1981)
Type:
Rat
Dose:
10, 100, 1000 ng/kg/week
Strain:
CD-COBS rats
Route:
Oral gavage
Body weight:
BW set to 125g
Regime:
1 dose/week for 45 weeks
Sex:
Female
Simulation
time:
7,584 hours
(45 weeks + 24 hours before sacrifice)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 15 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.43
Emond
1,018
2,040 (@ 7,392 hours)
982
CADM
-
14.29
Emond
4.XI.X
14.<>4l> ( a hum's)
4.242
CADM
142.86
Emond
2~.55l>
125. 'tin ( a ^>2 Ikimis)
2 1 .iw
-------
142.86
Emond
1,336
1,695 (@ 7,398 hours)
1,088
CADM
2,106
2,266
2,266
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.43
Emond
6.04
7.76 (@ 7,396 hours)
5.88
CADM
-
14.29
Emond
CADM
2 ' "
2') 1 ( a hum's)
:i (.
142.86
Emond
66.8
So o ( a 1 hum's)
(>2 (i
CADM
1 C.3.1.2. Chu et al. (2007)
Type:
Rat
Dose:
2.5, 25, 250, and 1,000 ng/kg-day
Strain:
Sprague-Dawley
Route:
Oral gavage
Body weight:
200 g
Regime:
1 dose per day for 28 days
Sex:
Female
Simulation time:
672 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.5
Emond
696
1,295 (@ 648 hours)
1,036
CADM
-
25
Emond
4,222
8,403 (@ 648 hours)
5,727
CADM
-
250
Emond
26,889
62,067 (@ 648 hours)
35,103
CADM
-
1,000
Emond
93,213
230,320 (@ 648 hours)
122,200
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 17 DRAFT—DO NOT CITE OR QUOTE
-------
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.5
Emond
148
268 (@ 652 hours)
255
CADM
-
-
-
25
Emond
1,777
2,953 (@ 653 hours)
2,806
CADM
-
-
-
250
Emond
19,232
30,262 (@ 653 hours)
28,668
CADM
-
-
-
1,000
Emond
77,819
120,400 (@ 653 hours)
113,890
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.5
Emond
108
180 (@ 668 hours)
180
CADM
-
-
-
25
Emond
660
1,020 (@ 659 hours)
1,015
CADM
-
-
-
250
Emond
4,210
6,433 (@ 655 hours)
6,354
CADM
-
-
-
1,000
Emond
14,576
22,610 (@ 655 hours)
22,280
CADM
-
-
-
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.5
Emond
16.1
27.5 (@ 652 hours)
26.9
CADM
-
-
-
25
Emond
138
222 (@ 652 hours)
214
CADM
-
-
-
250
Emond
1,239
1,935 (@ 652 hours)
1,842
CADM
-
-
-
1,000
Emond
4,801
7,444 (@ 652 hours)
7,067
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 18 DRAFT—DO NOT CITE OR QUOTE
-------
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.5
Emond
4.15
6.51 (@ 652 hours)
6.21
CADM
-
25
Emond
CADM
:u 5
2X 5 ( a (>52 hum's)
2" 4
250
Emond
CADM
in i
"(> ii ( a (>52 hum's)
"4 "
1,000
Emond
CADM
•jo 2
•w 0 ( a (>5 ' hoursi
<>X i
1 C.3.1.3. Crofton et al. (2005)
Type:
Rats
Dose:
0, 0.1, 3, 10, 30, 100, 300, 1000, 3000,
and 10,000 ng/kg-day
Strain:
Long Evans
Route:
Oral gavage
Body weight:
4 weeks old
BW set to 190 g
Regime:
One dose per day for four days
Sex:
Female
Simulation time:
96 hours
2
3 aThe CADM model was not ran because the dosing duration is lower than the resolution of the model (1 week)
4
5
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.1
Emond
11 1
22 4 72 hours')
l^i 5
CADM
-
3
Emond
2(
(><>5 ( a "2 Ikuiis)
321
CADM
.
10
Emond
(> i
1 ,X~ i i a "2 hours)
X<>2
CADM
30
Emond
1 .l>u5
5.2(12 (
CADM
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-l 19 DRAFT—DO NOT CITE OR QUOTE
-------
100
Emond
5,104
15,972 (@ 72 hours)
5,605
CADM
-
300
Emond
44.1>X2 ( a ~2 Ikuiis)
1 \5 ( a ~1 Ikuiis)
^".554
CADM
-
3000
Emond
•w.<45
42(1.X5<> ( a ~2 Ikuiis)
huxi.u
CADM
-
10,000
Emond
321,480
1,392,100 (@ 72
hours)
334,220
CADM
-
-
LIVER CONCENTRATIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
0.1
Emond
0.919
1.55 (@ 75 hours)
1.18
CADM
-
-
-
3
Emond
37.4
62.6 (@ 76 hours)
53.3
CADM
-
-
-
10
Emond
145
242 (@ 77 hours)
214
CADM
-
-
-
30
Emond
494
818 (@78 hours)
742
CADM
-
-
-
100
Emond
1,839
3,025 (@ 78 hours)
2,793
CADM
-
-
-
300
Emond
5,925
9,692 (@ 78 hours)
9,028
CADM
-
-
-
1000
Emond
20,717
33,738 (@ 79 hours)
31,564
CADM
-
-
-
3000
Emond
63,511
103,140 (@ 79 hours)
96,545
CADM
-
-
-
10,000
Emond
212,890
344,910 (@ 79 hours)
321,960
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-120 DRAFT—DO NOT CITE OR QUOTE
-------
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.1
Emond
1.00
1.93 (@ 96 hours)
1.93
CADM
-
-
-
3
Emond
24.6
45.9 (@ 96 hours)
45.9
CADM
-
-
-
10
Emond
70.3
129 (@ 96 hours)
129
CADM
-
-
-
30
Emond
177
317 (@ 96 hours)
317
CADM
-
-
-
100
Emond
480
838 (@ 96 hours)
838
CADM
-
-
-
300
Emond
1,206
2,065 (@ 96 hours)
2,065
CADM
-
-
-
1000
Emond
3,452
5,836 (@ 96 hours)
5,836
CADM
-
-
-
3000
Emond
9,522
16,050 (@ 96 hours)
16,050
CADM
-
-
-
10,000
Emond
30,657
51,918 (@ 96 hours)
51,918
CADM
-
-
-
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.1
Emond
0.138
0.224 (@ 79 hours)
0.223
CADM
-
-
-
3
Emond
4.04
6.56 (@ 78 hours)
6.44
CADM
-
-
-
10
Emond
13.3
21.5 (@ 78 hours)
21.0
CADM
-
-
-
30
Emond
39.3
63.5 (@ 78 hours)
61.5
CADM
-
-
-
100
Emond
129
208 (@ 78 hours)
200
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-121 DRAFT—DO NOT CITE OR QUOTE
-------
300
Emond
384
618 (@ 77 hours)
590
CADM
-
-
-
1000
Emond
1,270
2,041 (@ 77 hours)
1,942
CADM
-
-
-
3000
Emond
3,793
6,094 (@ 77 hours)
5,784
CADM
-
-
-
10,000
Emond
12,595
20,226 (@ 77 hours)
19,154
CADM
-
-
-
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
0.1
Emond
n
n 115 ((ft 75 hours')
n
CADM
PiPPPPPPiPPiiiiiP
-
3
Emond
:
2 4~ ( a "(> Ikuiis)
2
CADM
PPPPPPPPSPPPPPPP^
-
10
Emond
4
(> 42 ( a "(> Ikuiis)
5
CADM
-
30
Emond
10
14.1 (@ 76 hours)
12
CADM
-
100
Emond
22
29.9 (@ 76 hours)
27
CADM
PP?5?5^^^
PPP^PPPPPSPPPPPP^
-
300
Emond
41
51.9 (@ 77 hours)
49
CADM
PPPPPPPPPPPPPPPPB
-
1000
Emond
68
80.2 (@ 1 hours)
77
CADM
PPPPPPPPPPPPPPPPB
-
3000
Emond
•JO
'JS (I ( (1 1 IkUIIM
'X.
CADM
PPPPPPPPPPPPPPPPP::
-
10,000
Emond
104
los ( (1 1 IkUIIM
lo"
CADM
-
1
2
3 C.3.1.4. Fattore et al. (2000)
Type:
Rat
Dose:
20, 200, 2,000 ng/kg-day
Strain:
Sprague Dawley
Route:
Dietary
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-122 DRAFT—DO NOT CITE OR QUOTE
-------
Body weight:
7 weeks old (BW
150g)
Regime:
13 weeks
Sex:
Female and male
Simulation time:
2,184 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
20
Emond
5,282
8,259 (@ 2,160 hours)
6,135
CADM
-
200
Emond
31,761
56,170 (@ 2,160 hours)
35,183
CADM
-
2,000
Emond
262,030
497,250 (@ 2,160 hours)
287,690
CADM
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
20
Emond
2,448
3,228 (@ 2,164 hours)
3,078
CADM
4,471
5,639
5,639
200
Emond
24,136
30,245 (@ 2,164 hours)
28,709
CADM
45,337
56,499
56,499
2,000
Emond
234,170
288,020 (@ 2,164 hours)
272,590
CADM
454,031
565,103
565,103
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
20
Emond
890
1,113 (@ 2,166 hours)
1,101
CADM
1,545
1,796
1,756
200
Emond
5,355
6,542 (@ 2,165 hours)
6,430
CADM
13,351
15,604
15,292
2,000
Emond
44,176
54,246 (@ 2,165 hours)
53,140
CADM
131,259
153,534
150,516
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-123 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
20
Emond
187
242 (@ 2,164 hours)
233
CADM
261
324
324
200
Emond
1,556
1,940 (@ 2,164 hours)
1,850
CADM
2,496
3,084
3,084
2,000
Emond
14,432
17,797 (@ 2,164 hours)
16,891
CADM
24,836
30,674
30,674
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
20
Emond
24.9
29.8 (@ 2,164 hours)
28.8
CADM
-
200
Emond
<.<> 4
76.0 (@ 2,164 hours)
"4 "
CADM
-
2,000
Emond
It >4
I<>(. ( (i 2.1(4 hum's)
|0(>
CADM
-
1 C.3.1.5. Hassoun et al. (1998)
Type:
Mice
Dose:
0,0.45, 1.5, 15, 150 ng/kg-day.
Background exposure dose (default) = 0.05
ng/kg-day
Strain:
B6C3F1
Route:
Oral gavage
Body weight:
8 to 9 weeks old
(BW set to 23g)
Regime:
5 days/week for 13 weeks
Sex:
Female
Simulation
time:
2208 hours* (2,184h + 24h post exposure)
aNo background has been considered here for this simulation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-124 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.321
Emond
90.5
167 (@ 2,112 hours)
123
CADM
-
1.07
Emond
24ti
441 if/ 2.1 12 Ik-iiim
2T
CADM
10.7
Emond
I/O"
2."5 ' 2.1 12 hours)
I/1'!.
CADM
107
Emond
7,328
ll>.4l><> ( a 2.1 12 hours)
6,587
CADM
-
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.321
Emond
19.5
33.0 (@ 2,116 hours)
28.1
CADM
14.8
24.5
23.2
1.07
Emond
66.7
106 (@ 2,116 hours)
87.4
CADM
59.4
91.9
84.2
10.7
Emond
680
966 (@ 2,117 hours)
736
CADM
768
1,000
825
107
Emond
6,768
9,000 (@ 2,117 hours)
6,482
CADM
8,343
10,306
7,863
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.321
Emond
58.6
91.9 (@ 2,135 hours)
89.4
CADM
56.5
85.9
82.7
1.07
Emond
156
228 (@ 2,130 hours)
219
CADM
152
210
199
10.7
Emond
884
1,149 (@ 2,124 hours)
1,075
CADM
690
815
735
107
Emond
4,818
5,946 (@ 2,120 hours)
5,347
CADM
2,770
3,224
2,684
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-125 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.321
Emond
5.97
9.50 (@ 2,117 hours)
8.93
CADM
7.43
11.4 (@ 2,121 hours)
10.9
1.07
Emond
16.9
25.3 (@ 2,116 hours)
23.2
CADM
20.9
29.3
27.7
10.7
Emond
117
158 (@ 2,116 hours)
135
CADM
119
145
127
107
Emond
849
1,100 (@ 2,116 hours)
865
CADM
727
875
694
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.321
Emond
0.564
0.885 (@ 2,116 hours)
0.771
CADM
-
-
1.07
Emond
1 4"
2 15 (a 2.1 l<> Ikuiis)
1.83
CADM
10.7
Emond
" 5X
S i ( a 2.1 l<> Ikuiis)
8.07
CADM
107
Emond
M) --
'5 i a 2.11" Ikuiis)
29.8
CADM
1
2
3 C.3.1.6. Hassoun et al. (2000)
Type:
Rat
Dose:
0, 3, 10, 22, 46, 100 ng/kg-day (2.14, 7.14,
15.7, 32.9, and 71.4 ng/kg-day adjusted
doses)
Strain:
Sprague Dawley
Route:
Oral gavage
Body
weight:
8 weeks old
(BW=215g)
Regime:
5 days/week for 13 weeks
Sex:
Female
Simulation time:
2184 hours
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-126 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
2.14
Emond
1,068
1,720 (@ 2,112 hours)
1,303
CADM
-
-
7.14
Emond
2.542
4,:4<.u/ :.ii: Mmm
2.^1)1
CADM
-
-
15.7
Emond
4.489
"X'o ( a 2.112 Ikumm
4.'U"
CADM
-
-
32.9
Emond
7.718
I4.2(i(i i a 2.112 Ikuiis)
S.2
CADM
-
-
71.4
Emond
13.960
2".^(."( a 2.1 12 Ikumm
I4.(>
CADM
-
-
LIVER CONCENTRATIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
2.14
Emond
267
399 (@ 2,116 hours)
349
CADM
-
-
-
7.14
Emond
888
1,259 (@ 2,117 hours)
1,079
CADM
-
-
-
15.7
Emond
1,948
2,689 (@ 2,117 hours)
2,278
CADM
-
-
-
32.9
Emond
4,055
5,484 (@ 2,117 hours)
4,607
CADM
-
-
-
71.4
Emond
8,775
11,692 (@ 2,117 hours)
9,754
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
2.14
Emond
179
243 (@2,126 hours)
235
CADM
-
-
-
7.14
Emond
427
553 (@2,124 hours)
528
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-127 DRAFT—DO NOT CITE OR QUOTE
-------
15.7
Emond
755
958 (@ 2,123 hours)
908
CADM
-
-
-
32.9
Emond
1,299
1,627 (@ 2,122 hours)
1,529
CADM
-
-
-
71.4
Emond
2,350
2,928 (@ 2,121 hours)
2,727
CADM
-
-
-
BODY BURDEN (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
2.14
Emond
27.4
38.9 (@ 2,116 hours)
35.7
CADM
-
-
-
7.14
Emond
76.9
105 (@ 2,116 hours)
93.7
CADM
-
-
-
15.7
Emond
153
205 (@ 2,116 hours)
180
CADM
-
-
-
32.9
Emond
295
390 (@ 2,116 hours)
339
CADM
-
-
-
71.4
Emond
600
785 (@ 2,116 hours)
674
CADM
-
-
-
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
2.14
Emond
6.28
8.48 (@ 2,116 hours)
7.67
CADM
-
7.14
Emond
1 '"
1" 5 ( a 1.1 l<> Ikuiis)
15 "
CADM
-
15.7
Emond
:: (i
1~ 1 ( a 1.1 l<> Ikuiis)
24 4
CADM
-
32.9
Emond
s
^ lid 2.1 l<> Ikuiis)
i5 (¦
CADM
-
71.4
Emond
4" 5
55 (i ( a 2.1 l<> Ikuiis)
5() (.
CADM
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-128 DRAFT—DO NOT CITE OR QUOTE
-------
1 C.3.1.7. Kitchin and Woods (1979)
Type:
Rats
Dose:
0, 0.6, 2, 4, 20, 60, 200, 600, 2000, 5000,
20000 ng/kg-day
Strain:
Sprague-Dawley
Route:
Oral gavage
Body weight:
200 to 250 g (BW set to
225 g)
Regime:
Single dose
Sex:
Female
Simulation
time:
24 hours
2 aThe CADM model was not ran because the dosing duration is lower than the resolution of the model (1 week).
3
4
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
0.6
Emond
46.8
69.5 (@ 0 hours)
18.0
CADM
-
2
Emond
i::
2 '2 ( a ii hum's)
5" 1
CADM
-
4
Emond
::i
4<>' ( (i (i Ikhiim
|u<)
CADM
-
20
Emond
X'K.
2.' IS ( a o Ikmiis)
4(.:
CADM
-
60
Emond
(¦.'U'J ( a ii Ikuiis)
I.I (.5
CADM
-
200
Emond
6,393
2 V IS5 ( a (i Ikuiis)
CADM
-
600
Emond
1
<>l>.<>5~ ( (i (i Ikuiis)
7,345
CADM
-
2,000
Emond
5( ).<)')<)
2 '2.55D ( a ii Ikuiis)
I'M.'-
CADM
-
5,000
Emond
120,130
581,930 (@ 0 hours)
43,511
CADM
-
20,000
Emond
475,600
2,332,100 (@0 hours)
158,970
CADM
PSfSiiP?5§§§S
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-129 DRAFT—DO NOT CITE OR QUOTE
-------
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.6
Emond
3.99
3.81 (@ 4 hours)
1.60
CADM
-
-
-
2
Emond
11.7
12.9 (@ 4 hours)
6.01
CADM
-
-
-
4
Emond
23.4
26.3 (@ 4 hours)
13.2
CADM
-
-
-
20
Emond
129
143 (@ 6 hours)
85.2
CADM
-
-
-
60
Emond
422
463 (@ 8 hours)
305
CADM
-
-
-
200
Emond
1,525
1,666 (@ 9 hours)
1,194
CADM
-
-
-
600
Emond
4,822
5,258 (@ 10 hours)
3,987
CADM
-
-
-
2,000
Emond
16,606
18,081 (@ 11 hours)
14,296
CADM
-
-
-
5,000
Emond
41,973
45,674 (@ 11 hours)
36,821
CADM
-
-
-
20,000
Emond
167,820
182,580 (@11 hours)
149,280
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.6
Emond
2.11
3.03 (@ 72 hours)
3.03
CADM
-
-
-
2
Emond
5.54
9.57 (@ 72 hours)
9.57
CADM
-
-
-
4
Emond
10.1
18.2 (@ 72 hours)
18.2
CADM
-
-
-
20
Emond
42.1
76.4 (@ 72 hours)
76.4
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-130 DRAFT—DO NOT CITE OR QUOTE
-------
60
Emond
110
192 (@ 72 hours)
192
CADM
-
-
-
200
Emond
317
512 (@ 72 hours)
512
CADM
-
-
-
600
Emond
851
1,250 (@ 72 hours)
1,250
CADM
-
-
-
2,000
Emond
2,621
3,481 (@58 hours)
3,462
CADM
-
-
-
5,000
Emond
6,361
8,049 (@ 45 hours)
7,887
CADM
-
-
-
20,000
Emond
25,402
31,187 (@35 hours)
29,738
CADM
-
-
-
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.6
Emond
0.429
0.341 (@ 9 hours)
0.331
CADM
-
-
-
2
Emond
1.18
1.14 (@ 8 hours)
1.09
CADM
-
-
-
4
Emond
2.24
2.27 (@ 8 hours)
2.15
CADM
-
-
-
20
Emond
10.7
11.3 (@ 8 hours)
10.4
CADM
-
-
-
60
Emond
31.8
33.8 (@ 7 hours)
30.3
CADM
-
-
-
200
Emond
105
112 (@ 7 hours)
98
CADM
-
-
-
600
Emond
315
337 (@ 7 hours)
288
CADM
-
-
-
2,000
Emond
1,049
1,123 (@ 7 hours)
945
CADM
-
-
-
5,000
Emond
2,621
2,806 (@ 7 hours)
2,343
CADM
-
-
-
20,000
Emond
10,469
11,215 (@ 7 hours)
9,299
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-131 DRAFT—DO NOT CITE OR QUOTE
-------
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.6
Emond
0.284
0.407 (@ 3 hours)
0.238
CADM
-
2
Emond
o ~:x
1 u" ( a ~i Ikuiim
ii 4~<>
CADM
-
4
Emond
i
1 lU ((i ' Ikuiim
0 "•«
CADM
PPPPPPPPP#^
-
20
Emond
4 '«)
" "4 ui 2 Ikuiim
: «>5
CADM
-
60
Emond
11:
IS 4 ( a 2 Ikuiis)
¦ {)--
CADM
200
Emond
25.1
4(1 S ( a 1 Ikuiim
h>"
CADM
PPP!!PPPP5PPPPPPS
600
Emond
45 S
(>S 2 ( a 1 Ikuiim
V, 0
CADM
-
2,000
Emond
73.3
93.1 (@ 1 hours)
59.1
CADM
-
5,000
Emond
90.9
104 (@ 1 hours)
79.9
CADM
PPP:^/PIPIP^^
-
20,000
Emond
106
110 (@ 1 hours)
101
CADM
1
2
3 C.3.1.8. Kocibaetal. (1976)
Type:
Rats
Dose:
1, 10, 100, 1000 ng/kg-day
Strain:
Sprague-Dawley
(Spartan)
Route:
Oral gavage
Body weight:
170-190 g(bw=180g)
Regime:
5 days/week for 13 weeks
Sex:
Female
Simulation
time:
4,368 hours
(13wk exposed + 13 wk post exposures)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-132 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.714
Emond
398
761 (@ 2,112 hours)
163
CADM
-
7.143
Emond
i.sr
4. llK> ( (i 1.1 12 hum's)
CADM
-
71.43
Emond
').<)< )2
2<>.X~2 ( a 2.112 Ikums)
X2<)
CADM
-
714.3
Emond
(ill. ^xx
22(>.4"u ( a 2.1 12 Ikums)
2.u"2
CADM
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.714
Emond
70.9
140 (@ 2,116 hours)
21.4
CADM
89.0
192
12.1
7.143
Emond
595
1,259 (@ 2,117 hours)
62.4
CADM
970
2,007
29.0
71.43
Emond
5,391
11,693 (@ 2,117 hours)
183
CADM
9,841
20,170
88.0
714.3
Emond
51,476
112,580 (@2,117 hours)
670
CADM
98,617
201,814
455
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.714
Emond
68.3
114 (@ 2,129 hours)
28.8
CADM
120
190
43.0
7.143
Emond
313
553 (@ 2,124 hours)
66.2
CADM
456
787
67.0
71.43
Emond
1,552
2,925 (@2,121 hours)
148
CADM
3,036
5,748
117
714.3
Emond
10,415
21,127 (@ 2,120 hours)
379
CADM
28,382
55,013
274
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-133 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.714
Emond
9.03
16.1 (@ 2,116 hours)
3.41
CADM
11.5
20.0
3.75
7.143
Emond
53.7
105 (@ 2,116 hours)
8.44
CADM
65.3
126
6.22
71.43
Emond
377
785 (@ 2,116 hours)
20.8
CADM
553
1,113
12.0
714.3
Emond
3,230
6,961 (@ 2,116 hours)
62.4
CADM
5,401
10,967
37.0
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.714
Emond
2.44
4.17 (@ 2,116 hours)
1.02
CADM
-
7.143
Emond
lu |
1" 5 ( a 2.1 l<> Ikuiis)
:
CADM
-
71.43
Emond
55 (i ( a 2.1 l<> Ikuiis)
4l>5
CADM
-
714.3
Emond
<.<>_
•JX 2 ( a 2.11" Ikuiis)
1 1 "
CADM
-
1 C.3.1.9. Kociba et al. (1978) Female
Type:
Rats
Dose:
0, 1, 10, 100 ng/kg-day
Strain:
Sprague-Dawley
(Spartan)
Route:
Dietary
Body weight:
170-190 g (bw=180)
Regime:
104 weeks
Sex:
Female
Simulation time:
17,472 hours
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-134 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
853
1,058 (@ 17,448 hours)
929
CADM
-
10
Emond
vh:
5.(>l)Xu/ 1 ~44X hum's)
CADM
-
100
Emond
'1 (i 1 ~44X hum's)
:u.44i
CADM
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
192
226 (@ 17,452 hours)
218
CADM
292
333
333
10
Emond
1,618
1,742 (@ 17,452 hours)
1,665
CADM
2,981
3,342
3,342
100
Emond
14,892
15,673 (@ 17,452 hours)
14,907
CADM
29,917
33,432
33,432
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
147
165 (@ 17,457 hours)
164
CADM
196
229
181
10
Emond
680
713 (@ 17,454 hours)
706
CADM
861
1,015
789
100
Emond
3,663
3,788 (@ 17,454 hours)
3,731
CADM
6,756
7,939
6,203
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
21.2
24.3 (@ 17,452 hours)
23.8
CADM
26.0
27.0
27.0
10
Emond
131
140 (@ 17,452 hours)
136
CADM
169
176
176
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-135 DRAFT—DO NOT CITE OR QUOTE
-------
100
Emond
989
1,039 (@ 17,452 hours)
994
CADM
1,546
1,601
1,601
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
5.11
5.77 (@ 17,452 hours)
5.59
CADM
-
10
Emond
2() o
:i 11 a r.45: Mmm
20 4
CADM
-
100
Emond
')
(> 1 5 ( a 1 ".452 Ikuiis)
(.0 1
CADM
1 C.3.1.10. Kocibaetal. (1978) Male
Type:
Rats
Dose:
0, 1, 10, 100 ng/kg-day
Strain:
Sprague-Dawley
(Spartan)
Route:
Dietary
Body weight:
Body weight
approximated to be
250 g
Regime:
104 weeks
Sex:
Male
Simulation time:
17,472 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1
Emond
860
1,079 (@ 17,448 hours)
938
CADM
-
10
Emond
VM5
5.15' id 1 ~44X hum's)
V'J|(.
CADM
-
100
Emond
21.' U
'2.<>5X (
-------
LIVER CONCENTRATIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1
Emond
194
229 (@ 17,452 hours)
221
CADM
-
10
Emond
\r:^ i a r.45: Ikumm
1.(4'J
CADM
-
100
Emond
I4.X'JX
15.(>" 1(^/1 ".452 Ikuiis)
I4/JI2
CADM
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1
Emond
148
167 (@ 17,456 hours)
166
CADM
-
10
Emond
(.SO
1 ".454 Ikuiis)
"(H
CADM
-
100
Emond
V<"
vSii' ( a 1 ".45 ' Ikuiis)
V"4"
CADM
-
-
BODY BURDEN (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1
Emond
21 4
24 6 (fa) 17.452 hours')
24 1
CADM
-
10
Emond
i
1 ( a 1 ".452 Ikuiis)
1 U
CADM
-
100
Emond
991
1 .(>41(^/1 ".452 Ikuiis)
•W5
CADM
-
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1
Emond
5.15
5.83 (@ 17,452 hours)
5.64
CADM
-
10
Emond
:o o
2 1 (i ( a 1 ".452 Ikuiis)
20 ^
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-137 DRAFT—DO NOT CITE OR QUOTE
-------
100
Emond
60.0
61.5 (@ 17,452 hours)
60.1
CADM
-
1 C.3.1.11. Latchoumycandane and Mathur (2002)
Type:
Rat
Dose:
0, 1, 10, lOOng/kg-day
Strain:
Wistar
Route:
Mouth pipetting
Body weight:
45 days old
(BW set to 200g)
Regime:
1/day for 45 days
Sex:
Male
Simulation
time:
1,104 hours (1,080 daily exposure and 24 hours
before sacrifice)
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
4^7
754 (((i> 1.056 hours')
630
CADM
-
10
Emond
4.5u5 ( 5<> hum's)
v:_4
CADM
100
Emond
15,092
29,672 (@ 1,056 hours)
17,698
CADM
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
79.7
138 (@ 1,060 hours)
128
CADM
116
217
217
10
Emond
911
1,423 (@ 1,060 hours)
1,282
CADM
1,669
2,550
2,550
100
Emond
9,650
14,015 (@ 1,061 hours)
12,439
CADM
17,681
25,915
25,915
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
70.7
113 (@ 1,072 hours)
112
CADM
150
220
220
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-138 DRAFT—DO NOT CITE OR QUOTE
-------
10
Emond
420
608 (@ 1,065 hours)
592
CADM
744
1,009
1,009
100
Emond
2,467
3,425 (@ 1,062 hours)
3,273
CADM
5,719
7,866
7,866
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
9.68
15.9 (@ 1,060 hours)
15.2
CADM
14.0
22.2
22.2
10
Emond
77.5
117 (@ 1,060 hours)
108
CADM
106
157
157
100
Emond
651
933 (@ 1,060 hours)
842
CADM
988
1,439
1,439
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
2.67
4.12 (@ 1,060 hours)
3.85
CADM
-
10
Emond
1 ^ S
IS S ( (1 I.(Hill hum's)
r 5
CADM
100
Emond
4X S
I .()(¦() hum's)
5(i ii
CADM
1
2
3
4
5
6
7
8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-139 DRAFT—DO NOT CITE OR QUOTE
C.3.1.12. Lietal. (1997)
Type:
Rats
Dose:
0, 3, 10, 30, 100, 300, 1000, 3000,
10000, 30000 ng/kg/day
Strain:
Sprague-Dawley
Route:
Gastric intubation
Body weight:
22 day old, 55 to 58 g
(BW set to 56.5 g)
Regime:
One dose for one day
Sex:
Female
Simulation time:
24 hours
aThe CADM model was not ran because the dosing duration is lower than the resolution of the model (1 week)
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Time-weighted Ave
Max
Terminal
3
Emond
147
259 (@ 1 hours)
98.9
CADM
-
10
Emond
440
Sf.2 (7i 1 Ik.iirO
295
CADM
-
30
Emond
1.15<>
2.5SI i a 1 hum's)
"5"
CADM
-
100
Emond
3.232
S.5S5 ( a 1 hums)
2,()2(>
CADM
-
-
300
Emond
S.2<><>
25."So i
-------
1,000
Emond
7,938
8,671 (@ 9 hours)
8,094
CADM
-
-
-
3,000
Emond
24,474
26,639 (@ 9 hours)
25,267
CADM
-
-
-
10,000
Emond
82,349
89,464 (@ 9 hours)
85,597
CADM
-
-
-
30,000
Emond
245,610
265,670 (@ 10 hours)
255,390
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Model
Metric
Time-weighted Ave
Max
Terminal
3
Emond
8.75
12.7 (@ 24 hours)
12.7
CADM
-
-
-
10
Emond
26.6
38.0 (@ 24 hours)
38.0
CADM
-
-
-
30
Emond
70.8
98.9 (@ 24 hours)
98.9
CADM
-
-
-
100
Emond
202
273 (@ 24 hours)
273
CADM
-
-
-
300
Emond
530
689 (@ 24 hours)
689
CADM
-
-
-
1,000
Emond
1,573
1,958 (@ 24 hours)
1,958
CADM
-
-
-
3,000
Emond
4,433
5,358 (@ 24 hours)
5,358
CADM
-
-
-
10,000
Emond
14,428
17,119 (@ 24 hours)
17,119
CADM
-
-
-
30,000
Emond
44,361
51,948 (@ 22 hours)
51,898
CADM
-
-
-
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Model
Metric
Time-weighted Ave
Max
Terminal
3
Emond
1.60
1.70 (@ 8 hours)
1.68
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-141 DRAFT—DO NOT CITE OR QUOTE
-------
10
Emond
5.33
5.66 (@ 8 hours)
5.56
CADM
-
-
-
30
Emond
15.9
16.9 (@ 8 hours)
16.5
CADM
-
-
-
100
Emond
52.8
56.2 (@ 7 hours)
54.5
CADM
-
-
-
300
Emond
158
169 (@ 7 hours)
163
CADM
-
-
-
1,000
Emond
525
561 (@ 7 hours)
539
CADM
-
-
-
3,000
Emond
1,574
1,684 (@ 7 hours)
1,611
CADM
-
-
-
10,000
Emond
5,240
5,610 (@ 7 hours)
5,360
CADM
-
-
-
30,000
Emond
15,758
16,815 (@ 7 hours)
16,041
CADM
-
-
-
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Time-weighted Ave
Max
Terminal
3
Emond
1
1.37 (@ 3 hours)
1
CADM
-
10
Emond
3
4 In ( a 2 Ikuiis)
2
CADM
-
30
Emond
6
In 5 ( a 2 Ikuiis)
s
CADM
-
100
Emond
l(>
25 *> ui 2 Ikuiis)
i:
CADM
-
300
Emond
31.25
50.1 (@ 1 hours)
24.58
CADM
-
1,000
Emond
56.75
79.8 (@ 1 hours)
47.65
CADM
-
3,000
Emond
81.29
98.4 (@ 1 hours)
73.34
CADM
-
10,000
Emond
99.77
108 (@ 1 hours)
95.70
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-142 DRAFT—DO NOT CITE OR QUOTE
-------
30,000
Emond
108.04
111 (@ 1 hours)
106.23
CADM
-
1 C.3.1.13. NTP (1982)—Female Rats, Chronic
Type:
Rat
Dose:
10, 50 and 500 ng/kg/wk,
two doses per week
Strain:
Osborne-Mendel
Route:
Oral gavage
Body weight
6 weeks old
(BW set to 25 Og)
Regime:
Biweekly
Sex:
Female
Simulation time
17,976 hours (107 weeks)= (104 weeks of
exposure + 3 weeks observation post-treatment)
aThe CADM model simulates for 104 weeks only (17,472 hours). As a result, the terminal values from the CADM
model are overestimated compared to the Emond model, which considered an additional 3 weeks post exposure.
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.4
Emond
1,072
1,719 (@ 17,388 hours)
685
CADM
-
7.1
Emond
3,111
6,054 (@ 17,388 hours)
1,622
CADM
-
71
Emond
16,207
45,310 (@ 17,388 hours)
6,253
CADM
-
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.4
Emond
263
310 (@ 17,394 hours)
143
CADM
15,318
20,170
7,102
7.1
Emond
1,163
1,338 (@ 17,394 hours)
474
CADM
30,700
40,353
14,200
71
Emond
10,596
12,182 (@ 17,395 hours)
3,134
CADM
30,700
40,353
14,200
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-143 DRAFT—DO NOT CITE OR QUOTE
-------
FA T CONCENTRA TIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.4
Emond
185
200 (@ 17,412 hours)
124
CADM
4,655
5,748
2,107
7.1
Emond
537
569 (@ 17,409 hours)
297
CADM
9,064
11,224
3,964
71
Emond
2,798
2,973 (@ 17,404 hours)
1,173
CADM
17,879
22,172
7,671
BODY BURDEN (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.4
Emond
27.7
31.2 (@ 17,393 hours)
16.9
CADM
855
1,113
403
7.1
Emond
98.5
110 (@ 17,393 hours)
46.6
CADM
1,695
2,208
787
71
Emond
720
814 (@ 17,393 hours)
241
CADM
3,375
4,395
1,556
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.4
Emond
7 28r^ 17.^92 hours')
4 17
CADM
-
7.1
Emond
l<> 5
IS 5 ( a 1Ikuiis)
CADM
-
71
Emond
52.3
5<>4u/ 1Ikuiis)
29.1
CADM
-
1 C.3.1.14. NTP (1982)—Male Rats, Chronic
Type:
Rat
Dose:
10, 50 and 500 ng/kg/wk,
two doses per week
Strain:
Osborne-Mendel
Route:
Oral gavage
Body weight
6 weeks old
(BW set to 350g)
Regime:
Biweekly
(Simulation has been perform using female BW
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-144 DRAFT—DO NOT CITE OR QUOTE
-------
Sex:
Simulation time
17,976 hours (107 weeks)= (104 weeks of
Male
exposure + 3 weeks observation post-treatment)
aThe CADM model simulates for 104 weeks only (17,472 hours). As a result, the terminal values from the CADM
model are overestimated compared to the Emond model, which considered an additional 3 weeks post exposure.
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.4
Emond
Mi":
l."0( ,7 l~.'88 lkuiiM
(.81
CADM
7.1
Emond
VI l(>
(>. 't> 1 ( (1 1 '88 IkUII's)
i.(.::
CADM
71
Emond
16,272
47,951 (@ 17,388 hours)
6,269
CADM
-
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.4
Emond
263
306 (@ 17,394 hours)
141
CADM
-
-
7.1
Emond
u<.:
l.''4u/ 1 Ikimis)
4"'
CADM
71
Emond
|u 51>X
12,170 ((a Ikuiim
'.I4U
CADM
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.4
Emond
185
199 (@ 17,412 hours)
123
CADM
-
7.1
Emond
538
569 (@ 17,409 hours)
298
CADM
-
71
Emond
2,809
2,983 (@ 17,404 hours)
1,185
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-145 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.4
Emond
27.7
30.9 (@ 17,393 hours)
16.8
CADM
-
7.1
Emond
<>X (.
1 In ( a 1 ' Ikimis)
4(> (¦
CADM
71
Emond
S 1(1 ( (1 1 i'Ji IkmII's)
242
CADM
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.4
Emond
6.33
7.22 (@ 17,392 hours)
4.14
CADM
-
7.1
Emond
I<> 4
1S 4 ( a 1 Ikimis)
1(1
CADM
71
Emond
5(> ^ ui 1 ' Ikimis)
y> 1
CADM
1
1 C.3.1.15. NTP (1982)—Female Mice, Chronic
Type:
Mice
Dose:
40, 200 and 2000ng/kg/wk,
two doses during the week
Strain:
B6C3F1
Route:
Oral gavage
Body weight
6 weeks old
(BW set to 23g)
Regime:
Biweekly
Sex:
Female
Simulation time
17,976 hours (107 weeks)= (104 weeks of
exposure + 3 weeks observation post-treatment)
"The mice chronic exposure could not be simulated with the CADM model because this model simulates for only
123 days.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-146 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
5.7
Emond
1,064
2,684 (@ 17,220 hours)
569
CADM
-
28.6
Emond
VIX4
Hi.'J 15 < r/ 1 'XX Ikuiis)
1.334
CADM
-
286
Emond
r.4<)(,
t a 1 ",22o Ikuiis)
4.899
CADM
-
-
LIVER CONCENTRATIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
5.7
Emond
486
587 (@ 17,227 hours)
209
CADM
-
-
28.6
Emond
2.(>2l> < <-/ 1". ''15 Ikuiis)
(.82
CADM
-
-
286
Emond
2ii.5 15
24. '5' id 1 'li(i Ikmiim
4.2-2
CADM
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
5.7
Emond
7RQ (?a) 17.^24 hours')
4^6
CADM
-
-
28.6
Emond
:.iiu
2.'< r/ 1 ~.4o4 Ikiiii's)
1,059
CADM
-
-
286
Emond
12.0(1.
12.X(i 1(^/1 ",4oo Ikmii's)
4.151
CADM
-
-
BODY BURDEN (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
5.7
Emond
91.2
103 (@ 17,225 hours)
48.5
CADM
-
28.6
Emond
'25
'"0 ( (1 1 ". IkUII's)
130
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-147 DRAFT—DO NOT CITE OR QUOTE
-------
286
Emond
2,367
2,740 (@ 17,393 hours)
615
CADM
-
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
5.7
Emond
6.13
7.32 (@ 17,225 hours)
3.44
CADM
-
28.6
Emond
i(> i
IS ( (1 1W IkmII's)
-<.x
CADM
-
286
Emond
51 S
(>" S ( a 2 hum's)
(.
CADM
-
1 C.3.1.16. NTP (1982)—Male Mice, Chronic
Type:
Mice
Dose:
10, 50 and 500ng/kg/wk,
two doses during the week
Strain:
B6C3F1
Route:
Oral gavage
Body weight
6 weeks old
(BW set to 25g)
Regime:
Biweekly
Sex:
Male
Simulation time
17,976 hours (107 weeks)= (104 weeks of
exposure + 3 weeks observation post-treatment)
aThe mice chronic exposure could not be simulated with the CADM model because this model simulates for only
123 days.
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.4
Emond
420
842 (@ 17,136 hours)
270
CADM
-
7.1
Emond
1 24<)
\^<>2u/ 1 ^t>4 hum's)
(44
CADM
-
71
Emond
(-.1 IS
25.~'<>u/ 1^SS Ikuiis)
2.2H4
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-148 DRAFT—DO NOT CITE OR QUOTE
-------
LIVER CONCENTRATIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.4
Emond
137
165 (@ 17,142 hours)
76.7
CADM
-
7.1
Emond
id \~r-w Ikumm
24"
CADM
-
71
Emond
5.-1
<>.^2Xu/ 1'l>5 hum's)
i.'s:
CADM
-
-
FA T CON CENTRA TIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.4
Emond
289
314 (@ 17,243 hours)
202
CADM
-
7.1
Emond
S54
IX ( (i 1 ".4(1" hum's)
4 <><¦
CADM
-
71
Emond
4.:r
4.4l><> < <-/ 1 ~.4t>2 hums)
I."W
CADM
-
BODY BURDEN (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.4
Emond
t,2 2
^ m 17.141 hours')
21 1
CADM
-
7.1
Emond
III')
12^ id 1 ~.'ii'J Ikmiim
55 S
CADM
-
71
Emond
"u|
S()2 ( a 1 Ikmiim
2'5
CADM
-
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.4
Emond
2.54
3.04 (@ 17,141 hours)
1.67
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-149 DRAFT—DO NOT CITE OR QUOTE
-------
7.1
Emond
7.06
8.41 (@ 17,309 hours)
3.87
CADM
-
71
Emond
:<.x
'2 4 i r/ 2 hours)
i: i
CADM
-
1 C.3.1.17. NTP (2006) 31 Weeks
Type:
Rat
Dose:
0, 3, 10, 22, 46, 100 ng/kg-day
Strain:
Sprague Dawley
Route:
Oral gavage
Body
weight:
8 weeks old
(BW=215g)
Regime:
5 days/weeks for 31 weeks
Sex:
Female and male
Simulation time:
5208 hours (31 weeks)
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
1,284
1,792 (@ 3,960 hours)
1,360
CADM
-
-
7.14
Emond
2.932
4.'5<>u/ Ikuiis)
CADM
-
PPP^PPPPPPSIfPPPPPPPP^
-
15.7
Emond
5.075
~.1>5X ( (i V'Xii) hours)
5.ii vj
CADM
-
-
32.9
Emond
8.629
14.4l<>u/ V'Xii) linni\)
X.4I"
CADM
71.4
Emond
15.503
2~.~'X i a 5.1 linni\)
14.X
CADM
-
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
341
425 (@ 3,964 hours)
371
CADM
-
-
-
7.14
Emond
1,077
1,312 (@ 4,133 hours)
1,125
CADM
-
-
-
15.7
Emond
2,298
2,760 (@ 3,965 hours)
2,336
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-150 DRAFT—DO NOT CITE OR QUOTE
-------
32.9
Emond
4,698
5,599 (@ 3,965 hours)
4,711
CADM
-
-
-
71.4
Emond
10,036
11,910 (@5,141 hours)
9,956
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
220
256 (@ 4,141 hours)
245
CADM
-
-
-
7.14
Emond
502
571 (@ 4,139 hours)
545
CADM
-
-
-
15.7
Emond
868
979 (@ 4,138 hours)
926
CADM
-
-
-
32.9
Emond
1,476
1,657 (@ 4,137 hours)
1,558
CADM
-
-
-
71.4
Emond
2,653
2,979 (@ 5,144 hours)
2,776
CADM
-
-
-
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
34.2
41.2 (@ 3,964 hours)
37.6
CADM
-
-
-
7.14
Emond
91.7
109 (@ 4,132 hours)
97.2
CADM
-
-
-
15.7
Emond
178
210 (@ 3,964 hours)
184
CADM
-
-
-
32.9
Emond
339
398 (@ 4,132 hours)
346
CADM
-
-
-
71.4
Emond
683
799 (@5,140 hours)
687
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-151 DRAFT—DO NOT CITE OR QUOTE
-------
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
2.14
Emond
7.48
8.84 (@ 3,964 hours)
7.98
CADM
-
7.14
Emond
15 <>
1" K> ((i 4.1 '2 Ikuiis)
I(> 1
CADM
-
15.7
Emond
24 4
2" 5 ( a \'J(4 Ikuiis)
24 S
CADM
-
32.9
Emond
35.7
-'Hi hi \l><>4 Ikuiis)
}(,()
CADM
-
-
71.4
Emond
5()l)
55 4 (
4. '5'J ( a S.S'2 Ikuiis)
2,993
CADM
-
15.7
Emond
5.25'J
~.1>5X i a v'KiO Ikuiis)
5.U52
CADM
-
32.9
Emond
X.'HX
I4.4(>u ( a S.S'2 Ikuiis)
X.4'X
CADM
-
71.4
Emond
16,001
27,846 (@ 8,832 hours)
14,916
CADM
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-152 DRAFT—DO NOT CITE OR QUOTE
-------
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
365
425 (@ 3,964 hours)
373
CADM
-
-
-
7.14
Emond
1,138
1,312 (@ 8,837 hours)
1,127
CADM
-
-
-
15.7
Emond
2,407
2,760 (@ 3,965 hours)
2,344
CADM
-
-
-
32.9
Emond
4,902
5,611 (@ 8,837 hours)
4,726
CADM
-
-
-
71.4
Emond
10,443
11,943 (@ 8,837 hours)
9,989
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
233
256 (@ 8,845 hours)
247
CADM
-
-
-
7.14
Emond
525
572 (@ 8,843 hours)
546
CADM
-
-
-
15.7
Emond
904
979 (@ 8,842 hours)
928
CADM
-
-
-
32.9
Emond
1,533
1,661 (@ 8,841 hours)
1,562
CADM
-
-
-
71.4
Emond
2,750
2,987 (@ 8,840 hours)
2,785
CADM
-
-
-
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
36.4
41.2 (@ 3,964 hours)
37.8
CADM
-
-
-
7.14
Emond
96.4
109 (@ 8,836 hours)
97.3
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-153 DRAFT—DO NOT CITE OR QUOTE
-------
15.7
Emond
186
210 (@ 8,836 hours)
185
CADM
-
-
-
32.9
Emond
354
399 (@ 8,836 hours)
347
CADM
-
-
-
71.4
Emond
709
802 (@ 8,836 hours)
689
CADM
-
-
-
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
8 17
QWff) 17.572 hours')
8 4^
CADM
-
7.14
Emond
16.6
IS o ( a 1 ~.5~2 Ikuiis)
16.2
CADM
-
15.7
Emond
25 (.
2" (> ( a 1 ~5"2 Ikuiis)
24
CADM
-
32.9
Emond
37.3
^ " ( a 1 ~5"2 Ikuiis)
CADM
-
71.4
Emond
52.7
55.5 (@ 17,572 hours)
51.2
CADM
1 C.3.1.19. NTP (2006) 2 Years
Type:
Rat
Dose:
0, 3, 10, 22, 46, 100 ng/kg-day
Strain:
Sprague Dawley
Route:
Oral gavage
Body
weight:
8 weeks old
(BW=215g)
Regime:
5 days/weeks for 105 weeks
Sex:
Female and male
Simulation time:
17,640 hours* (105 weeks)
aThe CADM model simulates for 104 weeks only (17,472 hours). As a result, the terminal values from the CADM
model may be underestimated compared to the Emond model, which considers the full 105 weeks of exposure.
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
1.408
1.910 i a I~5<>S Ikuiis)
1.444
CADM
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-154 DRAFT—DO NOT CITE OR QUOTE
-------
7.14
Emond
3,137
4,389 (@ 17,568 hours)
3,007
CADM
-
-
15.7
Emond
5.393
X.Ui'Ju/ 1 ",5(>S Ikuiis)
5.()-<>
CADM
-
-
32.9
Emond
9.129
14.542 ( a 1 ~5<>X hum's)
S.4(>S
CADM
71.4
Emond
16.361
2",99l ( a l",5o8 hours;
14,951
CADM
-
I
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
385
460 (@ 17,572 hours)
403
CADM
632
715
715
7.14
Emond
1,177
1,320 (@ 17,573 hours)
1,135
CADM
2,127
2,387
2,387
15.7
Emond
2,487
2,779 (@ 17,573 hours)
2,361
CADM
4,691
5,252
5,252
32.9
Emond
5,051
5,637 (@ 17,573 hours)
4,749
CADM
9,822
10,984
10,984
71.4
Emond
10,734
11,976 (@ 17,573 hours)
10,018
CADM
21,366
23,880
23,880
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
2.14
Emond
243
271 (@ 17,581 hours)
261
CADM
302
355
277
7.14
Emond
541
575 (@ 17,579 hours)
549
CADM
667
787
611
15.7
Emond
930
985 (@ 17,578 hours)
934
CADM
1,242
1,463
1,138
32.9
Emond
1,574
1,667 (@ 17,577 hours)
1,568
CADM
2,369
2,787
2,173
71.4
Emond
2,821
2,995 (@ 17,576 hours)
2,792
CADM
4,890
5,748
4,489
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-155 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
2.14
Emond
38.1
44.0 (@ 17,572 hours)
40.4
CADM
46.0
48.0
48.0
7.14
Emond
99.5
109 (@ 17,572 hours)
97.9
CADM
125
130
130
15.7
Emond
192
211 (@ 17,572 hours)
186
CADM
257
267
267
32.9
Emond
364
400 (@ 17,572 hours)
348
CADM
520
538
538
71.4
Emond
729
804 (@ 17,572 hours)
691
CADM
1,110
1,149
1,149
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
2.14
Emond
8.17
9.30 (@ 17,572 hours)
8.43
CADM
-
7.14
Emond
1 (¦ (¦
IS () ( a 1 ~.5~2 Ikuiis)
i(.:
CADM
-
15.7
Emond
25 (.
2" (> ( a 1 ~5"2 Ikuiis)
24
CADM
-
32.9
Emond
•) j jy
1 ~5"2 Ikuiis)
CADM
-
71.4
Emond
52 "
55 5 UI r.5"2 Ik.iiis)
51:
CADM
-
-
1
2
3 C.3.1.20. Sewall et al. (1995)
Type:
Rat
Dose:
49, 149.8, 490, and 1750 ng/kg every two
weeks or 3.5, 10.7, 35, and 125 ng/kg-day
Strain:
Sprauge-Dawley
Route:
Oral gavage
Body weight:
12 wk old
(BW set to 25 Og)
Regime:
Once every 2 weeks for 30 weeks
Sex:
Female
Simulation time:
5040 hours
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-156 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
3.5
Emond
1,813
7,535 (@ 4,704 hours)
1,587
CADM
-
10.7
Emond
Vli.
2 1.2')' ( a 4."<>4 Ikuiis)
VIS'J
CADM
-
35
Emond
•>. I<>'
(>(>. 1 ( a 4."<>4 Ikuiis)
(>.lU5
CADM
-
125
Emond
:4.(.us
22X. i"u ( a 4.~t>4 Ikiiii's)
17,298
CADM
-
LIVER CONCENTRATIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
3.5
Emond
550
901 (@ 4,711 hours)
459
CADM
-
-
-
10.7
Emond
1,605
2,632 (@ 4,712 hours)
1,229
CADM
-
-
-
35
Emond
5,072
8,350 (@4,712 hours)
3,618
CADM
-
-
-
125
Emond
17,683
29,256 (@ 4,713 hours)
12,011
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
3.5
Emond
310
383 (@ 4,765 hours)
290
CADM
-
-
-
10.7
Emond
670
827 (@ 4,763 hours)
590
CADM
-
-
-
35
Emond
1,569
1,957 (@ 4,760 hours)
1,304
CADM
-
-
-
125
Emond
4,217
5,376 (@ 4,757 hours)
3,303
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-157 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
3.5
Emond
51.4
72.5 (@ 4,710 hours)
45.3
CADM
-
-
-
10.7
Emond
130
189 (@ 4,710 hours)
106
CADM
-
-
-
35
Emond
364
546 (@ 4,710 hours)
274
CADM
-
-
-
125
Emond
1,164
1,793 (@4,710 hours)
824
CADM
-
-
-
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
3.5
Emond
10.2
15.8 (@ 2 hours)
9.18
CADM
-
10.7
Emond
I'J S
^4 4 ( (i 1 hours)
I"u
CADM
-
35
Emond
TO
2 i r/ 1 hours)
'1 4
CADM
-
125
Emond
(.' i
l><>i a 1 hnui\)
55:
CADM
1
2
3 C.3.1.21. Smialowicz et al. (2008)
Type:
Mice
Dose:
0, 1.5, 15, 150, 450 ng/kg-day
Strain:
B6C3F1
Route:
Oral gavage
Body weight:
13 wk old
(BW set to 28g)
Regime:
5 days/week for 13 weeks
Sex:
Female
Simulation time:
2184
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-158 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.07
Emond
241
449 (@ 2,112 hours)
307
CADM
-
10.7
Emond
l.'ox
:.x:i i a :.i i: mmm
1 4(.o
CADM
-
107
Emond
2ii.() 1(11 a 2.112 hum's)
<..TX
CADM
-
321
Emond
r.-nx
54. '4(. i <¦/ 2.1 12 Ik.misi
15,650
CADM
-
LIVER CONCENTRATIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.07
Emond
67.1
107 (@ 2,116 hours)
91.5
CADM
59.0
92.0
88.0
10.7
Emond
683
971 (@ 2,117 hours)
787
CADM
767
1,000
907
107
Emond
6,784
9,010 (@ 2,117 hours)
7,043
CADM
8,349
10,306
8,998
321
Emond
20,218
26,379 (@ 2,117 hours)
20,405
CADM
25,344
31,006
26,967
FA T CONCENTRA TIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1.07
Emond
156
229 (@ 2,130 hours)
225
CADM
151
210
204
10.7
Emond
885
1,155 (@ 2,124 hours)
1,111
CADM
689
815
774
107
Emond
4,831
5,979 (@2,120 hours)
5,591
CADM
2,771
3,224
2,937
321
Emond
11,420
14,037 (@ 2,119 hours)
12,920
CADM
6,337
7,509
6,688
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-159 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.07
Emond
17.0
25.5 (@ 2,116 hours)
23.9
CADM
21.0
29.0
29.0
10.7
Emond
117
159 (@ 2,116 hours)
141
CADM
119
145
135
107
Emond
852
1,103 (@ 2,116 hours)
923
CADM
727
875
778
321
Emond
2,304
2,958 (@2,116 hours)
2,419
CADM
1,961
2,370
2,080
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1.07
Emond
1.48
2.17 (@ 2,116 hours)
1.90
CADM
-
10.7
Emond
" (¦()
') X(i ( (i 1.1 l<> hum's)
x4:
CADM
-
107
Emond
}<) ;
'<¦ ii ( a 2.11" hums)
i
CADM
-
321
Emond
51 1
5X 1 (
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
315
889 (@ 8,568 hours)
308
CADM
-
100
Emond
~.X 14
(i\(i~' ( (1 > IkUII's)
(..iil4
CADM
1,000
Emond
5(1. | (>5
(i |u.4l>t> ( a S.5(>S Ikuiim
U.I 55
CADM
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
94.1
131 (@ 8,575 hours)
91.5
CADM
-
100
Emond
7.337
In. I ^2 ( a ~.l><>5 Ikuiim
5.(>(>'>
CADM
-
1,000
Emond
"U. ISO
l>~.(>55 ( a S.5 Ikuiis)
5l.'JX(i
CADM
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
215
247 (fa) s.fin hours')
2^0
CADM
-
100
Emond
5,337
5.1>I2 ( a S.5lU Ikuiis)
4,997
CADM
-
1,000
Emond
^S.S25 ( a S.5SS Ikuiis)
"0.5 1 (•
CADM
-
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
23.4
28.4 (@ 8,574 hours)
24.3
CADM
-
100
Emond
1. IX1) ( a ",^()2 Ikuiis)
"SI
CADM
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-161 DRAFT—DO NOT CITE OR QUOTE
-------
1,000
Emond
7,564
10,044 (@ 8,574 hours)
5,965
CADM
-
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
1.93
2.65 (@ 8,573 hours)
1.90
CADM
-
100
Emond
'1 s
5X 4 ( r/ 2 Ikuiis)
-1 _ _
CADM
1,000
Emond
"S (.
|(>i ( a 2 Ikuiis)
72.7
CADM
-
1 C.3.1.23. Van Birgelen (1995)
Type:
Rat
Dose:
0, 13.5, 26.4, 46.9, 320, 1024 ng/kg-
day
Strain:
Sprague Dawley
Route:
Oral gavage
Body weight:
150 g
Regime:
Once per day for 13 weeks
Sex:
Female
Simulation time:
2184 hours (13 weeks)
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
13.5
Emond
3,969
6,098
@ 2,160 hours)
4,665
CADM
-
26.4
Emond
(.4"
'<¦)
IU.25S ( a 2.1(i<) Ikuiis)
"45"
CADM
-
-
46.9
Emond
'VKiX
l<>.2X4 ( a 2.1(>u Ikuiis)
MM ^
CADM
-
-
320
Emond
4~.(>H(i
Xu Ikuiis)
52.5X1
CADM
-
-
1024
Emond
1
<2u
25X.,)|(> ( a 2.1(>u Ikuiis)
151.(.X()
CADM
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-162 DRAFT—DO NOT CITE OR QUOTE
-------
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
13.5
Emond
1,655
2,208 (@ 2,164 hours)
2,107
CADM
-
-
-
26.4
Emond
3,228
4,216 (@ 2,164 hours)
4,017
CADM
-
-
-
46.9
Emond
5,719
7,366 (@ 2,164 hours)
7,008
CADM
-
-
-
320
Emond
38,484
47,999 (@ 2,164 hours)
45,537
CADM
-
-
-
1024
Emond
121,640
150,410 (@2,164 hours)
142,510
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
13.5
Emond
669
843 (@ 2,167 hours)
835
CADM
-
-
-
26.4
Emond
1,092
1,357 (@ 2,166 hours)
1,342
CADM
-
-
-
46.9
Emond
1,680
2,071 (@ 2,166 hours)
2,045
CADM
-
-
-
320
Emond
8,027
9,816 (@ 2,165 hours)
9,639
CADM
-
-
-
1024
Emond
23,234
28,519 (@ 2,165 hours)
27,954
CADM
-
-
-
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
13.5
Emond
132
173 (@ 2,164 hours)
167
CADM
-
-
-
26.4
Emond
240
308 (@ 2,164 hours)
296
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-163 DRAFT—DO NOT CITE OR QUOTE
-------
46.9
Emond
404
513 (@ 2,164 hours)
492
CADM
-
-
-
320
Emond
2,437
3,031 (@ 2,164 hours)
2,887
CADM
-
-
-
1024
Emond
7,521
9,310 (@ 2,164 hours)
8,846
CADM
-
-
-
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
13.5
Emond
1QQ
24 2 m 2.1 M hours')
2^ 4
CADM
-
26.4
Emond
29.0
"4 ' ( a 1.1(4 Ikuiis)
33.2
CADM
-
46.9
Emond
s
45 (i ( a 2.1(4 Ikuiis)
4^ "
CADM
-
320
Emond
-<> 1
S5 1 ui 2.1(4 Ikuiis)
S4 1
CADM
-
1024
Emond
97.5
101 (@ 2,164 hours)
101
CADM
-
1 C.3.1.24. Vanden Heuvel et al (1994)
2
3
4
5
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.05
Emond
3.13
5.90 (@ 0 hours)
1.50
Aylward
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-164 DRAFT—DO NOT CITE OR QUOTE
Type:
Rat
Dose:
0.05, 0.1, 1, 10, 100, 1000, 10000 ng/kg-day
Strain:
Sprague Dawley
Route:
Oral gavage
Body weight:
10 weeks old
(BW 225 to 275g)
(BW=250g)
Regime:
Single dose
Sex:
Female
Simulation time:
24 hours*
al week is the minimum that can be simulated with the Aylward model
-------
0.1
Emond
6.21
11.9 (@ 0 hours)
2.97
Aylward
-
-
1
Emond
5XU
1 IS ( (i o hum's)
2S ()
Aylward
10
Emond
4S4
1. IX' ( a o hum's)
2U
Aylward
-
100
Emond
v5<.<>
1 1 .')(>' ( (i (i hum's)
l,o22
Aylward
-
-
1,000
Emond
I I'J.Xiii) ( a () hours)
,).,)S4
Aylward
10,000
Emond
24(1. <•(>()
0 hum's)
"2. (No
Aylward
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg)
Model
Metric
Time-weighted Ave
Max
Terminal
0.05
Emond
0.230
0.311 (@ 3 hours)
0.114
Aylward
0.0140
0.1
Emond
ii 4<>5
(i (>24 ( a ' hum's)
0.232
Aylward
0.0320
1
Emond
5.04
(> '4 ( a 4 hum's)
2.61
Aylward
0.950
10
Emond
59.7
(>" ') ( a 5 hum's)
34.0
Aylward
52.7
100
Emond
733
XOO ( (1 S hums)
477
Aylward
1,342
1,000
Emond
S.2I5
S.'HS id |() hum's)
5,941
Aylward
15,967
10,000
Emond
84,520
91,628 (@ 11 hours)
64,335
Aylward
162,773
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg)
Model
Metric
Time-weighted Ave
Max
Terminal
0.05
Emond
0.137
0.261 (@ 83 hours)
0.259
Aylward
0.780
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-165 DRAFT—DO NOT CITE OR QUOTE
-------
0.1
Emond
0.272
0.518 (@ 84 hours)
0.515
Aylward
1.57
1
Emond
: —
4 l><> ( a S5 Ikuiis)
4.86
Aylward
15.3
10
Emond
:: (i
41 4 ( a X1) Ikuiis)
41.0
Aylward
125
100
Emond
m
( a SS Ikuiis)
288
Aylward
739
1,000
Emond
1.'54
1 .'>(>5 ( l> Ikuiis)
1,824
Aylward
5,779
10,000
Emond
i:.5"i
15.5'J' ( a 4(1 Ikuiis)
13,735
Aylward
55,825
BODY BURDEN (ng/kg)
Dose
(ng/kg)
Model
Metric
Time-weighted Ave
Max
Terminal
0.05
Emond
0.0267
0.028 (@ 9 hours)
0.0272
Aylward
-
0.0450
0.1
Emond
0 1)5^4
ii <>5<> ( a '> ImiiiM
0.0542
Aylward
0.0900
1
Emond
u 5^:
(i 5<> 1 ( a ') Ikuiis)
0.531
Aylward
-
0.900
10
Emond
5 2')
5 5l> ( a S Ikuiis)
5.02
Aylward
9.00
100
Emond
53.0
5<> ' ( a ~ Ikuiis)
46.1
Aylward
-
90.0
1,000
Emond
52"
5(>2 ( a " Ikuiis)
424
Aylward
-
900
10,000
Emond
5,258
5,610 (@ 7 hours)
4,082
Aylward
-
9,000
BOUND LIVER (ng/kg)
Dose
(ng/kg)
Model
Metric
Time-weighted Ave
Max
Terminal
0.05
Emond
0.0192
0 (@ 4 hours)
0.00963
Aylward
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-166 DRAFT—DO NOT CITE OR QUOTE
-------
0.1
Emond
0.0380
0 (@ 4 hours)
0.0191
Aylward
-
1
Emond
U ^51
1 ( a ' Ikuiis)
I) ISO
Aylward
-
-
10
Emond
: "5
4 ( a Ikuiis)
1 4X
Aylward
-
100
Emond
i(> i
2<> ( a 2 Ikuiis)
<> 4S
Aylward
-
-
1,000
Emond
5~ ~
((i 2 Ikuiis)
40 "
Aylward
-
10,000
Emond
|uu
|u" ( a 2 Ikuiis)
•jo 4
Aylward
-
-
1 C.3.1.25. White et al. (1986)
Type:
Mice
Dose:
10, 50, 100, 500, 1000, 2000 ng/kg-day
Strain:
B6C3F1
Route:
Oral gavage
Body weight:
7 weeks old (BW set
to 23g)
Regime:
1/day for 14 days
Sex:
Female
Simulation time:
336 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
10
Emond
603
1,502 (@312 hours)
785
CADM
-
50
Emond
CADM
2.25H
(>. 'X~ ( a ^12 Ikuiis)
2."42
100
Emond
CADM
1.T4
1 1 ( a ^12 Ikuiis)
4,(.5(i
500
Emond
CADM
14.—2
5v 1XX ( a ^12 Ikuiis)
l<..^4
1,000
Emond
CADM
2<>.X44
1 <>2/>(•<> ( a ^12 Ikuiis)
2'J.22'J
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-167 DRAFT—DO NOT CITE OR QUOTE
-------
2,000
Emond
49,896
201,110 (@312 hours)
53,697
CADM
-
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
10
Emond
216
375 (@317 hours)
343
CADM
217
468 (336h)
463
50
Emond
1,279
2,164 (@317 hours)
1,997
CADM
1,775
3,261 (336h)
3,261
100
Emond
2,707
4,525 (@ 317 hours)
4,184
CADM
3,999
6,923 (336h)
6,923
500
Emond
14,802
24,165 (@317 hours)
22,383
CADM
22,705
36,362 (336h)
36,362
1,000
Emond
30,278
49,034 (@317 hours)
45,414
CADM
46,309
73,145 (336h)
73,145
2,000
Emond
61,381
98,703 (@317 hours)
91,363
CADM
93,577
146,695 (336h)
146,695
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
10
Emond
279
507 (@ 336 hours)
507
CADM
316
537 (336h)
537
50
Emond
1,056
1,846 (@ 336 hours)
1,846
CADM
1,029
1,564 (336h)
1,564
100
Emond
1,854
3,195 (@ 333 hours)
3,195
CADM
1,662
2,470 (336h)
2,470
500
Emond
7,008
11,868 (@324 hours)
11,816
CADM
5,711
8,594 (336h)
8,594
1,000
Emond
12,746
21,566 (@ 323 hours)
21,424
CADM
10,498
15,993 (336h)
15,993
2,000
Emond
23,691
40,177 (@ 322 hours)
39,843
CADM
19,990
30,726 (336h)
30,726
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-168 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
10
Emond
37.7
65.9 (@317 hours)
63.8
CADM
47.9
85.9 (336h)
85.9
50
Emond
175
297 (@317 hours)
284
CADM
207
342 (336h)
342
100
Emond
338
570 (@316 hours)
542
CADM
388
624 (336h)
624
500
Emond
1,597
2,637 (@ 316 hours)
2,480
CADM
1,761
2,754 (336h)
2,754
1,000
Emond
3,137
5,153 (@316 hours)
4,830
CADM
3,455
5,387 (336h)
5,387
2,000
Emond
6,186
10,118 (@316 hours)
9,459
CADM
6,836
10,643 (336h)
10,643
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
10
Emond
3.49
5.32 (@316 hours)
4.82
CADM
-
50
Emond
1 1 4
l(> 4 ( a ' 1" Ikuiis)
15 1
CADM
100
Emond
IS 1
25 1 < <-/ ' 1" Ikuiis)
4
CADM
500
Emond
44:
5<> lid ' 1" Ikuiis)
5 i S
CADM
1,000
Emond
"I "¦> ( (1 ' 1 ~ IkiIII's)
<.<> "
CADM
2,000
Emond
"4 4
S(i 1 ( a 11" Ikuiis)
X4 ^
CADM
1
2
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-169 DRAFT—DO NOT CITE OR QUOTE
-------
1 C.3.2. Gestational Studies
2 C.3.2.1. Bell et aL (2007)
Type:
Rat
Dose:
2.4, 8, and 46 ng/kg-day with a 0.03 ng/kg-day
background
Strain:
Han/Wistar
Route:
Dietary
Body weight:
6 weeks
(BW= 85g)
Regime:
12 weeks prior to mating, during the two week mating
period, and during gestation
Sex:
Female
Simulation
time:
2,352 hr (98 days) prior to gestation + 504 hr (21 days)
during gestation for a total of 2,856 hours
aTime averages are computed during the gestation period only.
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2.43
1,998
4,977,500
2,452 (@2,352
hours)
1,745
8.03
4,539
11,602,000
5,781 (@2,352
hours)
4,023
46.03
15,952
41,518,000
22,096 (@ 2,352
hours)
14,275
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2.43
381
914,700
437 (@ 2,356 hours)
321
8.03
1,201
2,970,500
1,351 (@ 2,356
hours)
1,044
46.03
6,638
16,802,000
7,260 (@ 2,356
hours)
5,980
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2.43
233
585,680
263 (@ 2,336 hours)
211
8.03
528
1,365,300
589 (@2,335 hours)
487
46.03
1,851
4,885,900
2,039 (@ 2,334
hours)
1,739
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-170 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2.43
43.0
94,428
44.5 (@ 2,836 hours)
43.4
8.03
113
258,160
118 (@ 2,836 hours)
114
46.03
506
1,204,800
529 (@ 2,836 hours)
509
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2.43
17.2
8,674
39.7 (@ 2,530 hours)
6.53
8.03
37.7
19,002
86.7 (@ 2,529 hours)
14.4
46.03
118
59,628
271 (@ 2,527 hours)
45.9
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2.43
8.13
20,295
8.98 (@ 2,356 hours)
7.24
8.03
16.8
43,248
18.2 (@ 2,356 hours)
15.4
46.03
42.7
112,990
44.7 (@ 2,356 hours)
40.5
1 C.3.2.2. Hojo et al. (2002)
Type:
Rat
Dose:
20, 60 and 180 ng/kg
Strain:
Sprague Dawley
Route:
Oral gavage
Body weight
20 ng/kg BW = 27 lg
60 ng/kg BW = 275g
180 ng/kg BW = 262g
Regime:
Single dose on GD8
Sex:
Female
Simulation time
24 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
1,285
177,790
3,534 (@ 192 hours)
402
60
3,295
452,060
10,477 (@ 192 hours)
1,002
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-171 DRAFT—DO NOT CITE OR QUOTE
-------
180
8,465
1,114,200
31,887 (@ 192 hours)
2,396
LIVER CONCENTRATIONS (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
128
20,554
144 (@ 198 hours)
43.2
60
420
72,340
465 (@ 200 hours)
147
180
1,364
250,820
1,497 (@ 201 hours)
497
FAT CONCENTRATIONS (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
32.5
17,253
63.0 (@ 281 hours)
49.4
60
86.4
44,093
161 (@ 284 hours)
124
180
226
108,730
398 (@ 286 hours)
301
BODY BURDEN (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
10.6
3,054
11.3 (@ 200 hours)
8.67
60
31.8
8,702
33.8 (@ 199 hours)
23.6
180
95.0
24,747
101 (@ 199 hours)
63.4
FETUS (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
15.9
2,334
18.4 (@ 206 hours)
1.64
60
39.8
5,829
45.7 (@ 205 hours)
4.10
180
96.3
13,866
110 (@ 203 hours)
9.72
BOUND LIVER (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
4.88
759
7.74 (@ 194 hours)
1.75
60
11.2
1,848
18.5 (@ 194 hours)
4.26
180
23.6
4,157
38.5 (@ 193 hours)
9.65
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-172 DRAFT—DO NOT CITE OR QUOTE
-------
1 C.3.2.3. Ikeda et al. (2005)
Type:
Rat
Dose:
400 ng/kg single dose and 80 ng/kg weekly
maintenance dose
Strain:
Sprague Dawley
Route:
Oral gavage
Body weight:
10 weeks
(BW= 25 Og)
Regime:
Initial single loading dose, 2 weekly maintenance
doses prior to gestation and 2 weekly maintenance
doses during gestation
Sex:
Female
Simulation
time:
504 hr (21 days) prior to gestation + 504 hr (21 days)
during gestation for a total simulation of 1,008 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
16.5
18,103
18,249,000
80,047 (@ 144 hours)
8,009
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
16.5
7,755
7,817,300
17,016 (@ 150 hours)
2,698
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
16.5
2,087
2,103,900
3,663 (@ 184 hours)
1,028
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
16.5
548
552,590
1,085 (@ 149 hours)
262
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
16.5
45.9
46,290
245 (@ 679 hours)
30.2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-173 DRAFT—DO NOT CITE OR QUOTE
-------
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
16.5
44.0
44,361
63.8 (@ 149 hours)
26.8
1 C.3.2.4. Kattainen et al. (2001)
Type:
Rat
Dose:
30, 100, 300, and 1,000 ng/kg
Strain:
Han/Wistar (Kuopio)
and Long/Evans
(Turku/AB) crossing.
Route:
Oral gavage
Body weight:
BW not specified
(BW set to 190g)*
Regime:
Single dose on GDI5
Sex:
Female
Simulation
time:
24 hours
aDerelanko and Hollinger (1995).
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
1,763
151,690
4,703 (@ 336 hours)
632
100
4,944
423,680
15,679 (@ 336 hours)
1,761
300
12,712
1,054,600
47,253 (@ 336 hours)
4,327
1,000
37,039
2,878,700
158,470 (@ 336 hours)
11,429
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
193
19,784
219 (@ 342 hours)
78.9
100
713
79,889
793 (@ 344 hours)
324
300
2,298
276,990
2,533 (@ 345 hours)
1,150
1,000
8,054
1,032,300
8,830 (@ 345 hours)
4,412
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-174 DRAFT—DO NOT CITE OR QUOTE
-------
FAT CONCENTRATIONS (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
42.8
12,439
82.8 (@ 426 hours)
77.5
100
123
34,712
230 (@ 431 hours)
217
300
327
86,670
571 (@ 431 hours)
536
1,000
981
238,680
1,551 (@425 hours)
1,435
BODY BURDEN (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
15.9
2,562
16.9 (@ 343 hours)
14.1
100
52.7
8,273
56.2 (@ 343 hours)
44.2
300
158
24,176
168 (@ 343 hours)
125
1,000
524
78,767
561 (@ 343 hours)
395
FETUS (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
4.86
828
6.90 (@ 372 hours)
2.53
100
13.2
2,221
18.2 (@ 372 hours)
6.89
300
31.5
5,200
42.3 (@ 371 hours)
16.2
1,000
82.2
12,907
106 (@ 369 hours)
39.6
BOUND LIVER (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
6.58
634
10.7 (@ 338 hours)
2.73
100
15.8
1,642
26.3 (@ 338 hours)
7.28
300
31.6
3,538
50.6 (@ 337 hours)
16.3
1,000
57.1
7,095
80.1 (@ 337 hours)
34.8
1 C.3.2.5. Keller et al. (2007)
Type:
Mouse
Dose:
10, 100, and 1000 ng/kg
Strain:
CBA/J and C3H/HeJ
Route:
Oral
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-175 DRAFT—DO NOT CITE OR QUOTE
-------
Body weight:
Not specified (24 g
used in the simulation)
Regime:
Single dose on GDI3
Sex:
Female
Simulation time:
504 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
10
296
18,384
788 (@ 312 hours)
48.4
100
2,365
149,060
7,884 (@312 hours)
374
1,000
18,764
1,083,900
78,825 (@312 hours)
2,454
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
10
30.6
2,046
39.8 (@316 hours)
4.90
100
371
28,867
421 (@319 hours)
62.7
1,000
4,214
388,320
4,697 (@ 321 hours)
833
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
10
22.4
7,075
41.1 (@386 hours)
35.9
100
188
57,462
333 (@ 396 hours)
291
1,000
1,591
425,300
2,441 (@ 392 hours)
2,064
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
10
5.57
1,024
5.99 (@319 hours)
4.99
100
54.3
9,170
59.0 (@318 hours)
41.9
1,000
530
79,818
581 (@318 hours)
323
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-176 DRAFT—DO NOT CITE OR QUOTE
-------
FETUS (ng/kg) andAVC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
10
2.57
386
3.80 (@337 hours)
0.795
100
21.8
3,109
30.0 (@ 334 hours)
6.42
1,000
179
22,097
233 (@ 329 hours)
42.6
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
10
1.74
115
3.14 (@315 hours)
0.305
100
11.5
857
23.5 (@314 hours)
2.30
1,000
46.7
4,430
79.8 (@314 hours)
13.3
C.3.2.6. Li et al. (2006) 3-Day
Type:
Mouse
Dose:
2, 50, and 100 ng/kg-day
Strain:
NIH
Route:
Oral gavage
Body weight:
25-28 g (used 27 g in
the simulation)
Regime:
Daily exposure from GD1 to GD3
Sex:
Female
Simulation time:
72 hours
2
3
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2
87.5
6,305
216 (@ 48 hours)
75.1
50
1,564
112,720
4,906 (@ 48 hours)
1,312
100
2,823
203,490
9,547 (@ 48 hours)
2,313
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2
8.98
647
15.1 (@ 52 hours)
9.10
50
333
23,971
539 (@ 53 hours)
402
100
718
51,738
1,156 (@ 53 hours)
888
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-177 DRAFT—DO NOT CITE OR QUOTE
-------
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2
17.0
1,227
31.1 (@ 72 hours)
31.1
50
315
22,704
548 (@ 72 hours)
548
100
576
41,460
984 (@ 72 hours)
984
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2
2.29
165
3.51 (@ 55 hours)
3.43
50
53.6
3,863
82.2 (@ 54 hours)
77.1
100
105
7,598
162 (@ 53 hours)
150
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2
0.00
0
0.000 (@ 72 hours)
0.00
50
0.0
0
0.000 (@ 72 hours)
0.00
100
0.0
0
0.000 (@ 72 hours)
0.00
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
2
0.538
38.8
0.864 (@51 hours)
0.498
50
8.24
594
13.5 (@ 2 hours)
8.16
100
13.6
981
23.7 (@ 2 hours)
13.6
1
2
3 C.3.2.7. Markowski el ill. (2001)
Type:
Rat
Dose:
20, 60 and 180 ng/kg
Strain:
Holtzman rats
Route:
Oral gavage
Body weight:
BW not specified
(BW set to 190g)*
Regime:
Single dose on GDI8
Sex:
Female
Simulation
time:
24 hours
aDerelanko and Hollinger (1995).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-178 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
1,234
71,255
3,029 (@ 408 hours)
471
60
3,184
184,690
9,096 (@ 408 hours)
1,317
180
8,152
465,030
27,457 (@ 408 hours)
3,193
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
123
8,315
142 (@ 414 hours)
56.5
60
409
29,656
459 (@ 415 hours)
213
180
1,333
103,210
1,478 (@ 416 hours)
790
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
28.0
4,437
55.6 (@ 498 hours)
55.5
60
74.0
11,462
144 (@ 504 hours)
144
180
195
28,948
363 (@ 504 hours)
363
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
10.6
1,013
11.3 (@ 415 hours)
10.2
60
31.7
2,989
33.7 (@415 hours)
29.5
180
94.7
8,834
101 (@ 415 hours)
85.7
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
1.26
157
1.93 (@ 448 hours)
1.43
60
3.21
395
4.79 (@ 449 hours)
3.63
180
7.80
943
11.3 (@ 449 hours)
8.69
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-179 DRAFT—DO NOT CITE OR QUOTE
-------
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
20
4.75
299
7.61 (@410 hours)
2.12
60
11.0
729
18.2 (@410 hours)
5.47
180
23.2
1,621
38.1 (@ 409 hours)
12.9
1 C.3.2.8. Mietinnen et al. (2006)
Type:
Rat
Dose:
30, 100, 300 and 1000 ng/kg
Strain:
cross-breeding of
Han/Wistar and Long-
Evans rats
Route:
Oral gavage
Body weight:
BW 11 weeks
(BW set to 180g)
Regime:
Single dose on GDI5
Sex:
Female
Simulation
time:
24 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
1,756
151,180
4,641 (@ 336 hours)
721
100
4,922
422,480
15,471 (@ 336 hours)
1,758
300
12,657
1,052,000
46,647 (@ 336 hours)
4,994
1,000
36,874
2,872,800
156,480 (@ 336 hours)
11,423
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
193
19,697
219 (@ 342 hours)
78.8
100
711
79,610
791 (@ 344 hours)
323
300
2,293
276,280
2,529 (@ 345 hours)
1,149
1,000
8,044
1,030,600
8,822 (@ 345 hours)
4,409
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-180 DRAFT—DO NOT CITE OR QUOTE
-------
FAT CONCENTRATIONS (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
43.1
12,461
82.9 (@ 425 hours)
77.4
100
124
34,793
231 (@ 430 hours)
217
300
329
86,906
572 (@ 430 hours)
536
1,000
988
239,390
1,555 (@ 424 hours)
1,436
BODY BURDEN (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
15.9
2,560
16.9 (@ 343 hours)
14.1
100
52.7
8,269
56.2 (@ 343 hours)
44.1
300
158
24,169
168 (@ 343 hours)
125
1,000
524
78,769
561 (@ 343 hours)
395
FETUS (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
4.83
824
6.87 (@ 372 hours)
2.52
100
13.1
2,213
18.1 (@372 hours)
6.87
300
31.3
5,182
42.1 (@371 hours)
16.2
1,000
81.7
12,867
105 (@ 369 hours)
39.5
BOUND LIVER (ng/kg) and AUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
30
6.56
632
10.7 (@ 338 hours)
2.72
100
15.8
1,639
26.3 (@ 338 hours)
7.27
300
31.6
3,533
50.5 (@ 337 hours)
16.3
1,000
57.0
7,090
80.1 (@ 337 hours)
34.8
1 C.3.2.9. Murray et al. (1979) Gestational Portion
Type:
Rat
Dose:
1,10, and 100 ng/kg-day
Strain:
Sprague Dawley
Route:
Diet oral dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-181 DRAFT—DO NOT CITE OR QUOTE
-------
Body weight:
6- to 7 week
(Bw= 85g)
Regime:
Once per day for 90 days prior to gestation and
during gestation
Sex:
Female
Simulation
time:
2160 hr (90 days ) prior gestation + 504 hr (21
days) for a total simulation of 2664 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
1
897
2,389,400
1,291 (@ 2,160 hours)
926
10
4,691
12,497,000
6,780 (@ 2,160 hours)
4,708
100
26,219
69,849,000
42,272 (@ 2,160 hours)
25,849
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
1
129
342,940
186 (@ 2,164 hours)
133
10
1,271
3,385,700
1,657 (@ 2,164 hours)
1,298
100
12,492
33,279,000
15,332 (@ 2,164 hours)
12,876
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
1
105
280,460
142 (@ 2,146 hours)
112
10
551
1,467,700
682 (@ 2,143 hours)
569
100
3,080
8,204,300
3,682 (@ 2,142 hours)
3,162
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
1
15.4
41,059
21.8 (@2,644 hours)
21.4
10
108
286,920
141 (@ 2,644 hours)
137
100
847
2,257,100
1,060 (@ 2,644 hours)
1,017
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-182 DRAFT—DO NOT CITE OR QUOTE
-------
FETUS (ng/kg) andAVC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
1
1.77
4,720
21.7 (@ 2,3 3 9 hours)
3.54
10
8.22
21,889
99.8 (@ 2,337 hours)
16.7
100
37.4
99,722
453 (@ 2,334 hours)
77.0
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
1
3.79
10,101
5.06 (@ 2,163 hours)
3.96
10
17.1
45,522
20.5 (@ 2,164 hours)
17.6
100
55.1
146,790
61.0 (@ 2,164 hours)
56.8
1 C.3.2.10. Murray et al. (1979) Adult Portion
Type:
Rat
Dose:
1,10, and 100 ng/kg-day
Strain:
Sprague Dawley
Route:
Dietary
Body weight:
BW set to 4.5 g
Regime:
120 days
Sex:
Female
Simulation time:
2880 hours
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
1
Emond
619
832 (@ 2,856 hours)
785
CADM
-
10
Emond
3,241
4,181 (@ 2,856 hours)
3,717
CADM
-
100
Emond
18,038
24,433 (@ 2,856 hours)
19,844
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-183 DRAFT—DO NOT CITE OR QUOTE
-------
LIVER CONCENTRATIONS (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1
Emond
128
180 (@ 2,859 hours)
173
CADM
-
10
Emond
i.:-'
l.t.ISi (i 2.8
(>(>5 ( a 2.8(4 Ikuiis)
<>5~
CADM
-
100
Emond
\(>'J5
\<>(>4 ( a 2.8(i2 Ikuiis)
V5U
CADM
-
BODY BURDEN (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1
Emond
148
20 0 (ia)2.^Ct hours')
1 gfi
CADM
-
10
Emond
It >5
1 'ti ( a 2.8
CADM
-
100
Emond
837
1 .tin' ( a 2.X(i() Ikmii's)
l>5~
CADM
-
BOUND LIVER (ng/kg)
Dose
Model
Metric
(ng/kg-day)
Adjusted dose
Time-weighted Ave
Max
Terminal
1
Emond
3.77
4.95 (@ 2,859 hours)
4.77
CADM
-
10
Emond
r i
2d ' ( a 2.85l> Ikuiis)
I<> 5
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-184 DRAFT—DO NOT CITE OR QUOTE
-------
100
Emond
55.3
60.9 (@ 2,860 hours)
59.4
CADM
ff:::::::::::::::::::-™:::::::::::::::::::::
-
1 C.3.2.11. Nohara et al. (2000)
Type:
Rat
Dose:
12.5, 50, 200 or 800 ng TCDD/kg
Strain:
Holtzman rats
Route:
Oral gavage
Body weight:
BW not specified
(BW set to 190g)*
Regime:
Single dose on GDI5
Sex:
Female
Simulation
time:
24 hours
aDerelanko and Hollinger (1995).
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
816
69,459
1,933 (@ 336 hours)
290
50
2,724
235,070
7,736 (@ 336 hours)
981
200
8,912
752,170
31,022 (@ 336 hours)
3,110
800
30,121
2,378,900
125,030 (@ 336 hours)
9,532
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
73.9
7,084
86.2 (@341 hours)
28.3
50
336
35,736
378 (@ 343 hours)
143
200
1,492
175,300
1,651 (@ 344 hours)
722
800
6,387
810,340
7,011 (@ 345 hours)
3,449
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
19.7
5,736
38.1 (@419 hours)
35.4
50
67.6
19,362
129 (@ 427 hours)
121
200
229
62,032
410 (@ 431 hours)
385
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-185 DRAFT—DO NOT CITE OR QUOTE
-------
800
803
197,830
1,288 (@ 425 hours)
1,194
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
6.63
1,088
7.05 (@ 343 hours)
6.10
50
26.4
4,212
28.1 (@ 343 hours)
22.9
200
105
16,259
112 (@343 hours)
85.1
800
420
63,228
449 (@ 343 hours)
319
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
2.25
385
3.26 (@371 hours)
1.17
50
7.43
1,263
10.5 (@ 372 hours)
3.89
200
22.8
3,802
31.0 (@372 hours)
11.8
800
68.1
10,862
88.5 (@ 369 hours)
33.6
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
3.24
298
5.12 (@338 hours)
1.27
50
9.66
959
16.0 (@ 338 hours)
4.18
200
24.8
2,695
40.7 (@ 337 hours)
12.2
800
51.9
6,315
75.0 (@ 337 hours)
30.6
1 C.3.2.12. Ohsako et al. (2001)
Type:
Rat
Dose:
12.5, 50, 200, and 800 ng/kg-day
Strain:
Holtzmann
Route:
Oral gavage
Body weight
10 weeks (200g)
Regime:
Single dose on GDI5
Sex:
Female
Simulation time
24 hours
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-186 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
845
63,918
2,016 (@360 hours)
304
50
2,763
212,870
7,928 (@ 360 hours)
1,020
200
9,022
677,090
31,557 (@360 hours)
3,239
800
30,504
2,148,100
127,220 (@ 360 hours)
9,983
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
76.8
6,595
89.0 (@ 365 hours)
30.1
50
340
32,557
383 (@ 367 hours)
152
200
1,504
157,600
1,657 (@ 368 hours)
768
800
6,426
724,530
7,026 (@ 369 hours)
3,689
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
19.6
4,897
38.4 (@ 446 hours)
36.9
50
65.8
16,240
128 (@ 455 hours)
124
200
223
51,709
404 (@ 458 hours)
393
800
780
165,660
1,270 (@453 hours)
1,224
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
6.84
966
7.24 (@ 367 hours)
6.38
50
26.6
3,693
28.4 (@ 367 hours)
23.7
200
106
14,210
112 (@ 367 hours)
88.3
800
421
55,466
449 (@ 367 hours)
334
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
1.69
274
2.50 (@ 397 hours)
1.16
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-187 DRAFT—DO NOT CITE OR QUOTE
-------
50
5.48
881
7.91 (@ 398 hours)
3.79
200
16.8
2,629
23.3 (@ 398 hours)
11.4
800
50.2
7,518
66.4 (@ 396 hours)
32.3
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
12.5
3.34
274
5.25 (@ 362 hours)
1.33
50
9.76
863
16.1 (@362 hours)
4.34
200
25.0
2,396
40.7 (@ 361 hours)
12.7
800
52.1
5,566
75.1 (@361 hours)
31.7
C.3.2.13. Schantzetal. (1995) and Amin et al. (2000)
Type:
Rat
Dose:
25 and 100 ng/kg-day
Strain:
Sprague Dawley
Route:
Oral gavage
Body weight:
BW not specified
(BW set to 25 Og)
Regime:
Daily doses from GD 10-16
Sex:
Female
Simulation time:
384 hours; time averages are calculated
from the beginning of the dosing
4
5
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
2,670
384,750
6,800 (@ 360 hours)
3,190
100
8,341
1,201,700
24,522 (@ 360 hours)
9,706
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
512
73,705
871 (@ 365 hours)
778
100
2,371
341,460
4,009 (@ 366 hours)
3,662
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-188 DRAFT—DO NOT CITE OR QUOTE
-------
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
169
24,329
307 (@ 384 hours)
307
100
532
76,559
949 (@ 384 hours)
949
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
45
6,492
76.6 (@ 365 hours)
74
100
176
25,401
298 (@ 365 hours)
287
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
25
3,628
30.4 (@ 343 hours)
27
100
74.0
10,655
88.1 (@ 342 hours)
77.8
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
10
1,440
14.4 (@ 364 hours)
13
100
25
3,628
34.2 (@ 364 hours)
32
1
2
3 C.3.2.14. Seoetal. (1995)
Type:
Rat
Dose:
25 and 100 ng/kg-day
Strain:
Sprague Dawley
Route:
Oral gavage
Body weight:
BW not specified
(BW set to 190g)
Regime:
Daily from GD 10-16
Sex:
Female
Simulation time:
384 hours; time averages are calculated
from the beginning of the dosing
4
5
6
7
8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-189 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
2,655
766,430
6,796 (@ 384 hours)
2,748
100
8,319
2,372,700
24,284 (@ 384 hours)
8,333
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
506
163,400
972 (@ 389 hours)
606
100
2,358
767,640
4,486 (@ 389 hours)
2,871
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
173
66,734
358 (@ 436 hours)
339
100
545
207,420
1,105 (@433 hours)
1,037
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
45.3
16,124
87.5 (@ 389 hours)
73.6
100
177
61,908
339 (@ 389 hours)
271
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
24.7
5,826
29.8 (@ 343 hours)
10.6
100
72.6
16,930
86.6 (@ 342 hours)
30.2
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
25
9.92
2,937
15.4 (@ 388 hours)
11.0
100
25.1
7,349
36.1 (@ 388 hours)
27.7
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-190 DRAFT—DO NOT CITE OR QUOTE
-------
1 C.3.2.15. Shi et al. (2007) Gestational Portion
Type:
Rat
Dose:
1, 5, 50 and 200 ng/kg
Strain:
Sprague Dawley
Route:
Oral gavage
Body weight:
BW not specified
(BW set to 190g)*
Regime:
Single dose on GDI4 and GD21
Sex:
Female
Simulation time:
504 hours
aDerelanko and Hollinger (1995).
BLOOD CONCENTRATIONS (ng/kg) (Serum lipid adjusted) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
0.143
17.9
9,014
173 (@ 480 hours)
74.7
0.714
81.1
40,871
840 (@ 480 hours)
329
7.14
621
312,880
8,016 (@480 hours)
2,310
28.6
1,975
995,020
31,730 (@312 hours)
6,960
LIVER CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
0.143
1.16
583
8.44 (@ 484 hours)
5.63
0.714
6.87
3,462
46.8 (@ 485 hours)
35.2
7.14
96.9
48,840
576 (@ 486 hours)
499
28.6
465
234,480
2,581 (@487 hours)
2,328
FAT CONCENTRATIONS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
0.143
1.31
662
5.66 (@ 504 hours)
5.66
0.714
6.02
3,032
25.2 (@ 504 hours)
25.2
7.14
46.9
23,608
188 (@ 504 hours)
188
28.6
150
75,504
591 (@ 504 hours)
591
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-191 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
0.143
0.229
116
1.08 (@ 487 hours)
1.07
0.714
1.12
565
5.32 (@ 487 hours)
5.23
7.14
10.7
5,389
50.8 (@ 487 hours)
49.4
28.6
41.3
20,788
196 (@ 487 hours)
190
FETUS (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
0.143
0.103
52.0
0.430 (@ 343 hours)
0.151
0.714
0.470
237
1.91 (@344 hours)
0.681
7.14
3.53
1,781
13.8 (@ 345 hours)
5.04
28.6
10.6
5,354
41.0 (@ 345 hours)
15.1
BOUND LIVER (ng/kg) andAUC ((ng/kg) • hr)
Dose
(ng/kg-day)
Adjusted dose
Metric
Time-weighted Ave
Area Under the
Curve
Max
Terminal
0.143
0.0780
39.3
0.566 (@483 hours)
0.341
0.714
0.348
175
2.31 (@483 hours)
1.49
7.14
2.44
1,231
16.0 (@ 314 hours)
9.67
28.6
6.67
3,360
40.8 (@313 hours)
24.8
C.3.2.16. Shi et al. (2007) Adult Portion
Type:
Rat
Dose:
1, 5, 50 and 200 ng/kg
Strain:
Sprague Dawley
Route:
Oral gavage
Body weight:
BW set to 4.5 g
Regime:
Weekly doses for 11 months
Sex:
Female
Simulation time:
8040 hours
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-192 DRAFT—DO NOT CITE OR QUOTE
-------
BLOOD CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.143
Emond
188
262 (@ 7,561 hours)
210
CADM
-
0.714
Emond
5l>2
S44 ( a ~.5<><> Ikuiis)
(,(H
CADM
-
7.14
Emond
:.xx:
5.U2 ' ( a ~5<>u Ikuiis)
CADM
-
28.6
Emond
~.<|<0
l<>. In ' i a ".5(>(i Ikuiis)
<>.X25
CADM
-
LIVER CONCENTRATIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.143
Emond
26.1
36.5 (@ 7,564 hours)
29.6
CADM
-
-
-
0.714
Emond
118
159 (@ 7,564 hours)
120
CADM
-
-
-
7.14
Emond
1,068
1,415 (@ 7,565 hours)
970
CADM
-
-
-
28.6
Emond
4,119
5,450 (@ 7,565 hours)
3,574
CADM
-
-
-
FA T CONCENTRA TIONS (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.143
Emond
32.5
40.0 (@ 7,583 hours)
36.7
CADM
-
-
-
0.714
Emond
102
120 (@ 7,584 hours)
106
CADM
-
-
-
7.14
Emond
497
571 (@ 7,584 hours)
475
CADM
-
-
-
28.6
Emond
1,322
1,527 (@ 7,584 hours)
1,217
CADM
-
-
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-193 DRAFT—DO NOT CITE OR QUOTE
-------
BODY BURDEN (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.143
Emond
3.94
4.99 (@ 7,566 hours)
4.45
CADM
-
-
-
0.714
Emond
14.0
17.2 (@ 7,566 hours)
14.5
CADM
-
-
-
7.14
Emond
90.8
112 (@ 7,566 hours)
84.4
CADM
-
-
-
28.6
Emond
300
374 (@ 7,566 hours)
266
CADM
-
-
-
BOUND LIVER (ng/kg)
Dose
(ng/kg-day)
Adjusted dose
Model
Metric
Time-weighted Ave
Max
Terminal
0.143
Emond
1.18
1.60 (@ 7,563 hours)
1.31
CADM
-
0.714
Emond
4 "5 ( (1 ",5(> ' Ikuiis)
^ "0
CADM
-
7.14
Emond
15 (>
ll> " ((i ".5(4 Ikuiis)
14 "
CADM
-
28.6
Emond
33.5
4<> " ( a ".5(4 Ikuiis)
'i:
CADM
-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-194 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table C-l. Model input parameters potentially addressed by selected
2 articles
3
Articles
Model input parameters potentially addressed
Absorption
Desorption
Distribution
Elimination
Kinetics
Induction CYP1A1
Interspecies
differences
Age Differences
Aryl hydrocarbon
receptor (AhR)
Mode of action
Partition
coefficient
Aylward et al., 2004
•
•
•
•
•
Aylward et al., 2005
•
•
•
•
•
Aylward et al., 2009
•
Bohonowych and Denison, 2007
•
•
•
Boverhof et al., 2005
•
•
Connor and Aylward, 2006
•
•
•
Heinzl et al., 2007
•
•
Irigaray et al., 2005
•
•
Kerger et al., 2006
•
•
•
Kerger et al., 2007
•
Kim et al., 2003
•
Korenaga et al., 2007
•
•
Korkalainen et al., 2004
•
•
Kransler et al., 2007
•
•
Maruyama et al., 2002
•
•
•
Maruyama et al., 2003
•
•
•
Maruyama and Aoki, 2006
•
•
•
Millbrath et al., 2009
•
•
•
•
Moser and McLachlan, 2002
•
•
Mullerova and Kopecky, 2007
•
Nadal et al., 2009
•
•
Nohara et al., 2006
•
•
Olsman et al., 2007
•
Saghir et al., 2005
•
•
•
Schecter et al., 2003
•
•
Staskal et al., 2005
•
•
Toyoshiba et al., 2004
•
•
•
Wilkes et al., 2009
•
4 Partition coefficient estimates and CYP parameter value estimates were derived from Wang et al. (1997, 2000) and
5 Santostefano et al. (1998).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-195 DRAFT—DO NOT CITE OR QUOTE
-------
1 C.4. REFERENCES
2 Amin, S; Moore, RW; Peterson, RE; et al. (2000) Gestational and lactational exposure to TCDD or coplanar PCBs
3 alters adult expression of saccharin preference behavior in female rats. Neurotoxicol Teratol 22(5):675-682.
4 Aylward, LL; Brunet, RC; Carrier, G; et al. (2005a) Concentration-dependent TCDD elimination kinetics in
5 humans: toxicokinetic modeling for moderately to highly exposed adults from Seveso, Italy, and Vienna, Austria,
6 and impact on dose estimates for the NIOSH cohort. J Exp Anal Environ Epidemiol 15:51-65.
7 Aylward, LL; Brunet, RC; Starr, TB; et al. (2005b) Exposure reconstruction for the TCDD-exposed NIOSH cohort
8 using a concentration- and age-dependent model of elimination. Risk Anal 25(4):945-956.
9 Aylward, LL; Bodner, KM; Collins, JJ; et al. (2009) TCDD exposure estimation for workers at a New Zealand
10 2,4,5-T manufacturing facility based on serum sampling data. J Expo Sci Environ Epidemiol. 3 June;
11 doi:10.1038/jes.2009.31. Available online at http://www.nature.com/jes/journal/vaop/ncurrent/full/jes200931a.html.
12 Bell, DR; Clode, S; Fan, MQ; et al. (2007) Toxicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin in the developing male
13 Wistar(Han) rat. I: No decrease in epididymal sperm count after a single acute dose. Toxicol Sci (l):214-23.
14 Bohonowych, JE; Denison, MS. (2007) Persistent binding of ligands to the aryl hydrocarbon receptor.
15 Toxicological Sciences 98(1):99-109.
16 Boverhof, DR; Burgoon, LD; Tashiro, C; et al. (2005) Temporal and dose-dependent hepatic gene expression
17 patterns in mice provide new insights into TCDD-mediated hepatotoxicity. Toxicol Sci 85(2): 1048-1063.
18 Cantoni, L; Salmona, M; Rizzardini, M. (1981) Porphyrogenic effect of chronic treatment with 2,3,7,8-
19 tetrachlorodibenzo-p-dioxin in female rats. Dose-effect relationship following urinary excretion of porphyrins.
20 Toxicol Appl Pharmacol 57:156-163.
21 Carrier, G; Brunet, RC; Brodeur, J. (1995a) Modeling of the toxicokinetics of polychlorinated dibenzo-p-dioxins
22 and dibenzofuranes in mammalians, including humans: II kinetics of absorption and disposition of PCDDs/PCDFs.
23 Toxicol Appl Pharmacol 131(267):276.
24 Carrier, G; Brunet, RC; Brodeur, J. (1995b) Modeling of the toxicokinetics of polychlorinated dibenzo-p-dioxins
25 and dibenzofurans in mammalians, including humans. Toxicol Appl Pharmacol 131:253-266.
26 Chu, I; Valli, VE; Rousseaux, CG. (2007) Combined effects of 2,3,7,8-tetrachlorodibenzo-pdioxin and
27 polychlorinated biphenyl congeners in rats. Toxicol Environ Chem 89(l):71-87.
28 Connor, KT; Aylward, LL. (2006) Human response to dioxin: aryl hydrocarbon receptor (AHR) molecular structure,
29 function, and dose-response data for enzyme induction indicate an impaired human AhR. J Toxicol Environ Health
30 Part B: Critical Reviews 9(2):147-171.
31 Crofton, KM; Craft, ES; Hedge, JM; et al. (2005) Thyroid-hormone-disrupting chemicals: evidence for dose-
32 dependent additivity or synergism. Environ Health Perspect 113(11): 1549-1554.
33 Derelanko, MJ; Hollinger, MA. (1995) CRC Handbook of Toxicology, pp.1-948. New York, NY.
34 Diliberto, JJ; Burgin, D; Birnbaum, LS. (1997) Role of CYP1A2 in hepatic sequestration of dioxin: studies using
35 CYP1A2 knock-out mice. BiochemBiophys Res Commun236(2):431-433.
36 Emond, C; Birnbaum, LS; DeVito, M. (2004) Physiologically based pharmacokinetic model for developmental
37 exposures to TCDD in the rat. Toxicol Sci 80(1): 115-133.
3 8 Emond, C; Michalek, JE; Birnbaum, LS; et al. (2005) Comparison of the use of a physiologically based
3 9 pharmacokinetic model and a classical pharmacokinetic model for dioxin exposure assessments. Environ Health
40 Perspect 113(12): 1666-1668.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-196 DRAFT—DO NOT CITE OR QUOTE
-------
1 Emond, C; Birnbaum, LS; Devito, MJ. (2006) Use of a physiologically based pharmacokinetic model for rats to
2 study the influence of body fat mass and induction of CYP1A2 on the pharmacokinetics of TCDD. Environ Health
3 Perspect 114(9): 1394-1400.
4 Fattore, EE; Trossvik, C; Hakansson, H. (2000) Relative potency values derived from hepatic vitamin A reduction in
5 male and female Sprague-Dawley rats following subchronic dietary exposure to individual polychlorinated dibenzo-
6 p-dioxin and dibenzofuran congeners and a mixture thereof. Toxicol Appl Pharmacol 165(3): 184-194.
7 Hassoun, EA; Wilt, SC; Devito, MJ; et al. (1998) Induction of oxidative stress in brain tissues of mice after
8 subchronic exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol Sci 42:23-27.
9 Hassoun, EA; Li, F; Abushaban, A; et al. (2000) The relative abilities of TCDD and its congeners to induce
10 oxidative stress in the hepatic and brain tissues of rats after subchronic exposure. Toxicology 145:103-113.
11 Heinzl, H; Mittlback, M; Edler, L. (2007) On the translation of uncertainty from toxicokinetic to toxicodynamic
12 models - the TCDD example. Chemosphere 67(9):S365-S374.
13 Hojo, R; Stern, S; Zareba, G; et al. (2002) Sexually dimorphic behavioral responses to prenatal dioxin exposure.
14 Ikeda, M; Mitsui, T; Setani, K; et al. (2005) In utero and lactational exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin
15 in rats disrupts brain sexual differentiation. Toxicol Appl Pharmacol 205(1):98-105.
16 ILSI (International Life Sciences Institute). (1994) Physiological parameter values for PBPK models. Washington,
17 DC: Risk Science Institute.
18 Irigaray, P; Mejean, L; Laurent, F. (2005) Behaviour of dioxin in pig adipocytes. Food Chem Toxicol
19 43(3):457-460.
20 Kattainen, H; Tuukkanen, J; Simanainen, U; et al. (2001) In utero/lactational 2,3,7,8-tetrachlorodibenzo-p-dioxin
21 exposure impairs molar tooth development in rats. Toxicol Appl Pharmacol 174(3):216-224.
22 Keller, JM; Huet-Hudson, YM; Leamy, LJ. (2007) Qualitative effects of dioxin on molars vary among inbred mouse
23 strains. Arch Oral Biol 52(5):450-454.
24 Kerger, BD; Leung, HW; Scott, P; et al. (2006) Age- and concentration-dependent elimination half-life of 2,3,7,8-
25 tetrachlorodibenzo-p-dioxin in Seveso children. Environ Health Perspect 114(10): 1596-1602.
26 Kerger, BD; Leung, HW; Scott, PK; et al. (2007) Refinements on the age-dependent half-life model for estimating
27 child body burdens of polychlorodibenzodioxins and dibenzofurans. Chemosphere 67(9):S272-S278.
28 Kim, AH; Kohn, MC; Nyska, A; et al. (2003) Area under the curve as a dose metric for promotional responses
29 following 2,3,7,8-tetrachlorodibenzo-p-dioxin exposure. Toxicol Appl Pharmacol 191 (1): 12—21.
30 Kitchin, KT; Woods, JS. (1979) 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) effects on hepatic microsomal
31 cytochrome P-448-mediated enzyme activities. Toxicol Appl Pharmacol 47:537-546.
32 Kociba, RJ; Keeler, PA; Park, GN; et al. (1976) 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD): results of a 13-week
33 oral toxicity study in rats. Toxicol Appl Pharmacol 35:553-574.
34 Kociba, RJ; Keyes, DG; Beyer, JE; et al. (1978) Results of a two-year chronic toxicity and oncogenicity study of
35 2,3,7,8-tetrachlorodibenzo-p-dioxin in rats. Toxicol Appl Pharmacol 46(2):279-303.
36 Kociba, RJ; Keyes, DG; Beyer, JE; et al. (1979) Long-term toxicological studies of 2,3,7,8-tetrachlorodibenzo-p-
37 dioxin (TCDD) in laboratory animals. Ann NY Acad Sci 320: 397-404.
38 Korenaga, T; Fukusato, T; Ohta, M; et al. (2007) Long-term effects of subcutaneously injected 2,3,7,8-
39 tetrachlorodibenzo-p-dioxin on the liver of rhesus monkeys. Chemosphere 67(9):S399-S404.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-197 DRAFT—DO NOT CITE OR QUOTE
-------
1 Korkalainen, M; Tuomisto, J; Pohjanvirta, R. (2004) Primary structure and inducibility by 2,3,7,8-
2 tetrachlorodibenzo-p-dioxin (TCDD) of aryl hydrocarbon receptor repressor in a TCDD-sensitive and a TCDD-
3 resistant rat strain. Biochem Biophys Res Commun 315(1):123—131.
4 Kransler, KM; McGarrigle, BP; Olson, JR. (2007) Comparative developmental toxicity of 2,3,7,8-
5 tetrachlorodibenzo-p-dioxin in the hamster, rat, and guinea pig. Toxicology 229(3):214-225.
6 Krishnan, D. (2007) Neurobehavioral and neuroendocrine assessment of rats perinatally exposed to polychlorinated
7 biphenyls: a possible model for autism. Bowling Green State University.
8 Krishnan, K. (2008) Physiologically based pharmacokinetic modelling in toxicology. In: Hayes, AW; ed. Principles
9 and methods of toxicology. 5 th ed. New York, NY: CRC Press, pp. 231-291.
10 Latchoumycandane, C; Mathur, PP. (2002) Effects of vitamin E on reactive oxygen species-mediated 2,3,7,8-
11 tetrachlorodi-benzo-p-dioxin toxicity in rat testis. J Appl Toxicol 22(5):345-351
12 Latchoumycandane, C; Chitra, KC; Mathur, PP. (2002) The effect of 2,3,7,8-tetrachlorodibenzo-p-dioxin on the
13 antioxidant system in mitochondrial and microsomal fractions of rat testis. Toxicology 171(2-3): 127-135.
14 Li, B; Liu, H; Dai, L; et al. (2006) The early embryo loss caused by 2,3,7,8-tetrachlorodibenzo-p-dioxin may be
15 related to the accumulation of this compound in the uterus. Reprod Toxicol 21(3):301-306.
16 Luecke, RH; Pearce, BA; Wosilait, WD; et al. (2007) Postnatal growth considerations for PBPK modeling. J
17 Toxicol Environ Health A 70(12): 1027-37.
18 Markowski, VP; Zareba, G; Stern, S; et al. (2001) Altered operant responding for motor reinforcement and the
19 determination of benchmark doses following perinatal exposure to low-level 2,3,7,8-tetrachlorodibenzo-p-dioxin.
20 Environ Health Perspect 109(6):621-627.
21 Maruyama, W; Aoki, Y. (2006) Estimated cancer risk of dioxins to humans using a bioassay and physiologically
22 based pharmacokinetic model. Toxicol Appl Pharmacol 214(2):188-198.
23 Maruyama, W; Yoshida, K; Tanaka, T; et al. (2002) Determination of tissue-blood partition coefficients for a
24 physiological model for humans, and estimation of dioxin concentration in tissues. Chemosphere 46(7): 975-985.
25 Maruyama, W; Yoshida, K; Tanaka, T; et al. (2003) Simulation of dioxin accumulation in human tissues and
26 analysis of reproductive risk. Chemosphere 53(4):301-313.
27 Miettinen, HM; Sovari, R; Alaluusua, S; et al. (2006) The effect of perinatal TCDD exposure on caries susceptibility
28 in rats. Toxicol Sci 91(2):568-575.
29 Milbrath, MO; Wenger, Y; Chang, CW; et al. (2009) Apparent half-lives of dioxins, furans, and polychlorinated
30 biphenyls as a function of age, body fat, smoking status, and breast-feeding. Environ Health Perspect
31 -117(3):417425.
32 Moser, GA; McLachlan, MS. (2002) Modeling digestive tract absorption and desorption of lipophilic organic
33 contaminants in humans. Environ Sci Technol 36(15):3318-25.
34 Mullerova, D; Kopecky, J. (2007) White adipose tissue: storage and effector site for environmental pollutants.
35 Physiol Res 56(4):375-381.
36 Murray, FJ; Smith, FA; Nitschke, KD; et al. (1979) Three-generation reproduction study of rats given 2,3,7,8-
37 tetrachlorodibenzo-p-dioxin (TCDD) in the diet. Toxicol Appl Pharmacol 50:241-252.
3 8 Murray, TJ; Yang, X; Sherr, D. (2006) Growth of a human mammary tumor cell line is blocked by galangi, a
39 naturally occurring bioflavonoid, and is accompanied by down-regulation of cyclins D3, E, and A. Breast Cancer
40 Res 8:R17 (doi:10.1186/bcrl391)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-198 DRAFT—DO NOT CITE OR QUOTE
-------
1 Nadal, M; Perello, G; Schuhmacher, M; et al. (2008) Concentrations of PCDD/PCDFs in plasma of subjects living
2 in the vicinity of a hazardous waste incinerator: Follow-up and modeling validation. Chemosphere 73(6):901-906.
3 Nadal, M; Domingo, JL; Garcia, F; et al. (2009) Levels of PCDD/F in adipose tissue on non-occupationally exposed
4 subjects living near a hazardous waste incinerator in Catalonia, Spain. Chemosphere 74(11):1471-1476.
5 NAS (National Academy of Sciences). (2006) Health risks from dioxin and related compounds: evaluation of the
6 EPA reassessment. Washington, DC: National Academies Press. Available online at
7 http://www.nap.edu/catalog.php?record_id=l 1688.
8 Nohara, K; Fujimaki, H; Tsukumo, S; et al. (2000) The effects of perinatal exposure to low doses of 2,3,7,8-
9 tetrachlorodibenxo-p-dioxin on immune organs of rats. Toxicology 154(1-3): 123-133
10 Nohara, K; Ao, K; Miyamoto, Y; et al. (2006) Comparison of the 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)-
11 induced CYP1A1 gene expression profile in lymphocytes from mice, rats, and humans: Most potent induction in
12 humans. Toxicology 225(2-3):204-213.
13 NTP (National Toxicology Program). (1982) Bioassay of 2,3,7,8-tetrachlorodibenzo-p-dioxin for possible
14 carcinogenicity (gavage study). Tech. Rept. Ser. No. 201. U.S. Department of Health and Human Services, Public
15 Health Service, Research Triangle Park, NC.
16 NTP (National Toxicology Program). (2006) Studies of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) in female
17 Harlan Sprague-Dawley rats (gavage studies) Tech. Rep. Ser. No. 521. U.S. Department of Health and Human
18 Services, Public Health Service, Research Triangle Park, NC.
19 O'Flaherty, EJ. (1992) Modeling bone mineral metabolism, with special reference to calcium and lead.
20 Neurotoxicity 13(4):789-797.
21 Ohsako, S; Miyabara, Y; Nishimura, N; et al. (2001) Maternal exposure to a low dose of 2,3,7,8-tetrachlorodibenzo-
22 p-dioxin (TCDD) suppressed the development of reproductive organs of male rats: dose-dependent increase of
23 mRNA levels of 5alpha-reductase type 2 in contrast to decrease of androgen receptor in the pubertal ventral prostate.
24 Toxicol Sci 60(l):132-43.
25 Olsman, H; Engwall, M; Kammann, U; et al. (2007) Relative differences in aryl hydrocarbon receptor-mediated
26 response for 18 polybrominated and mixed halogenated dibenzo-p-dioxins and -furans in cell lines from four
27 different species. Environ Toxicol Chem 26(11):2448-2454.
28 Pelekis, M; Gephart, LA; Lerman, SE. (2001) Physiological-model-based derivation of the adult and child
29 pharmacokinetic intraspecies uncertainty factors for volatile organic compounds. Reg Toxicol Pharmacol
30 33(1): 12-20.
31 Saghir, SA; Lebofsky, M; Pinson, DM; et al. (2005) Validation of Haber's Rule (dose X time = constant) in rats and
32 mice for monochloroacetic acid and 2,3,7,8-tetrachlorodibenzo-p-dioxin under conditions of kinetic steady state.
33 Toxicology 215(l-2):48-56.
34 Santostefano, MJ; Wang, X; Richardson, VM; et al. (1998) A pharmacodynamic analysis of TCDD-induced
3 5 cytochrome P450 gene expression in multiple tissues: dose- and time-dependent effects. Toxicol Appl Pharmacol
36 151:294-310.
37 Schantz, SL; Seo, BW; Moshtaghian, J; et al. (1996) Effects of gestational and lactational exposure to TCDD or
38 coplanarPCBs on spatial learning. Neurotoxicol Teratol 18(3):305—313.
39 Schecter, A; Pavuk, M; Papke, O; et al. (2003) Dioxin, dibenzofuran, and coplanar PCB levels in Laotian blood and
40 milk from agent orange-sprayed and nonsprayed areas, 2001. J Toxicol Environ Health Part A: Current Issues
41 66(21):2067-2075.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-199 DRAFT—DO NOT CITE OR QUOTE
-------
1 Sewall, CH; Flagler, N; Vanden Heuvel, JP; et al. (1995) Alterations in thyroid function in female Sprague-Dawley
2 rats following chronic treatment with 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol Appl Pharmacol 132:237-244.
3 Shi, Z; Valdez, K; Ting, A; et al. (2007) Ovarian endocrine disruption underlies premature reproductive senescence
4 following environmentally relevant chronic exposure to the aryl hydrocarbon receptor agonist 2,3,7,8-
5 tetrachlorodibenzo-p-dioxin. Biol Reprod 76(2): 198-202.
6 Smialowicz, RJ; DeVito, MJ; Williams, WC; et al. (2008) Relative potency based on hepatic enzyme induction
7 predicts immunosuppressive effects of a mixture of PCDDS/PCDFS and PCBS. Toxicol Appl Pharmacol
8 227(3):477-484.
9 Staskal, DF; Diliberto, JJ; Devito, MJ; et al. (2005) Inhibition of human and rat CYP1A2 by TCDD and dioxin-like
10 chemicals. Toxicol Sci 84(2) :225-231.
11 Toth, K; Somfai-Relle, S; Sugar, J; et al. (1979) Carcinogenicity testing of herbicide 2,4,5-trichlorophenoxyethanol
12 containing dioxin and of pure dioxin in Swiss mice. Nature 278:548-549.
13 Toyoshiba, H; Walker, NJ; Bailer, AJ; et al. (2004) Evaluation of toxic equivalency factors for induction of
14 cytochromes P450 CYP1A1 and CYP1A2 enzyme activity by dioxin-like compounds. Toxicol Appl Pharmacol
15 194(2): 156-168.
16 U.S. EPA (Environmental Protection Agency). (2003) Exposure and human health reassessment of 2,3,7,8-
17 tetrachlorodibenzo-p-dioxin (TCDD) and related compounds [NAS review draft]. Volumes 1-3. National Center
18 for Environmental Assessment, Washington, DC; EPA/600/P-00/001 Cb, Volume 1. Available online at
19 http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
20 Van Birgelen, AP; Van der Kolk, J; Fase, KM; et al. (1995) Subchronic dose-response study of 2,3,7,8-
21 tetrachlorodibenzo-p-dioxin in female Sprague-Dawley rats. Toxicol Appl Pharmacol 132:1-13.
22 Vanden Heuvel, JP; Clark, GC; Tritscher, A; et al. (1994) Accumulation of polychlorinated dibenzo-p-dioxins and
23 dibenzofurans in liver of control laboratory rats. Fundam Appl Toxicol 23:465-469.
24 Wang, X; Santostefano, MJ; Evans, MV; et al. (1997) Determination of parameters responsible for pharmacokinetic
25 behavior of TCDD in female Sprague-Dawley Rats. Toxicol Appl Pharmacol 147:151-168.
26 Wang, X; Santostefano, MJ; Devito, MJ; et al. (2000) Extrapolation of a PBPK model for dioxins across dosage
27 regimen, gender, strain, and species. Toxicol Sci 56(l):49-60.
28 White, KL, Jr; Lysy, HH; McCay, JA; et al. (1986) Modulation of serum complement levels following exposure to
29 polychlorinated dibenzo-p-dioxins. Toxicol Appl Pharmacol 84:209-219.
30 Wilkes, JG; Hass, BS; Buzatu, DA; et al. (2008) Modeling and assaying dioxin-like biological effects for both
31 dioxin-like and certain non-dioxin-like compounds. Toxicol Sci 102(1): 187-195.
32
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 C-200 DRAFT—DO NOT CITE OR QUOTE
-------
DRAFT
DO NOT CITE OR QUOTE
January 2010
Agency/Interagency Review Draft
APPENDIX D
Epidemiological Kinetic Modeling
NOTICE
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
CONTENTS—APPENDIX D: Epidemiological Kinetic Modeling
LIST OF TABLES D-iii
APPENDIX D. EPIDEMIOLOGICAL KINETIC MODELING D-1
D.l. BACCARELLIET AL. (2008) MODELING D-l
D. 1.1. Input File for Exposure During Pregnancy D-l
D.1.2. Table of Results for Baccarelli et al. (2008) D-l
D.2. MOCARELLI ET AL. (2008) MODELING D-2
D.2.1. Input File for Exposure for Pulse to Measurement 0.5 Years After the
Seveso Pulse Dose D-2
D.2.2. Input File for Exposure from Pulse to the End of the Critical Window
3.8 Years After the Seveso Pulse Dose D-2
D.2.3. Input File for Continuous Exposure for 10 Years D-3
D.2.4. Tables of Results for Mocarelli et al. (2008) D-4
D.3. ALALUUSUA ET AL. (2004) MODELING D-4
D.3.1. Input File for Exposure for Pulse to Measurement 0.5 Years After the
Seveso Pulse Dose D-4
D.3.2. Input File for Exposure from Pulse to the End of the Critical Window
2.5 Years After the Seveso Pulse Dose D-5
D.3.3. Input File for Continuous Exposure for 5 Years D-5
D.3.4. Tables of Results for Alaluusua et al. (2004) D-6
D.4. ESKANAZI ET AL. (2002) MODELING D-7
D.4.1. Input File for Exposure for Pulse to Measurement 0.5 Years After the
Seveso Pulse Dose D-7
D.4.2. Input File for Exposure from Pulse to the End of the Critical Window
6.7 Years After the Seveso Pulse Dose D-8
D.4.3. Input File for Continuous Exposure for 13 Years D-8
D.4.4. Tables of Results for Eskanazi et al. (2002) D-9
D.5. REFERENCES D-10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-ii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES
1 D-l. Estimated continuous intake corresponding to maternal serum concentration in
2 Figure 2A D-l
3 D-2. Estimated maximum intake corresponding to maternal serum concentration in
4 Figure 2A D-2
5 D-3. Matching critical window average after pulse to critical window average for
6 continuous intake run D-4
7 D-4. Matching critical window peak after pulse to peak critical window concentration
8 for continuous intake run D-4
9 D-5. Matching critical window average after pulse to critical window average for
10 continuous intake run D-6
11 D-6. Matching critical window peak after pulse to peak critical window
12 concentration for continuous intake run D-7
13 D-7. Matching critical window average after pulse to critical window average for
14 continuous intake run D-9
15 D-8. Matching critical window peak after pulse to peak critical window
16 concentration for continuous intake run D-9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-iii DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
APPENDIX D. EPIDEMIOLOGICAL KINETIC MODELING
D.l. BACCARELLI ET AL. (2008) MODELING
D.l.l. Input File for Exposure During Pregnancy
CINT = 1 % 168 %100 %integration time
%Exposure scenario
EXPTIMEON = 0 % delay before begin exposure (HOUR)
EXPTIMEOFF =401190 %TIME EXPO SURE STOP (HOUR)
DAY CYCLE = 24 %TIME
BCK TIME ON =401190 %DELAY BEFORE BACKGROUND EXP (HOUR)
BCK TIME OFF =401190 %TIME OF BACKGROUND EXP STOP (HOUR)
IVLACK =401190
IVPERIOD =401190
%GESTATION CONTROL
MATTING = 262800 % BEGINNING MATTING (HOUR)at 30 years old
TIMELIMIT = 269184 %SIMULATION LIMIT TIME (HOUR)
TRANSTIME ON = 264312 % EXCHANGE MOTHER FETUS 1512 HOUR POST
MATTING
%Exposure dose
MSTOT =0.021 % ng of TCDD /kg of BW
MSTOTBCKGR =0. %0.1 % ORAL BACKGROUND EXPOSURE DOSE (nG/KG)
DOSEIV = 0. %10
DOSEIVLATE = 0. %10
% TRANFER MOTHER TO FETUS CLEARANCE
CLPLAFET = 0.001 % MOTHER TO FETUS TRANFERT CLE ARAN CE(L/HR)
D.1.2. Table of Results for Baccarelli et al. (2008)
Table D-l. Estimated continuous intake corresponding to maternal serum
concentration in Figure 2A
Variable
Value
Notes
Infant b-TSH
5 uU/mL
BMR
Maternal lipid adjusted serum
270 ng/kg
From Figure 2A
Intake
0.024 ng/kg-day
From Emond model, pregnancy at 30
years
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-l DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Table D-2. Estimated maximum intake corresponding to maternal serum
concentration in Figure 2A
Variable
Value
Notes
Infant b-TSH
—
—
Maternal lipid adjusted serum
309.5 ng/kg
Maximum from Figure 2A
Intake
0.030 ng/kg-day
From Emond model, pregnancy at 30
years
D.2. MOCARELLI ET AL. (2008) MODELING
D.2.1. Input File for Exposure for Pulse to Measurement 0.5 Years After the Seveso Pulse
Dose
CINT = 1. %
EXPTIMEON = 54312. % Delay before begin exposure (HOUR) 6.2 years
EXPTIMEOFF = 54335. %324120 % HOUR/YEAR !TIME EXPOSURE STOP
(HOUR) 6.2 years + 23 hours
DAY CYCLE = 24. % TIME
BCK TIME ON = 0. % DELAY BEFORE BACKGROUND EXP (HOUR)
BCK TIME OFF = 613200 % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT = 58692. % half a year (July 1976 until January 1977) past 6.2 years
MSTOTBCKGR = 3.7E-4 % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 232.4 % Seveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
MEANLIPID = 731 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 1.
FEMALE = 0.
Y0 = 0. % 0 years old at the beginning of the simulation
D.2.2. Input File for Exposure from Pulse to the End of the Critical Window 3.8 Years
After the Seveso Pulse Dose
CINT = 1. %
EXP TIME ON = 54312. % Delay before begin exposure (HOUR) 6.2 years
EXP TIME OFF = 54335. %324120 % HOUR/YEAR !TIME EXPOSURE STOP
(HOUR) 6.2 years + 23 hours
DAY CYCLE = 24. % TIME
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
BCKTIMEON = 0. % DELAY BEFORE BACKGROUND EXP (HOUR)
BCKTIMEOFF = 613200. % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT = 87600. % 10 years
MSTOTBCKGR = 3.7e-4 % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 232.5 % Serveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
MEANLIPID = 730 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 1.
FEMALE = 0.
Y0 = 0. % 0 years old at the beginning of the simulation
D.2.3. Input File for Continuous Exposure for 10 Years
CINT = 1. %
EXP TIME ON =0. % Delay before begin exposure (HOUR)
EXP TIME OFF = 87600. % HOUR/YEAR !TIME EXPOSURE STOP (HOUR)
DAY CYCLE = 24. % TIME
BCK TIME ON = 0. %324120 % DELAY BEFORE BACKGROUND EXP (HOUR)
BCKTIMEOFF = 613200 %324120 % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT = 87600. % 10 years
MSTOTBCKGR = 0. %3.35E-4 % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 3.903 % Seveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
MEANLIPID = 730 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 1.
FEMALE = 0.
Y0 = 0. % 0 years old at the beginning of the simulation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
D.2.4. Tables of Results for Mocarelli et al. (2008)
Table D-3. Matching critical window average after pulse to critical window
average for continuous intake run
Person modeled,
beginning at age 0
Lipid adjusted
serum (1976) ng/kg
from Figure 3E
Pulse dose, 0.5
year lag time
(ng/kg)
Average lipid
adjusted serum
3.8 years after
incident (ng/kg)
Continuous intake
for 10 years
(ng/kg-day)
Boy, lstquartile
68
8.135
57.72
0.008024
Boy, 4th quartile
733
232.5
580.5
0.2128
Table D-4. Matching critical window peak after pulse to peak critical
window concentration for continuous intake run
Person modeled,
beginning at age 0
Lipid adjusted
serum (1976) ng/kg
from Figure 3E
Pulse dose, 0.5
year lag time
(ng/kg)
Peak lipid
adjusted serum
after incident
(ng/kg)
Continuous intake
for 10 years
(ng/kg-day)
Boy, lstquartile
68
8.135
248.0
0.03194
Boy, 4th quartile
733
232.5
6674
3.904
D.3. ALALUUSUA ET AL. (2004) MODELING
D.3.1. Input File for Exposure for Pulse to Measurement 0.5 Years After the Seveso Pulse
Dose
CINT = 1. %
EXPTIMEON =21900. % Delay before begin exposure (HOUR) 2.5 years
EXPTIMEOFF = 21923. % 21900+23 % HOUR/YEAR ! TIME EXPOSURE STOP
(HOUR) 2.5 years and 23 hours
DAY CYCLE = 24. % TIME
BCK TIME ON = 0. % DELAY BEFORE BACKGROUND EXP (HOUR)
BCK TIME OFF = 613200. % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT = 26280. % half a year (July 1976 until January 1977) past 2.5 years
MSTOTBCKGR = 3.7e-4 % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 24.22 % Seveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
MEANLIPID = 730 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 1.
FEMALE = 0.
Y0 = 0. % 0 years old at the beginning of the simulation
D.3.2. Input File for Exposure from Pulse to the End of the Critical Window 2.5 Years
After the Seveso Pulse Dose
CINT = 1. %
EXPTIMEON =21900. % Delay before begin exposure (HOUR) 2.5 years
EXPTIMEOFF = 21923. % 324120 % HOUR/YEAR ! TIME EXPOSURE STOP
(HOUR) 2.5 years and 23 hours
DAY CYCLE = 24. % TIME
BCKTIMEON = 0. % 324120 % DELAY BEFORE BACKGROUND EXP (HOUR)
BCKTIMEOFF = 613200. % 324120 % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT =43800. % 5 years
MSTOTBCKGR = 3.7e-4 % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 24.22 % Seveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
MEANLIPID = 730 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 1.
FEMALE = 0.
Y0 = 0. % 0 years old at the beginning of the simulation
D.3.3. Input File for Continuous Exposure for 5 Years
CINT = 1. %
EXP TIME ON =0. % Delay before begin exposure (HOUR)
EXP TIME OFF = 43800. % 324120 % HOUR/YEAR !TIME EXPOSURE STOP (HOUR)
DAY CYCLE = 24. % TIME
BCK TIME ON = 0. % 324120 % DELAY BEFORE BACKGROUND EXP (HOUR)
BCK TIME OFF = 613200. % 324120 % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT =43800. % End of critical window (5 years)
MSTOTBCKGR = 0. % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 0.03486 % Seveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
MEANLIPID = 730 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 1.
FEMALE = 0.
Y0 = 0. % 0 years old at the beginning of the simulation
D.3.4. Tables of Results for Alaluusua et al. (2004)
Table D-5. Matching critical window average after pulse to critical window
average for continuous intake run
Person modeled,
beginning at age 0
Lipid adjusted
serum (1976) ng/kg
estimated from
tertile bins3
Pulse dose,
0.5 year lag
time (ng/kg)
Average lipid
adjusted serum 2.5
years after
incident (ng/kg)
Continuous
intake for 5
years (ng/kg-
day)
Boy, 1st tertile
130
24.22
110.8
0.03486
Boy, 2nd tertile
383
108.9
322.7
0.1578
Boy, 3rd tertile
1830
1041
1538
1.511
Girl, 1st tertile
130
23.03
110.8
0.03211
Girl, 2nd tertile
383
105.3
324.4
0.1481
Girl, 3rd tertile
1830
1015
1546
1.427
Boy and girl, averaged,
1st tertile
130
-
-
0.03349
Boy and girl, averaged,
2nd tertile
383
-
-
0.1530
Boy and girl, averaged,
3rd tertile
1830
-
-
1.469
aMean of tertile bin assuming a lognormal distribution of serum concentrations.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
Table D-6. Matching critical window peak after pulse to peak critical
window concentration for continuous intake run
Person modeled,
beginning at age 0
Lipid adjusted
serum (1976) ng/kg
estimated from
tertile bins
Pulse dose,
0.5 year lag
time (ng/kg)
Peak lipid
adjusted serum
after incident
(ng/kg)
Continuous
intake for 5
years (ng/kg-
day)
Boy, 1st tertile
130
24.22
618.8
0.2113
Boy, 2nd tertile
383
108.9
2700
1.783
Boy, 3rd tertile
1830
1041
24706
31.35
Girl, 1st tertile
130
23.02
588.0
0.1882
Girl, 2nd tertile
383
105.3
2610
1.642
Girl, 3rd tertile
1830
1015
24113
29.52
Boy and girl, averaged,
1st tertile
130
-
-
0.1998
Boy and girl, averaged,
2nd tertile
383
-
-
1.713
Boy and girl, averaged,
3rd tertile
1830
-
-
30.44
"Mean of tertile bin assuming a lognormal distribution of serum concentrations.
D.4. ESKANAZI ET AL. (2002) MODELING
D.4.1. Input File for Exposure for Pulse to Measurement 0.5 Years After the Seveso Pulse
Dose
CINT = 1. %
EXPTIMEON = 58692. % Delay before begin exposure (HOUR) 6.7 years
EXPTIMEOFF = 58715. % HOUR/YEAR !TIME EXPOSURE STOP (HOUR) 6.7 years +
23 hours
DAY CYCLE = 24. % TIME
BCK TIME ON = 0. %324120 % DELAY BEFORE BACKGROUND EXP (HOUR)
BCK TIME OFF = 613200. %324120 % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT = 63072. % half a year (July 1976 until January 1977) past 6.7 years
MSTOTBCKGR = 3.7e-4 % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 7193 % Seveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
MEANLIPID = 730 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 0.
FEMALE = 1.
Y0 = 0. % 0 years old at the beginning of the simulation
D.4.2. Input File for Exposure from Pulse to the End of the Critical Window 6.7 Years
After the Seveso Pulse Dose
CINT = 1. %
EXPTIMEON = 58692. % Delay before begin exposure (HOUR) 6.7 years
EXPTIMEOFF = 58715. %324120 % HOUR/YEAR !TIME EXPOSURE STOP
(HOUR) 6.7 years + 23 hours
DAY CYCLE = 24. % TIME
BCKTIMEON = 0. %324120 % DELAY BEFORE BACKGROUND EXP (HOUR)
BCKTIMEOFF = 613200 %324120 % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT = 113880. % 13 years
MSTOTBCKGR = 3.7e-4 % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT =7193 % Seveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
MEANLIPID = 730 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 0.
FEMALE = 1.
Y0 = 0. % 0 years old at the beginning of the simulation
D.4.3. Input File for Continuous Exposure for 13 Years
CINT = 1. %
EXP TIME ON =0. % Delay before begin exposure (HOUR)
EXP TIME OFF = 113880. %324120 % HOUR/YEAR !TIME EXPOSURE STOP
(HOUR) 13 years
DAY CYCLE = 24. % TIME
BCK TIME ON = 0. %324120 % DELAY BEFORE BACKGROUND EXP (HOUR)
BCK TIME OFF = 613200. %324120 % TIME OF BACKGROUND EXP STOP (HOUR)
TIMELIMIT = 113880. % 13 years
MSTOTBCKGR = 0. %3.35E-4 % ORAL BACKGROUND EXPOSURE DOSE (UG/KG)
% oral dose oral dose oral dose
MSTOT = 166 % Seveso, ORAL DAILY EXPOSURE DOSE (NG/KG)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
DOSEIV = 0 % 40 %50 %5 %0.5 %0.3 %0.2 %0.1%0.05%0.3 %NG/KG
% oral dose oral dose oral dose
MEANLIPID = 730 % 711 %664 %778 %468 %671 %730 %662 %592%615%730%
PAS_INDUC= 1 % NON INDUCTION (0) CONTROLE DE L'INDUCTION
%human variable parameter
MALE = 0.
FEMALE = 1.
Y0 = 0. % 0 years old at the beginning of the simulation
D.4.4. Tables of Results for Eskanazi et al. (2002)
Table D-7. Matching critical window average after pulse to critical window
average for continuous intake run
Person modeled,
beginning at age 0
Lipid adjusted serum
(adjusted to 1976-
1977 levels) ng/kg
from Figure 1A
Pulse dose,
0.5 year lag
time (ng/kg)
Average lipid
adjusted serum 6.7
years after incident
(ng/kg)
Continuous
intake for 13
years (ng/kg-day)
Girl, estrous cycle
28.5 days
166
28.40
114.0
0.01660
Girl, estrous cycle
29 days
693
215.5
455.1
0.1224
Girl, estrous cycle
29.5 days
2020
1008
1295
0.5693
Girl, estrous cycle
30 days
8450
7193
5179
4.054
Table D-8. Matching critical window peak after pulse to peak critical
window concentration for continuous intake run
Person modeled,
beginning at age 0
Lipid adjusted serum
(adjusted to 1976-
1977 levels) ng/kg
from Figure 1A
Pulse dose,
0.5 year lag
time (ng/kg)
Peak lipid
adjusted serum
after incident
(ng/kg)
Continuous intake
for 13 years
(ng/kg-day)
Girl, estrous cycle
28.5 days
166
28.40
838.2
0.1800
Girl, estrous cycle
29 days
693
215.5
6183
3.148
Girl, estrous cycle
29.5 days
2020
1008
28316
20.86
Girl, estrous cycle
30 days
8450
7193
198240
166.6
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 D.5. REFERENCES
3 Alaluusua, S; Calderara, P; Gerthoux, PM; et al. (2004) Developmental dental aberrations after the dioxin accident
4 inSeveso. Environ Health Perspect 112(13): 1313-1318.
5 Baccarelli, A; Giacomini, SM; Corbetta, C; et al. (2008) Neonatal thyroid function in Seveso 25 years after maternal
6 exposure to dioxin. PLoS Med 5(7):el61.
7 Eskenazi, B; Mocarelli, P; Warner, M; et al. (2002). Serum dioxin concentrations and endometriosis: a cohort study
8 in Seveso, Italy. Environ Health Perspect 110(7): 629-634.
9 Mocarelli, P; Gerthoux, PM; Patterson, DG, Jr.; et al. (2008) Dioxin exposure, from infancy through puberty,
10 produces endocrine disruption and affects human semen quality. Environ Health Perspect 116(l):70-77.
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 D-10 DRAFT—DO NOT CITE OR QUOTE
-------
-------
DRAFT
DO NOT CITE OR QUOTE
January 2010
Agency/Interagency Review Draft
APPENDIX E
Noncancer Benchmark Dose Modeling
NOTICE
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
CONTENTS—APPENDIX E: Noncancer Benchmark Dose Modeling
APPENDIX E. NONCANCER BENCHMARK DOSE MODELING E-1
E.l. BMDS INPUT TABLES E-l
E.l.l. Amin et al. (2000) E-l
E.l.2. Bell et al. (2007a) E-l
E.l.3. Cantoni et al. (1981) E-2
E.l.4. Crofton et al. (2005) E-2
E.l.5. DeCaprio et al. (1986) E-3
E.l.6. Hojoetal. (2002) E-4
E.l.7. Kattainen et al. (2001) E-4
E.l.8. Keller et al. (2007, 2008a, b) E-5
E.l.9. Kociba et al. (1978) E-5
E. 1.10. Latchoumycandane and Mathur (2002) E-6
E.l. 11. Li et al. (1997) E-6
E.l. 12. Li et al. (2006) E-6
E. 1.13. Markowski et al. (2001) E-7
E.l. 14. Mietinnin et al. (2006) E-7
E.l.15. National Toxicology Program (1982) E-8
E.l. 16. National Toxicology Program (2006) E-8
E.l.17. Ohsakoetal. (2001) I>10
E.l.18. Schantzetal. (1996) I>10
E.l. 19. Shi et al. (2007) E-ll
E.l.20. Smialowicz et al. (2008) E-ll
E.l.21. Tothetal. (1979) 11-12
E.l.22. Van Birgelen et al. (1995) 11-12
11.1.23. White etal. (1986) 11-13
11.2. ALTERNATE DOSE: BLOOD SERUM BMDS RESULTS 11-14
E.2.1. Amin et al. (2000): Saccharin Consumed, Female (0.25%) E-14
E.2.1.1. Summary Table of BMDS Modeling Results E-14
E.2.1.2. Figure for Selected Model: Linear, Nonconstant Variance E-15
E.2.1.3. Output File for Selected Model: Linear, Nonconstant
Variance E-15
E.2.1.4. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-18
E.2.1.5. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-18
E.2.2. Amin et al. (2000): Saccharin Consumed, Female (0.50%) E-21
E.2.2.1. Summary Table of BMDS Modeling Results E-21
E.2.2.2. Figure for Selected Model: Linear, Nonconstant Variance E-22
E.2.2.3. Output File for Selected Model: Linear, Nonconstant
Variance E-22
E.2.2.4. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-25
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-ii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
E.2.2.5. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-25
E.2.3. Amin et al. (2000): Saccharin Preference Ratio, Female (0.25%) E-28
E.2.3.1. Summary Table of BMDS Modeling Results E-28
E.2.3.2. Figure for Selected Model: Linear, Nonconstant Variance E-29
E.2.3.3. Output File for Selected Model: Linear, Nonconstant
Variance E-29
E.2.3.4. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-32
E.2.3.5. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-32
E.2.4. Amin et al. (2000): Saccharin Preference Ratio, Female (0.50%) E-35
E.2.4.1. Summary Table of BMDS Modeling Results E-35
E.2.4.2. Figure for Selected Model: Linear, Constant Variance E-35
E.2.4.3. Output File for Selected Model: Linear, Constant Variance E-36
E.2.4.4. Figure for Unrestricted Model: Power, Constant Variance,
Power Unrestricted E-38
E.2.4.5. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted E-38
E.2.5. Bell et al. (2007): Balano-Preputial Separation in Male Pups
(10% extra risk) E-41
E.2.5.1. Summary Table of BMDS Modeling Results E-41
E.2.5.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-42
E.2.5.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-42
E.2.5.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-44
E.2.5.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-44
E.2.6. Bell et al. (2007): Balano-Preputial Separation in Male Pups
(5% extra risk) E-46
E.2.6.1. Summary Table of BMDS Modeling Results E-46
E.2.6.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-47
E.2.6.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-47
E.2.6.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-49
E.2.6.5. Output File for Unrestricted Model: Log-Logistic,
Slope Unrestricted E-49
E.2.7. Cantoni et al. (1981): Urinary Copro-Porhyrins E-51
E.2.7.1. Summary Table of BMDS Modeling Results E-51
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-iii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
E.2.7.2. Figure for Selected Model: Exponential (M4), Nonconstant
Variance, Power Restricted >1 E-52
E.2.7.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-52
E.2.7.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-55
E.2.7.5. Output file for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-55
E.2.7.6. Figure for Unrestricted Model: Hill, Nonconstant Variance,
n Unrestricted E-58
E.2.7.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-58
E.2.7.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-61
E.2.7.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-61
E.2.8. Cantoni et al. (1981): Urinary Porphyrins E-64
E.2.8.1. Summary Table of BMDS Modeling Results E-64
E.2.8.2. Figure for Selected Model: Exponential (M2), Nonconstant
Variance, Power Restricted >1 E-65
E.2.8.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 E-65
E.2.9. Crofton et al. (2005): Serum T4 E-68
E.2.9.1. Summary Table of BMDS Modeling Results E-68
E.2.9.2. Figure for Selected Model: Exponential (M4), Constant
Variance, Power Restricted >1 E-69
E.2.9.3. Output File for Selected Model: Exponential (M4), Constant
Variance, Power Restricted >1 E-69
E.2.9.4. Figure for Unrestricted Model: Exponential (M5), Constant
Variance, Power Unrestricted E-72
E.2.9.5. Output File for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted E-73
E.2.9.6. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted E-76
E.2.9.7. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted E-76
E.2.9.8. Figure for Unrestricted Model: Power, Constant Variance,
Power Unrestricted E-79
E.2.9.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted E-79
E.2.10. Hojo et al. (2002): DRL Reinforce Per Min E-82
E.2.10.1. Summary Table of BMDS Modeling Results E-82
E.2.10.2. Figure for Selected Model: Exponental (M4) E-83
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-iv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
E.2.10.3. Output File for Selected Model: Exponential (M4) E-83
E.2.11. Hojo et al. (2002): DRL Response Per Min E-86
E.2.11.1. Summary Table of BMDS Modeling Results E-86
E.2.11.2. Figure for Selected Model: Exponential (M4) E-87
E.2.11.3. Output File for Selected Model: Exponential (M4) E-87
E.2.12. Kattainen et al. (2001): 3rd Molar Mesio-Distal Length (Molar
Development) E-90
E.2.12.1. Summary Table of BMDS Modeling Results E-90
E.2.12.2. Figure for Selected Model: Hill, Nonconstant Variance, n
Restricted >1, Bound Hit E-91
E.2.12.3. Output File for Selected Model: Hill, Nonconstant Variance,
n Restricted >1, Bound Hit E-91
E.2.12.4. Figure for Unrestricted Model: Hill, Nonconstant Variance,
n Unrestricted E-94
E.2.12.5. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-94
E.2.13. Kattainen et al. (2001): Females 3rd Molar Eruption E-97
E.2.13.1. Summary Table of BMDS Modeling Results E-97
E.2.13.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-97
E.2.13.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-98
E.2.13.4. Figure for Unrestricted Model: Log-Logistic, Unrestricted E-99
E.2.13.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-100
E.2.14. Kattainen et al., 2001: 3rd molar length in pups E-101
E.2.14.1. Summary Table of BMDS modeling results E-101
E.2.14.2. Figure for selected model: Hill E-102
E.2.14.3. Output for selected model: Hill E-102
E.2.14.4. Figure for additional model presented: Hill, unrestricted E-105
E.2.14.5. Output for additional model presented: Hill, unrestricted E-105
E.2.15. Keller et al. (2006): Missing Mandibular Molars in CBA J Mice E-108
E.2.15.1. Summary Table of BMDS Modeling Results E-108
E.2.15.2. Figure for Selected Model: Multistage, 1-Degree E-109
E.2.15.3. Output File for Selected Model: Multistage, 1-Degree E-109
E.2.16. Kociba et al. (1978): Urinary Coproporphyrins, Females (Table 2) E-lll
E.2.16.1. Summary Table of BMDS Modeling Results E-lll
E.2.17. Kociba et al. (1978): Uroporphyrin per Creatinine, Females E-113
E.2.17.1. Summary Table of BMDS Modeling Results E-113
E.2.17.2. Figure for Selected Model: Linear, Constant Variance E-l 14
E.2.17.3. Output File for Selected Model: Linear, Constant Variance E-l 14
1
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-v DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
l
E.2.18. Latchoumycandane and Mathur (2002): Daily Sperm Production E-l 17
E.2.18.1. Summary Table of BMDS Modeling Results E-117
E.2.18.2. Figure for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-l 18
E.2.18.3. Output File for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-l 18
E.2.18.4. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted E-121
E.2.18.5. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted E-121
E.2.19. Li et al. (1997): Follicle-Stimulating Hormone E-123
E.2.19.1. Summary Table of BMDS Modeling Results E-123
E.2.19.2. Figure for Selected Model: Power E-124
E.2.19.3. Output for Selected Model: Power E-124
E.2.19.4. Figure for Unrestricted Model: Power, Unrestricted E-l27
E.2.19.5. Output for Unrestricted Model: Power, Unrestricted E-l27
E.2.20. Li et al. (2006): Hormone Levels (Estradiol) E-130
E.2.20.1. Summary Table of BMDS Modeling Results E-130
E.2.20.2. Figure for Selected Model: Linear, Constant Variance E-131
E.2.20.3. Output File for Unrestricted Model: Linear, Constant
Variance E-131
E.2.20.4. Figure for Unrestricted Model: Exponential (M5), Constant
Variance, Power Unrestricted E-134
E.2.20.5. Output File for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted E-134
E.2.20.6. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted E-l 37
E.2.20.7. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted E-137
E.2.20.8. Figure for Unrestricted Model: Power, Constant Variance,
Power Unrestricted E-l40
E.2.20.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted E-140
E.2.21. Li et al. (2006): Hormone Levels (Progesterone) E-143
E.2.21.1. Summary Table of BMDS Modeling Results E-143
E.2.21.2. Figure for Selected Model: Exponential (M2), Nonconstant
Variance, Power Restricted >1 E-144
E.2.21.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 E-144
E.2.21.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-147
E.2.21.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-147
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-vi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
1
E.2.21.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-150
E.2.21.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-150
E.2.21.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-153
E.2.21.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-153
E.2.22. Markowski et al. (2001): FRIO Run Opportunities E-156
E.2.22.1. Summary Table of BMDS Modeling Results E-156
E.2.22.2. Figure for Selected Model: Exponential (M2) E-156
E.2.22.3. Output File for Selected Model: Exponential (M2) E-157
E.2.23. Markowski et al. (2001): FR2 Revolutions E-160
E.2.23.1. Summary Table of BMDS Modeling Results E-160
E.2.23.2. Figure for Selected Model: Exponential (M5) E-161
E.2.23.3. Output File for Selected Model: Exponential (M5) E-161
E.2.23.4. Figure for Unrestricted Model: Power, Unrestricted E-164
E.2.23.5. Output for Unrestricted Model: Power, Unrestricted E-164
E.2.24. Markowski et al. (2001): FR5 Run Opp E-167
E.2.24.1. Summary Table of BMDS Modeling Results E-167
E.2.24.2. Figure for Selected Model: Hill E-168
E.2.24.3. Output File for Selected Model: Hill E-168
E.2.24.4. Figure for Unrestricted Model: Power, Unrestricted E-171
E.2.24.5. Output File for Unrestricted Model: Power, Unrestricted E-171
E.2.25. Mietinnin et al. (2006): Cariogenic Lesions in Pups E-174
E.2.25.1. Summary Table of BMDS Modeling Results E-174
E.2.25.2. Figure for Selected Model: Log-Logistic E-174
E.2.25.3. Output File for Selected Model: Log-Logistic E-175
E.2.25.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-176
E.2.25.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-177
E.2.26. National Toxicology Program (1982): Male Mice, Toxic Hepatitis E-178
E.2.26.1. Summary Table of BMDS Modeling Results E-178
E.2.26.2. Figure for Selected Model: Multistage, 2nd Degree E-179
E.2.26.3. Output File for Selected Model: Multistage, 2nd Degree E-179
E.2.27. National Toxicology Program (2006): Alveolar Metaplasia E-181
E.2.27.1. Summary Table of BMDS Modeling Results E-181
E.2.27.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1 E-182
E.2.27.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1 E-182
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-vii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
1
E.2.28. National Toxicology Program (2006): Gingival Hyperplasia
Squamous, 2 Years E-184
E.2.28.1. Summary Table of BMDS Modeling Results E-184
E.2.28.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-185
E.2.28.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-185
E.2.28.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-187
E.2.28.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-187
E.2.29. National Toxicology Program (2006): Heart, Cardiomyopathy E-189
E.2.29.1. Summary Table of BMDS Modeling Results E-189
E.2.29.2. Figure for Selected Model: Multistage, 2-Degree, Betas
Restricted >0, Bound Hit E-189
E.2.29.3. Output File for Selected Model: Multistage, 2-Degree,
Betas Restricted >0, Bound Hit E-190
E.2.30. National Toxicology Program (2006): Hepatocyte Hypertrophy, 2
Years E-191
E.2.30.1. Summary Table of BMDS Modeling Results E-191
E.2.30.2. Figure for Selected Model: Multistage, 2-Degree, Betas
Restricted >0, Bound Hit E-192
E.2.30.3. Output File for Selected Model: Multistage, 2-Degree,
Betas Restricted >0, Bound Hit E-193
E.2.31. National Toxicology Program (2006): Liver, Eosinophilic Focus,
Multiple E-194
E.2.31.1. Summary Table of BMDS Modeling Results E-194
E.2.31.2. Figure for Selected Model: Probit E-195
E.2.31.3. Output File for Selected Model: Probit E-195
E.2.32. National Toxicology Program (2006): Liver, Fatty Change, Diffuse E-197
E.2.32.1. Summary Table of BMDS Modeling Results E-197
E.2.32.2. Figure for Selected Model: Weibull, Power Restricted >1 E-198
E.2.32.3. Output File for Selected Model: Weibull, Power
Restricted >1 E-198
E.2.33. National Toxicology Program (2006): Liver Necrosis E-200
E.2.33.1. Summary Table of BMDS Modeling Results E-200
E.2.33.2. Figure for Selected Model: Log-Probit E-200
E.2.33.3. Output File for Selected Model: Log-Probit E-201
E.2.34. National Toxicology Program (2006): Liver, Pigmentation E-202
E.2.34.1. Summary Table of BMDS Modeling Results E-202
E.2.34.2. Figure for Selected Model: Log-Probit, Slope
Restricted >1 E-203
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-viii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
l
E.2.34.3. Output File for Selected Model: Log-Probit, Slope
Restricted >1 E-203
E.2.35. National Toxicology Program (2006): Liver, Toxic Hepatopathy E-205
E.2.35.1. Summary Table of BMDS Modeling Results E-205
E.2.35.2. Figure for Selected Model: Multistage, 2-Degree, Betas
Restricted >0, Bound Hit E-205
E.2.35.3. Output File for Selected Model: Multistage, 2-Degree,
Betas Restricted >0, Bound Hit E-206
E.2.35.4. Figure for Unrestricted Model: Multistage, 2-Degree, Betas
Unrestricted E-208
E.2.35.5. Output File for Unrestricted Model: Multistage, 2-Degree,
Betas Unrestricted E-208
E.2.36. National Toxicology Program (2006): Lung, Alveolar to Bronchiolar
Epithelial Metaplasia (Alveolar Epithelium, Metaplasia, Bronchiolar) ....E-210
E.2.36.1. Summary Table of BMDS Modeling Results E-210
E.2.36.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1 E-211
E.2.36.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1 E-211
E.2.37. National Toxicology Program (2006): Oval Cell Hyperplasia, 2
Years E-213
E.2.37.1. Summary Table of BMDS Modeling Results E-213
E.2.37.2. Figure for Selected Model: Probit E-213
E.2.37.3. Output File for Selected Model: Probit E-214
E.2.38. National Toxicology Program (2006): Toxic Hepatopathy E-215
E.2.38.1. Summary Table of BMDS Modeling Results E-215
E.2.38.2. Figure for Selected Model: Multistage, 2-Degree, Betas
Restricted >0, Bound Hit E-216
E.2.38.3. Output File for Selected Model: Multistage, 2-Degree,
Betas Restricted >0, Bound Hit E-216
E.2.38.4. Figure for Unrestricted Model: Multistage, 2-Degree,
Betas Unrestricted E-218
E.2.38.5. Output File for Unrestricted Model: Multistage, 2-Degree,
Betas Unrestricted E-218
E.2.39. Ohsako et al. (2001): Anogenital Distance in Male Pups E-220
E.2.39.1. Summary Table of BMDS Modeling Results E-220
E.2.39.2. Figure for Selected Model: Hill I>221
E.2.39.3. Output File for Selected Model: Hill E-221
E.2.39.4. Figure for Unrestricted Model: Hill, Unrestricted E-224
E.2.39.5. Output File for Unrestricted Model: Hill, Unrestricted E-224
E.2.40. Schantz et al. (1996): Maze Errors Per Block, Female E-227
E.2.40.1. Summary Table of BMDS Modeling Results E-227
E.2.40.2. Figure for Selected Model: Linear, Constant Variance E-228
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-ix DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
l
E.2.40.3. Output File for Selected Model: Linear, Constant Variance E-228
E.2.40.4. Figure for Unrestricted Model: Power, Constant Variance,
Power Unrestricted E-231
E.2.40.5. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted E-231
11.2.41. Shi et al. (2007): Estradiol 11-234
E.2.41.1. Summary Table of BMDS Modeling Results E-234
E.2.41.2. Figure for Selected Model: Exponential (M4), Nonconstant
Variance, Power Restricted >1 E-235
E.2.41.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-235
E.2.41.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-238
E.2.41.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-238
E.2.41.6. Figure for Unrestricted Model: Hill, Nonconstant Variance,
n Unrestricted E-241
E.2.41.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-241
E.2.41.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-244
E.2.41.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-244
11.2.42. Smialowicz et al. (2008): PFC per 10 6 Cells 11-247
E.2.42.1. Summary Table of BMDS Modeling Results E-247
E.2.42.2. Figure for Selected Model: Linear, Nonconstant Variance E-248
E.2.42.3. Output File for Selected Model: Linear, Nonconstant
Variance E-248
E.2.42.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-251
E.2.42.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-251
E.2.42.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-254
E.2.42.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-254
E.2.42.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-257
E.2.42.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-257
E.2.43. Smialowicz et al. (2008): PFC per Spleen E-260
E.2.43.1. Summary Table of BMDS Modeling Results E-260
E.2.43.2. Figure for Selected Model: Linear, Nonconstant Variance E-261
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-x DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
1
E.2.43.3. Output File for Selected Model: Linear, Nonconstant
Variance E-261
E.2.43.4. Figure for UnrestrictedModel: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-264
E.2.43.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-264
E.2.43.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-267
E.2.43.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-267
E.2.43.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-270
E.2.43.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-270
E.2.44. Toth et al. (1978): Amyloidosis E-273
E.2.44.1. Summary Table of BMDS Modeling Results E-273
E.2.44.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1 E-274
E.2.44.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1 E-274
E.2.44.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-276
E.2.44.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-276
E.2.44.6. Figure for Unrestricted Model: Log-Probit, Slope
Restricted >1 E-278
E.2.44.7. Output File for Unrestricted Model: Log-Probit, Slope
Restricted >1 E-278
E.2.45. Toth et al. (1978): Skin Lesions E-280
E.2.45.1. Summary Table of BMDS Modeling Results E-280
E.2.45.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1 E-281
E.2.45.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1 E-281
E.2.45.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-283
E.2.45.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-283
E.2.46. Van Birgelen et al. (1995a): Hepatic Retinol E-285
E.2.46.1. Summary Table of BMDS Modeling Results E-285
E.2.46.2. Figure for Selected Model: Exponential (M4), Nonconstant
Variance, Power Restricted >1 E-287
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
1
E.2.46.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-287
E.2.46.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-290
E.2.46.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-290
E.2.46.6. Figure for Unrestricted Model: Hill, Nonconstant Variance,
n Unrestricted E-293
E.2.46.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-293
E.2.46.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-296
E.2.46.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-296
E.2.47. Van Birgelen et al. (1995a): Hepatic Retinol Palmitate E-299
E.2.47.1. Summary Table of BMDS Modeling Results E-299
E.2.47.2. Figure for Selected Model: Exponential (M4), Nonconstant
Variance, Power Restricted >1 E-300
E.2.47.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-300
E.2.47.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-303
E.2.47.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-303
E.2.47.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-306
E.2.47.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-306
E.2.47.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-309
E.2.47.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-309
E.2.48. Van Birgelen et al. (1995a): Plasma FT4 E-312
E.2.48.1. Summary Table of BMDS Modeling Results E-312
E.2.48.2. Figure for Selected Model: Exponential (M2), Nonconstant
Variance, Power Restricted >1 E-313
E.2.48.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 E-314
E.2.48.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-316
E.2.48.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-317
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
E.2.48.6. Figure for Unrestricted Model: Hill, Nonconstant Variance,
n Unrestricted E-319
E.2.48.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-320
E.2.48.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-322
E.2.48.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-322
E.2.49. Van Birgelen et al. (1995a): Plasma TT4 E-325
E.2.49.1. Summary Table of BMDS Modeling Results E-325
E.2.49.2. Figure for Selected Model: Exponential (M2), Constant
Variance, Power Restricted >1 E-327
E.2.49.3. Output File for Selected Model: Exponential (M2),
Constant Variance, Power Restricted >1 E-327
E.2.49.4. Figure for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted E-330
E.2.49.5. Output File for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted E-330
E.2.49.6. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted E-3 3 3
E.2.49.7. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted E-333
E.2.49.8. Figure for Unrestricted Model: Power, Constant Variance,
Power Unrestricted E-3 3 6
E.2.49.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted E-336
E.2.50. White et al. (1986): CH50 E-338
E.2.50.1. Summary Table of BMDS Modeling Results E-338
E.2.50.2. Figure for Selected Model: Hill, Nonconstant Variance, n
Restricted >1, Bound Hit E-340
E.2.50.3. Output File for Selected Model: Hill, Nonconstant
Variance, n Restricted >1, Bound Hit E-340
E.2.50.4. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-343
E.2.50.5. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-343
E.3. ADMINISTERED DOSE BMDS RESULTS E-346
E.3.1. Amin et al. (2000): Saccharin Consumed, Female (0.25%) E-346
E.3.1.1. Summary Table of BMDS Modeling Results E-346
E.3.1.2. Figure for Selected Model: Linear, Nonconstant Variance E-347
E.3.1.3. Output file for Selected Model: Linear, Nonconstant
Variance E-347
E.3.2. Amin et al. (2000): Saccharin Consumed, Female (0.50%) E-349
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xiii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
l
E.3.2.1. Summary Table of BMDS Modeling Results E-349
E.3.2.2. Figure for Selected Model: Linear, Nonconstant Variance E-350
E.3.2.3. Output File for Selected Model: Linear, Nonconstant
Variance E-350
E.3.3. Amin et al. (2000): Saccharin Preference Ratio, Female (0.25%) E-353
E.3.3.1. Summary Table of BMDS Modeling Results E-353
E.3.3.2. Figure for Selected Model: Linear, Nonconstant Variance E-353
E.3.3.3. Output File for Selected Model: Linear, Nonconstant
Variance E-354
E.3.4. Amin et al. (2000): Saccharin Preference Ratio, Female (0.50%) E-356
E.3.4.1. Summary Table of BMDS Modeling Results E-356
E.3.4.2. Figure for Selected Model: Linear, Nonconstant Variance E-357
E.3.4.3. Output File for Selected Model: Linear, Nonconstant
Variance E-357
E.3.5. Bell et al. (2007): Balano-Preputial Separation in Male Pups (10%
Extra Risk) E-359
E.3.5.1. Summary Table of BMDS modeling results E-359
E.3.5.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-360
E.3.5.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-360
E.3.5.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-362
E.3.5.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-362
E.3.6. Bell et al. (2007): Balano-Preputial Separation in Male Pups (5%
Extra Risk)
E.3.6.1. Summary Table of BMDS Modeling Results
E.3.6.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit
E.3.6.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit
E.3.6.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted
E.3.6.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-367
E.3.7. Cantoni et al. (1981): Urinary Copro-Porhyrins E-369
E.3.7.1. Summary Table of BMDS Modeling Results E-369
E.3.7.2. Figure for Selected Model: Probit E-369
E.3.7.3. Output File for Selected Model: Probit E-370
E.3.8. Cantoni et al. (1981): Urinary Porphyrins E-371
E.3.8.1. Summary Table of BMDS Modeling Results E-371
E.3.8.2. Figure for Selected Model: Linear, Nonconstant Variance E-372
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xiv DRAFT—DO NOT CITE OR QUOTE
E-364
E-364
E-365
E-365
E-367
-------
CONTENTS (continued)
1
E.3.8.3. Output File for Selected Model: Linear, Nonconstant
Variance E-373
E.3.9. Crofton et al. (2005): Serum T4 E-375
E.3.9.1. Summary Table of BMDS Modeling Results E-375
E.3.9.2. Figure for Selected Model: Exponential (M4), Constant
Variance, Power Restricted >1 E-376
E.3.9.3. Output File for selected Model: Exponential (M4),
Constant Variance, Power Restricted >1 E-376
E.3.10. DeCaprio et al. (1986): Absolute Kidney Weight, Males E-379
E.3.10.1. Summary Table of BMDS Modeling Results E-379
E.3.10.2. Figure for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-380
E.3.10.3. Output File for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-380
E.3.11. DeCaprio et al. (1986): Absolute Thymus Weight, Males E-383
E.3.11.1. Summary Table of BMDS Modeling Results E-383
E.3.11.2. Figure for Selected Model: Power, Nonconstant Variance,
Power Restricted >1, Bound Hit E-384
E.3.11.3. Output File for Selected Model: Power, Nonconstant
Variance, Power Restricted >1, Bound Hit E-384
E.3.12. DeCaprio et al. (1986): Body Weight, Females E-387
E.3.12.1. Summary Table of BMDS Modeling Results E-387
E.3.12.2. Figure for Selected Model: Exponential (M2), Constant
Variance, Power Restricted >1 E-388
E.3.12.3. Output File for Selected Model: Exponential (M2),
Constant Variance, Power Restricted >1 E-388
E.3.13. DeCaprio et al. (1986): Body Weight, Males E-391
E.3.13.1. Summary Table of BMDS Modeling Results E-391
E.3.13.2. Figure for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-392
E.3.13.3. Output File for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-392
E.3.14. DeCaprio et al. (1986): Relative Brain Weight, Males E-395
E.3.14.1. Summary Table of BMDS Modeling Results E-395
E.3.14.2. Figure for Selected Model: Exponential (M4), Constant
Variance, Power Restricted >1 E-396
E.3.14.3. Output File for Selected Model: Exponential (M4),
Constant Variance, Power Restricted >1 E-396
E.3.15. DeCaprio et al. (1986): Relative Liver Weight, Females E-399
E.3.15.1. Summary Table of BMDS Modeling Results E-399
E.3.15.2. Figure for Selected Model: Exponential (M2), Nonconstant
Variance, Power Restricted >1 E-400
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
l
E.3.15.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 E-400
E.3.16. DeCaprio et al. (1986): Relative Liver Weight, Males E-403
E.3.16.1. Summary Table of BMDS Modeling Results E-403
E.3.16.2. Figure for Selected Model: Exponential (M4), Nonconstant
Variance, Power Restricted >1 E-404
E.3.16.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-404
E.3.17. DeCaprio et al. (1986): Relative Thymus Weight, Males E-407
E.3.17.1. Summary Table of BMDS Modeling Results E-407
E.3.17.2. Figure for selected Model: Exponential (M2), Constant
Variance, Power Restricted >1 E-408
E.3.17.3. Output File for Selected Model: Exponential (M2),
Constant Variance, Power Restricted >1 E-408
E.3.18. Hojo et al. (2002): DRL Reinforce per Min E-411
E.3.18.1. Summary Table of BMDS Modeling Results E-411
E.3.18.2. Figure for Selected Model: Linear, Constant Variance E-412
E.3.18.3. Output File for Selected Model: Linear, Constant Variance E-412
E.3.19. Hojoetal. (2002): DRL Response per Min E-415
E.3.19.1. Summary Table of BMDS Modeling Results E-415
E.3.19.2. Figure for Selected Model: Exponential (M4), Constant
Variance, Power Restricted >1 E-416
E.3.19.3. Output File for Selected Model: Exponential (M4),
Constant Variance, Power Restricted >1 E-416
E.3.20. Kattainen et al. (2001): 3rd Molar Mesio-Distal Length (Molar
Development) E-419
E.3.20.1. Summary Table of BMDS Modeling Results E-419
E.3.20.2. Figure for Selected Model: Hill, Nonconstant Variance, n
Restricted >1, Bound Hit E-420
E.3.20.3. Output File for Selected Model: Hill, Nonconstant
Variance, n Restricted >1, Bound Hit E-420
E.3.21. Kattainen et al. (2001): Females 3rd Molar Eruption E-423
E.3.21.1. Summary Table of BMDS Modeling Results E-423
E.3.21.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-424
E.3.21.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-424
E.3.22. Keller et al. (2006): Missing Mandibular Molars in CBA J Mice E-426
E.3.22.1. Summary Table of BMDS Modeling Results E-426
E.3.22.2. Figure for Selected Model: Multistage, 1-Degree, Betas
Restricted >0 E-427
E.3.22.3. Output File for Selected Model: Multistage, 1-Degree,
Betas Restricted >0 E-427
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xvi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
1
E.3.23. Kociba et al. (1978): Urinary Coproporphyrins, Females (Table 2) E-429
E.3.23.1. Summary Table of BMDS Modeling Results E-429
E.3.23.2. Figure for Selected Model: Exponential (M4), Nonconstant
Variance, Power Restricted >1 E-430
E.3.23.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-430
E.3.24. Kociba et al. (1978): Uroporphyrin per Creatinine, Females E-433
E.3.24.1. Summary Table of BMDS Modeling Results E-433
E.3.24.2. Figure for Selected Model: Linear, Constant Variance E-434
E.3.24.3. Output File for Selected Model: Linear, Constant Variance E-434
E.3.25. Latchoumycandane and Mathur (2002): Daily sperm Production E-437
E.3.25.1. Summary Table of BMDS Modeling Results E-437
E.3.25.2. Figure for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-438
E.3.25.3. Output File for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-438
E.3.26. Li et al. (2006): Hormone Levels (Estradiol) E-441
E.3.26.1. Summary Table of BMDS Modeling Results E-441
E.3.26.2. Figure for Selected Model: Exponential (M2), Constant
Variance, Power Restricted >1 E-442
E.3.26.3. Output File for Selected Model: Exponential (M2),
Constant Variance, Power Restricted >1 E-442
E.3.27. Li et al. (2006): Hormone Levels (Progesterone) E-445
E.3.27.1. Summary Table of BMDS Modeling Results E-445
E.3.27.2. Figure for Selected Model: Exponential (M4), Nonconstant
Variance, Power Restricted >1 E-446
E.3.27.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-446
E.3.28. Markowski et al. (2001): FR10 Run Opp E-449
E.3.28.1. Summary Table of BMDS Modeling Results E-449
E.3.28.2. Figure for Selected Model: Exponential (M2), Constant
Variance, Power Restricted >1 E-450
E.3.28.3. Output File for Selected Model: Exponential (M2),
Constant Variance, Power Restricted >1 E-450
E.3.29. Markowski et al. (2001): FR2 Revolutions E-453
E.3.29.1. Summary Table of BMDS Modeling Results E-453
E.3.29.2. Figure for Selected Model: Exponential (M2), Constant
Variance, Power Restricted >1 E-454
E.3.29.3. Output File for Selected Model: Exponential (M2),
Constant Variance, Power Restricted >1 E-454
E.3.30. Markowski et al. (2001): FR5 Run Opp E-457
E.3.30.1. Summary Table of BMDS Modeling Results E-457
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xvii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
1
E.3.30.2. Figure for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-458
E.3.30.3. Output File for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-458
E.3.31. Mietinnin et al. (2006): Caries E-461
E.3.31.1. Summary Table of BMDS Modeling Results E-461
E.3.31.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-462
E.3.31.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-462
E.3.32. National Toxicology Program (1982): Male Mice, Toxic Hepatitis E-464
E.3.32.1. Summary Table of BMDS Modeling Results E-464
E.3.32.2. Figure for Selected Model: Logistic E-464
E.3.32.3. Output file for Selected Model: Logistic E-465
E.3.33. National Toxicology Program (2006): Alveolar Metaplasia E-466
E.3.33.1. Summary Table of BMDS Modeling Results E-466
E.3.33.2. Figure for Selected Model: Log-Probit, Slope
Restricted >1 E-467
E.3.33.3. Output File for Selected Model: Log-Probit, Slope
Restricted >1 E-467
E.3.34. National Toxicology Program (2006): Gingival Hyperplasia
Squamous, 2 Years E-469
E.3.34.1. Summary Table of BMDS Modeling Results E-469
E.3.34.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-470
E.3.34.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-470
E.3.34.4. Figure for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-472
E.3.34.5. Output File for Unrestricted Model: Log-Logistic, Slope
Unrestricted E-472
E.3.35. National Toxicology Program (2006): Heart, Cardiomyopathy E-474
E.3.35.1. Summary Table of BMDS Modeling Results E-474
E.3.35.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-474
E.3.35.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-475
E.3.36. National Toxicology Program (2006): Hepatocyte Hypertrophy, 2
Years E-476
E.3.36.1. Summary Table of BMDS Modeling Results E-476
E.3.36.2. Figure for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-477
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xviii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
l
E.3.36.3. Output File for Selected Model: Log-Logistic, Slope
Restricted >1, Bound Hit E-477
E.3.37. National Toxicology Program (2006): Liver, Eosinophilic Focus,
Multiple E-479
E.3.37.1. Summary Table of BMDS Modeling Results E-479
E.3.37.2. Figure for Selected Model: Weibull, Power Restricted >1,
Bound Hit E-479
E.3.37.3. Output File for Selected Model: Weibull, Power
Restricted >1, Bound Hit E-480
E.3.38. National Toxicology Program (2006): Liver, Fatty Change, Diffuse E-481
E.3.38.1. Summary Table of BMDS Modeling Results E-481
E.3.39. Figure for Selected Model: Multistage, 2-Degree, Betas
Restricted >0 E-482
E.3.40. Output File for Selected Model: Multistage, 2-Degree, Betas
Restricted >0 E-482
E.3.41. National Toxicology Program (2006): Liver Necrosis E-484
E.3.41.1. Summary Table of BMDS Modeling Results E-484
E.3.41.2. Figure for Selected Model: LogProbit E-485
E.3.41.3. Output File for Selected Model: Log-Probit E-485
E.3.42. National Toxicology Program (2006): Liver, Pigmentation E-487
E.3.42.1. Summary Table of BMDS Modeling Results E-487
E.3.42.2. Figure for Selected Model: Log-Probit, Slope
Restricted >1 E-487
E.3.42.3. Output File for Selected Model: Log-Probit, Slope
Restricted >1 E-488
E.3.43. National Toxicology Program (2006): Lung, Alveolar to Bronchiolar
Epithelial Metaplasia (Alveolar Epithelium, Metaplasia,
Bronchiolar) E-489
E.3.43.1. Summary Table of BMDS Modeling Results E-489
E.3.43.2. Figure for Selected Model: Log-Probit, Slope
Restricted >1 E-490
E.3.43.3. Output File for Selected Model: Log-Probit, Slope
Restricted >1 E-490
E.3.44. National Toxicology Program (2006): Oval Cell Hyperplasia, 2
Years E-492
E.3.44.1. Summary Table of BMDS Modeling Results E-492
E.3.44.2. Figure for Selected Model: Multistage, 2-Degree, Betas
Restricted >0 E-492
E.3.44.3. Output File for Selected Model: Multistage, 2-Degree,
Betas Restricted >0 E-493
E.3.45. National Toxicology Program (2006): Toxic Hepatopathy E-494
E.3.45.1. Summary Table of BMDS Modeling Results E-494
E.3.45.2. Figure for Selected Model: Gamma, Power Restricted >1 E-495
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xix DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
l
E.3.45.3. Output file for Selected Model: Gamma, Power
Restricted >1 E-495
E.3.45.4. Figure for Unrestricted Model: Weibull, Power
Restricted >1 E-497
E.3.45.5. Output File for Unrestricted Model: Weibull, Power
Restricted >1 E-497
E.3.46. Ohsako et al. (2001): Anogenital PND120 E-499
E.3.46.1. Summary Table of BMDS Modeling Results E-499
E.3.46.2. Figure for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-500
E.3.46.3. Output File for Selected Model: Hill, Constant Variance,
n Restricted >1, Bound Hit E-500
E.3.47. Schantz et al. (1996): Maze Errors Per Block, Female E-503
E.3.47.1. Summary Table of BMDS Modeling Results E-503
E.3.47.2. Figure for Selected Model: Linear, Constant Variance E-504
E.3.47.3. Output File for Selected Model: Linear, Constant
Variance E-504
E.3.48. Shi et al. (2007): Estradiol I>506
E.3.48.1. Summary Table of BMDS Modeling Results E-506
E.3.48.2. Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-508
E.3.48.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-508
E.3.49. Smialowicz et al. (2008): PFC per 10 6 Cells 11-511
E.3.49.1. Summary Table of BMDS Modeling Results E-511
E.3.49.2. Figure for Selected Model: Exponential (M2), Nonconstant
Variance, Power Restricted >1 E-512
E.3.49.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 E-512
E.3.50. Smialowicz et al. (2008): PFC per Spleen E-515
E.3.50.1. Summary Table of BMDS Modeling Results E-515
E.3.50.2. Figure for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 E-516
E.3.50.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 E-516
E.3.51. Toth et al. (1978): Amyloidosis E-519
E.3.51.1. Summary Table of BMDS Modeling Results E-519
E.3.51.2. Figure for Selected Model: Multistage, 1-Degree, Betas
Restricted >0 E-519
E.3.51.3. Output File for Selected Model: Multistage, 1-Degree,
Betas Restricted >0 E-519
E.3.52. Toth et al. (1978): Skin Lesions E-521
E.3.52.1. Summary Table of BMDS Modeling Results E-521
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xx DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
1
E.3.52.2. Figure for Selected Model: Log-Probit, Slope
Restricted >1 E-522
E.3.52.3. Output File for Selected Model: Log-Probit, Slope
Restricted >1 E-522
E.3.53. Van Birgelen et al. (1995a): Hepatic Retinol E-524
E.3.53.1. Summary Table of BMDS Modeling Results E-524
E.3.53.2. Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-525
E.3.53.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 E-526
E.3.53.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-528
E.3.53.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-529
E.3.53.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-531
E.3.53.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-532
E.3.53.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-534
E.3.53.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-535
E.3.54. Van Birgelen et al. (1995a): Hepatic Retinol Palmitate E-537
E.3.54.1. Summary Table of BMDS Modeling Results E-537
E.3.54.2. Figure for Selected Model: Hill, Nonconstant Variance, n
Unrestricted E-539
E.3.54.3. Output File for Selected Model: Hill, Nonconstant
Variance, n Unrestricted E-539
E.3.54.4. Figure for Unrestricted Model: Linear, Nonconstant
Variance E-542
E.3.54.5. Output File for Unrestricted Model: Linear, Nonconstant
Variance E-542
E.3.54.6. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-545
E.3.54.7. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-545
E.3.54.8. Figure for Unrestricted Model: Exponential (M5), Constant
Variance, Power Unrestricted E-548
E.3.54.9. Output File for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted E-548
E.3.55. Van Birgelen et al. (1995a): Plasma FT4 E-551
E.3.55.1. Summary Table of BMDS Modeling Results E-551
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xxi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
E.3.55.2. Figure for Selected Model: Exponential (M2), Nonconstant
Variance, Power Restricted >1 E-552
E.3.55.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 E-553
E.3.55.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-555
E.3.55.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted E-556
E.3.55.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-558
E.3.55.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted E-559
E.3.55.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-561
E.3.55.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted E-561
E.3.56. Van Birgelen et al. (1995a): Plasma TT4 E-564
E.3.56.1. Summary Table of BMDS Modeling Results E-564
E.3.56.2. Figure for Selected Model: Exponential (M2), Constant
Variance, Power Restricted >1 E-565
E.3.56.3. Output File for Selected Model: Exponential (M2),
Constant Variance, Power Restricted >1 E-566
E.3.56.4. Figure for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted E-569
E.3.56.5. Output File for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted E-569
E.3.56.6. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted E-572
E.3.56.7. Output File for Unrestricted Model: Hill, constant
Variance, n Unrestricted E-572
E.3.56.8. Figure for Unrestricted Model: Power, Constant Variance,
Power Unrestricted E-575
E.3.56.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted E-575
E.3.57. White et al. (1986): CH50 E-578
E.3.57.1. Summary Table of BMDS Modeling Results E-578
E.3.57.2. Figure for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-579
E.3.57.3. Output File for Selected Model: Hill, Constant Variance, n
Restricted >1, Bound Hit E-579
E.4. REFERENCES E-582
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-xxii DRAFT—DO NOT CITE OR QUOTE
-------
APPENDIX E.
NONCANCER BENCHMARK DOSE MODELING
E.l. BMDS INPUT TABLES
E.l.l. Amin et al. (2000)
Endpoint
Administered Dose (ng/kg-day)
0
25 a
100
Internal Dose (ng/kg blood) b
0
6,800
24,522
(n = 10)
(n = 10)
(n = 10)
Saccharin consumed, female (0.25%)
31.67 ±26.64
24.60 ± 11.98
10.70 ± 5.33
Saccharin consumed, female (0.50%)
22.40 ± 15.98
11.38 ± 7.66
4.54 ±3.33
Saccharin preference ratio, female (0.25%)
82.14 ± 13.35
58.12 ± 33.88
54.87 ± 19.51
Saccharin preference ratio, female (0.50%)
72.73 ± 24.64
44.48 ±32.85
33.77 ±24.64
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
E.1.2. Bell et al. (2007a)
Administered Dose (ng/kg-day)
0
2.4 a
8
46
Internal Dose (ng/kg blood) b
0
1,998
4,539
15,952
Endpoint
(n = 30)
(n = 30)
(n = 30)
(n = 30)
Balano-preputial separation, male
pups
1/30 (3%)
5/30 (17%)
6/30 (20%)
15/30 (50%)
aLOAEL.
b From the Emond PRPK model described in 3.3.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l DRAFT—DO NOT CITE OR QUOTE
-------
E.1.3. Cantoni et al. (1981)
Endpoint
Administered Dose (ng/kg-day)
0
1.43 a
14.3
143
Internal Dose (ng/kg blood) b
0
1,018
4,868
27,559
(n = 4)
(n = 4)
(n = 3)
(n = 3)
Urinary coporphyrins
0.74 ±0.35
1.81 ±0.83 c
2.73 ± 1.50 d
3.00 ± 2.60 d
Urinary porphyrins
2.27 ± 0.49
5.55 ± 0.85 c
7.62 ± 1.79 c
196.89 ±63.14
d
aLOAEL
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.05).
d Statistically significant as compared to control (p < 0.01).
E.1.4. Crofton et al. (2005)
Endpoint
Administered Dose (ng/kg-day)
0
0.1
3
10
30 a
100 b
300
1,000
3,000
10,000
Internal Dose (ng/kg blood)c
0
11.3
273
773
1,922
51,11
12,624
35,697
98,088
316,540
(n =
14)
(n = 6)
(n =
12)
(n = 6)
(n = 6)
(n = 6)
(n = 6)
(n = 6)
(n = 6)
(n = 4)
Serum T4
100.00
±
15.44
96.27
±
14.98
98.57
±
18.11
99.76
±
19.04
93.32
±
12.11
70.94
±
12.74
62.52
±
14.75
52.68
±
22.73
54.66
±
19.71
49.15
±
11.17
aNOAEL
bLOAEL
cFrom the Emond PRPK model described in 3.3.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-2 DRAFT—DO NOT CITE OR QUOTE
-------
l E.1.5. DeCaprio et al. (1986)
Endpoint
Administered Dose (ng/kg-day)
0
0.12
0.61 a
4.9 b
26
Internal Dose (ng/kg blood)c
n/a
n/a
n/a
n/a
n/a
(n = 10)
(n = 10)
(n = 11)
(n = 10)
(n = 4)
Absolute kidney weight,
males
5.49 ±
0.54
5.14 ±0.38
4.71 ±0.4
4.3±0.47d
-
Absolute thymus weight,
males
0.56 ±
0.16
0.45 ±0.07
0.44 ±0.11
0.35±0.53e
-
Body weight, males
713 ±
47.43
682 ± 50.6
651 ±63.02
603±63.25d
433 ± 76
f
Relative brain weight,
males
0.54 ±
0.05
0.56 ±0.05
0.6 ±0.05
0.65±0.05d
-
Relative liver weight, males
4.54 ±
0.73
4.1 ±0.44
5.36 ±2.02
5.63±0.92d
-
Relative thymus weight,
males
0.08 ±
0.02
0.07 ±0.01
0.07 ±0.01
0.06±0.01d
-
Endpoint
Administered Dose (ng/kg-day)
0
0.12
0.68
4.86
31
Internal Dose (ng/kg blood)c
0
n/a
n/a
n/a
n/a
(n = 8)
(n = 10)
(n = 9)
(n = 10)
(n = 4)
Body weight, females
602 ±
33.94
583 ±69.57
570 ± 66
531 ±44.27 d
351 ±98
f
Relative liver weight,
females
4.3 ±0.74
4.49 ± 1.11
4.27 ±0.48
5.54 ± 1.36
d
4.3 ±
0.74
aNOAEL
bLOAEL.
c Internal dose not calculated using the Emond PBPK (ginuea pigs).
d Statistically significant as compared to control (p < 0.05).
e Statistically significant as compared to control (p < 0.01).
Statistically significant as compared to control (p < 0.001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-3 DRAFT—DO NOT CITE OR QUOTE
-------
E.1.6. Hojo et al. (2002)
Endpoint
Administered Dose (ng/kg-day)
0
20 a
60
180
Internal Dose (ng/kg blood) b
0
1,285
3,295
8,465
(n = 5)
(n = 5)
(n = 6)
(n = 5)
DRL reinforce per min
0.09 ±0.45
0.54 ±0.82
1.27 ±0.54
0.74 ± 0.44
DRL response per min
18.46 ±7.99
-0.99 ± 10.96
-4.52 ±7.19
-0.41 ± 15.23
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
E.1.7. Kattainen et al. (2001)
Endpoint
Administered Dose (ng/kg-day)
0
30 a
100
300
1,000
Internal Dose (ng/kg blood) b
0
1,763
4,944
12,712
37,039
(n = 16)
(n = 17)
(n = 15)
(n = 12)
(n = 19)
3rd molar mesio-distal
length (molar development)
1.86 ±0.07
1.58 ± 0.19
C
1.6 ± 0.27 c
1.5 ± 0.22 c
1.35 ±0.51
C
Females 3rd molar eruption
1/16 (10%)
3/17 (20%)
4/15 (30%)
6/12 (50%)
C
13/19 (70%)
C
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-4 DRAFT—DO NOT CITE OR QUOTE
-------
E.1.8. Keller et al. (2007, 2008a, b)
Endpoint
Administered Dose (ng/kg-day)
0
10 a
100
1,000
Internal Dose (ng/kg blood) b
0
296
2,365
18,764
Missing mandibular molars in CBA J mice
0/29 (0%)
2/23
(10%)
6/29
(20%)
30/30
(100%)
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
E.1.9. Kociba et al. (1978)
Endpoint
Administered Dose (ng/kg-day)
0
1 a
10b
100
Internal Dose (ng/kg blood)c
0
853
3,942
21,246
(n = 5)
(n = 5)
(n = 5)
(n = 5)
Urinary coproporphyrin,
females
9.8 ± 1.3
8.6 ±2
16.4 ± 4.7 d
17.4 ± 4 d
Uroporphyrin per creatinine,
females
0.157 ±0.05
0.143 ±0.04
0.181 ±0.05
0.296 ± 0.07d
aNOAEL
bLOAEL.
cFrom the Emond PRPK model described in 3.3.
d Statistically significant as compared to control (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-5 DRAFT—DO NOT CITE OR QUOTE
-------
l E.1.10. Latchoumycandane and Mathur (2002)
Administered Dose (ng/kg-day)
0
la
10
100
Internal Dose (ng/kg blood) b
0
437
2,579
15,092
Endpoint
(n = 6)
(n = 6)
(n = 6)
(n = 6)
Daily sperm production
22.19 ±2.67
15.67 ±2.65 c
13.65 ± 2.19 c
13.1 ±3.16c
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.05).
2 E.1.11. Lietal. (1997)
Endpoint
Administered Dose (ng/kg-day)
0
3a
10 b
30
100
300
1,000
3,000
10,000
30,000
Internal Dose (ng/kg blood)c
0
147
440
1,156
3,232
8,266
23,875
66,081
212,650
649,740
(n = 10)
(n =10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
FSH
23.86
±
29.65
22.16
±
48.51
85.23
±
94.33
73.30
±
48.51
126.14
±
159.01
132.10
±
115.89
116.76
±
51.21
304.26
±
153.62
346.88
±
150.93
455.11
±
285.68
aNOAEL
bLOAEL.
cFrom the Emond PRPK model described in 3.3.
3
4
5 E.1.12. Li et al. (2006)
Administered Dose (ng/kg-day)
0
2 a
50
100
Internal Dose (ng/kg blood) b
0
87.5
1,564
2,823
Endpoint
(n = 10)
(n = 10)
(n = 10)
(n = 10)
Serum estradiol
10. ± 12.48
20 ± 19.97
24.74 ± 15.00
17.90 ±
18.31
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-6 DRAFT—DO NOT CITE OR QUOTE
-------
Serum progesterone
65.25 ± 11.10
43.36 ±40.48
C
27.46 ±33.30
25.19 ±
43.756
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.01).
E.1.13. Markowski et al. (2001)
Endpoint
Administered Dose (ng/kg-day)
0
20 a
60
180
Internal Dose (ng/kg blood) b
0
1,234
3,184
8,152
(n = 7)
(n = 4)
(n = 6)
(n = 7)
FR10 run opp
13.29 ± 8.65
11.25 ± 5.56
5.75 ±3.53
7 ±6.01
FR2 revolutions
119.29 ±69.9
108.5 ±61
56.5 ±31.21
68.14 ±33.23
FR5 run opp
26.14 ± 12.28
23.5 ±7.04
12.8 ±6.17
13.14 ± 7.14
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
E.1.14. Mietinnin et al. (2006)
Administered Dose (ng/kg-day)
0
30 a
100
300
1,000
Internal Dose (ng/kg blood) b
0
1,756
4,922
12,657
36,874
Endpoint
(n = 42)
II
(n = 15)
II
G
(n = 32)
Cariogenic lesions in pups
25/42
23/29
19/25
20/24
29/32
(60%)
(79%) b
(76%)
(83%) c
(91%) c
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-7 DRAFT—DO NOT CITE OR QUOTE
-------
E.1.15. National Toxicology Program (1982)
Administered Dose (ng/kg-day)
0
1.43 a
7.14
71.4
Internal Dose (ng/kg blood) b
0
420
1,240
6,118
Endpoint
(n = 73)
(n = 49)
(n = 49)
(n = 50)
Toxic hepatitis, male mice
1/73 (1.4%)
5/49 (10%)
5/49 (6.1%)
44/50 (88%)
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
E.1.16. National Toxicology Program (2006)
Endpoint
Administered Dose (ng/kg-day)
0
2.14 a
7.14
15.7
32.9
71.4
Internal Dose (ng/kg blood) b
0
1,408
3,137
5,393
9,128
16,361
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
Alveolar metaplasia
2/53
(0%)
19/54
(40%)°'
33/53
(60%)c
35/52
(70%)c
45/53
(80%)c
46/52
(90%)c
Gingival hyperplasia
squamous, 2 years
1/53
(2%)
7/54
(13%)d
14/53
(26%)c
13/53
(25%)c
15/53
(28%)c
16/53
(30%)c
Liver, hepatocyte hypertrophy,
2 years
0/53
(0%)
19/54
(40%)c'
19/53
(40%)c
42/53
(80%)c
41/53
(80%)c
52/53
(100%)c
Heart, cardiomyopathy
10/53
(19%)
12/54
(22%)
22/53c
(42%)
25/52c
(48%)
3 2/5 3c
(60%)
36/52c
(69%)
Liver, eosinophilic focus,
multiple
3/53
(6%)
8/54
(15%)
14/53
(26%)
17/53
(32%)
22/53
(42%)
42/53
(79%)
Liver, fatty change, diffuse
0/53
(0%)
2/54
(4%)
12/5 3c
(23%)
17/53c
(32%)
3 0/5 3c
(57%)
48/53c
(91%)
Liver, necrosis
1/53
(2%)
4/54
(7%)
4/53
(8%)
8/53d
(15%)
10/5 3c
(19%)
17/53c
(32%)
Liver, pigmentation
4/53
9/54
3 4/5 3c
48/53c
52/53c
53/53c
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-8 DRAFT—DO NOT CITE OR QUOTE
-------
Endpoint
Administered Dose (ng/kg-day)
0
2.14 a
7.14
15.7
32.9
71.4
Internal Dose (ng/kg blood) b
0
1,408
3,137
5,393
9,128
16,361
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(8%)
(17%)
(64%)
(91%)
(98%)
(100%)
Liver, toxic hepatopathy
0/53
(0%)
2/54
(4%)
8/53
(15%)
30/53
(57%)
45/50
(85%)
53/53
(100%)
Oval cell hyperplasia, 2 years
0/53
(0%)
4/54
(10%)d
3/53
(10%)
20/53
(40%)c
38/53
(70%)d
53/53
(100%)c
Lung, alveolar to bronchiolar
epithelial metaplasia
(Alveolar epithelium,
metaplasia, bronchiolar)
2/53
(4%)
19/54 c
(35%)
33/53c
(62%)
35/52c
(67%)
45/53c
(85%)
46/52c
(89%)
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.01).
Statistically significant as compared to control (p < 0.05).
1
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-9 DRAFT—DO NOT CITE OR QUOTE
-------
E.1.17. Ohsako et al. (2001)
Endpoint
Administered Dose (ng/kg-day)
0
12.5 a
50 b
200
800
Internal Dose (ng/kg blood)c
0
845
2,763
9,022
30,504
(n = 12)
(n = 10)
(n = 10)
(n = 10)
(n = 12)
Anogenital PND120
28.91 ±
3.54
28.08 ±
2.52
25.31 ±
3.59 d
26.07 ±
3.59 e
23.87 ±
2.36 d
aNOAEL for selected endpoint.
b LOAEL for selected endpoint.
cFrom the Emond PRPK model described in 3.3.
d Statistically significant as compared to control (p < 0.01).
e Statistically significant as compared to control (p < 0.05).
E.1.18. Schantz et al. (1996)
Administered Dose (ng/kg-day)
0
25
100
Internal Dose (ng/kg blood) a
0
6,800
24,522
Endpoint
(n = 10)
(n = 10)
(n = 10)
Maze errors per block
3.55 ±0.64
2.76 ±0.81 b
2.34 ±0.81 c
aFrom the Emond PRPK model described in 3.3.
b Statistically significant as compared to control (p < 0.05).
c Statistically significant as compared to control (p < 0.001).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-10 DRAFT—DO NOT CITE OR QUOTE
-------
E.1.19. Shi et al. (2007)
Endpoint
Administered Dose (ng/kg-day)
0
0.143 a
0.714 b
7.14
28.6
Internal Dose (ng/kg blood)c
0
188
592
2,882
7,665
(n = 10)
(n = 10)
(n = 10)
(n = 10)
(n = 10)
Serum estradiol
102.86 ±
41.41
86.19 ±
19.58
63.33 ±
29.36 d
48.1 ±
18.82 d
38.57 ±
22.59 d
aNOAEL.
bLOAEL.
cFrom the Emond PRPK model described in 3.3.
d Statistically significant as compared to control (p < 0.05).
E.1.20. Smialowicz et al. (2008)
Endpoint
Administered Dose (ng/kg-day)
0
1.07 a
10.7
107
321
Internal Dose (ng/kg blood) b
0
241
1,358
7,385
17,438
(n = 15)
(n = 14)
(n = 15)
(n = 15)
(n = 8)
PFC per 106 Cells
1491 ±716
1129±171c
945±516c
677 ± 465 c
161± 117 c
PFC per spleen
27.8 ± 13.4
21 ± 13.6 c
17.6 ± 9.4 c
12.6 ± 8.7 c
3 ±3.1 c
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
l E.1.21. Toth et al. (1979)
Endpoint
Administered Dose (ng/kg-day)
0
la
100
1,000
Internal Dose (ng/kg blood) b
0
316
7,814
50,105
(n =38)
(n = 44)
(n = 44)
(n = 43)
Amyloidosis
0/38 (0%)
5/44(11%)
10/44 (23%)
17/43 (40%)
Skin Lesions
0/38 (0%)
5/44(11%)
13/44 (30%)
25/43 (58%)
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
2
3
4 E.1.22. Van Birgelen et al. (1995)
Endpoint
Administered Dose (ng/kg-day)
0
14 a
26
47
320
1,024
Internal Dose (ng/kg blood) b
0
3,969
6,479
9,968
47,606
137,820
n = 8
n = 8
n = 8
n = 8
n = 8
n = 8
Hepatic retinol
14.9 ±8.77
8.4 ± 3.39c
8.2 ± 2.26c
5.1 ± 0.85c
2.2 ± 0.85 c
0.6 ± 0.57c
Hepatic retinol palmitate
472 ±
271.53
94 ± 67.88c
107 ± 76.37c
74 ± 39.6c
22 ± 22.63c
3 ± 2.83c
Plasma FT4
23.4 ± 3.11
24.5 ± 5.66
22.4 ±2.83
19.3 ±9.33
16.3 ± 4.24c
10.3 ±4.81c
Plasma TT4
40.9 ±6.79
41.4 ±5.37
41.4 ± 6.51
32.3 ± 7.35c
33.6 ± 6.22c
25.5 ± 7.64c
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-12 DRAFT—DO NOT CITE OR QUOTE
-------
l E.1.23. White et al. (1986)
Administered Dose (ng/kg-day)
0
10 a
50
100
500
1,000
2,000
Internal Dose (ng/kg blood) b
0
602
2,250
3,934
14,772
26,844
49,896
Endpoint
(n = 8)
(n = 8)
(n = 8)
(n = 8)
(n = 8)
(n = 8)
(n = 8)
CH50
91 ± 14.14
54 ± 8.5c
63 ± llc
56 ± 26 c
41 ± 17 c
32 ± 17 c
17 ± 17 c
aLOAEL.
bFrom the Emond PRPK model described in 3.3.
c Statistically significant as compared to control (p < 0.05).
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-13 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2. ALTERNATE DOSE: BLOOD SERUM BMDS RESULTS
2 E.2.1. Amin et al. (2000): Saccharin Consumed, Female (0.25%)
3 E.2.1.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
freedom
Variance
/7-value a
x2 Test
statistic
X2P-
value
b
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
Linearc
1
0.00
0.35
0.55
179.21
7.2E+03
4.8E+03
nonconstant
variance
Polynomial
1
0.00
0.35
0.55
179.21
7.2E+03
4.8E+03
nonconstant
variance
Power
1
0.00
0.35
0.55
179.21
7.2E+03
4.8E+03
nonconstant
variance, power
restricted >1,
bound hit
Power d
0
0.00
0.00
NA
180.86
6.6E+03
2.7E+03
nonconstant
variance, power
unrestricted
Linear
1
0.00
0.00
0.95
191.69
5.3E+03
3.5E+03
constant variance
Polynomial
1
0.00
0.00
0.95
191.69
5.3E+03
3.5E+03
constant variance
Power
1
0.00
0.00
0.95
191.69
5.3E+03
3.5E+03
constant variance,
power restricted
>1, bound hit
Power
0
0.00
0.00
NA
193.68
5.2E+03
1.3E+03
constant variance,
power
unrestricted
4
5 aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a
6 constant variance model should be selected.
7 bValues <0.1 fail to meet BMDS goodness-of-fit criteria.
8 cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix.
9 dAlternate model also presented in this appendix.
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.1.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
50 .
Linear
0 1000 2000 3000 4000 5000 6000 7000 8000
dose
13:42 11/16 2009
BMDL BMD
E.2.1.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_25_s_c.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_25_s_c.plt
Mon Nov 16 13:42:20 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose^2 + ...
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 5.29482
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
beta_0
beta 1
31. 5152
-0.0025051
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
beta_0
beta 1
lalpha
1
-0. 99
-0.029
0. 044
rho
-0. 99
1
0. 026
-0. 04
beta_0
-0.029
0. 026
1
-0. 94
beta_l
0. 044
-0. 04
-0. 94
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
-2.542
2.40977
31.2702
-0.00246009
Std. Err.
1.65042
0.541752
4.19399
0.000552567
Lower Conf. Limit
-5.77677
1.34795
23.0501
-0.0035431
Upper Conf. Limit
0.692762
3. 47158
39.4903
-0.00137708
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
2670 10
8341 10
31. 7
24 . 6
10.7
31. 3
24 . 7
10.8
20.6
12
5.33
17 . 8
13. 4
4 . 91
0.0717
-0.0253
-0.0363
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -92.841935 4 193.683870
A2 -85.255316 6 182.510632
A3 -85.429148 5 180.858295
fitted -85.605740 4 179.211479
R -98.136607 2 200.273213
Explanation of Tests
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 25.7626 4 <.0001
Test 2 15.1732 2 0.0005072
Test 3 0.347663 1 0.5554
Test 4 0.353184 1 0.5523
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 7 219.6 9
BMDL = 4 809.97
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.1.4. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
35
30
25
20
15
10
BMDL
BMD
0 1000 2000 3000 4000 5000 6000 7000 8000
dose
13:42 11/16 2009
E.2.1.5. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Pwr_Unrest_BMRl_25_s_c.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Pwr_Unrest_BMRl_25_s_c.plt
Mon Nov 16 13:42:20 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha = 5.29482
rho = 0
control = 31.6727
slope = -0.00381519
power = 0.953851
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 99
0. 34
-0.095
-0.061
rho
-0. 99
1
-0.42
0.11
0. 068
control
0. 34
-0.42
1
-0. 61
-0. 56
slope
-0.095
0.11
-0. 61
1
1
power
-0.061
0. 068
-0. 56
1
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
-2 .48291
2.38455
32 . 99
-0.0286289
0.736753
Std. Err.
2.08669
0.692047
5. 40753
0.0946744
0.351085
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-6.57274
1.02817
22.3915
-0.214187
0.0486403
1.60692
3. 74094
43.5886
0.156929
1. 42487
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
2670 10
8341 10
31. 7
24 . 6
10.7
33
23. 4
10.8
20.6
12
5. 33
18 . 7
12 . 4
4 . 94
-0.223
0. 302
-0. 08
Warning: Likelihood for fitted model larger than the Likelihood for model A3.
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2 : Yij = Mu (i ) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*In(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/N2
Likelihoods of Interest
Model
Log(likelihood)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
A1 -92.841935 4 193.683870
A2 -85.255316 6 182.510632
A3 -85.429148 5 180.858295
fitted -85.429148 5 180.858295
R -98.136607 2 200.273213
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 25.7626 4 <.0001
Test 2 15.1732 2 0.0005072
Test 3 0.347663 1 0.5554
Test 4 -8.2423e-013 0 NA
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
NA - Degrees of freedom for Test 4 are less than or equal to 0. The Chi-Square
test for fit is not valid
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 6606.37
BMDL = 2702.55
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-20 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.2. Amin et al. (2000): Saccharin Consumed, Female (0.50%)
2 E.2.2.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Valuea
x2 Test
Statistic
1 p-
Value
b
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linear0
1
<.0001
3.52
0.06
158.58
8.0E+03
5.2E+03
nonconstant
variance
polynomial
1
<0001
3.52
0.06
158.58
8.0E+03
5.2E+03
nonconstant
variance
power
1
<0001
3.52
0.06
158.58
8.0E+03
5.2E+03
nonconstant
variance, power
restricted >1,
bound hit
power d
0
<0001
0.00
NA
157.06
5.2E+03
9.1E+02
nonconstant
variance, power
unrestricted
3
4 aValues <0.1 means nonconstant variance model should be selected; values >0.1 means a
5 constant variance model should be selected
6 bValues <0.1 fail to meet BMDS goodness-of-fit criteria
7 cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
8 dAlternate model also presented in this appendix
9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-21 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.2.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
35
30
25
20
15
10
Linear
BMDL
1000 2000 3000 4000 5000 6000
dose
7000 8000
13:41 11/16 2009
E.2.2.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_50_s_c.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_50_s_c.plt
Mon Nov 16 13:41:55 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = beta 0 + beta l*dose + beta 2*dose/N2 +
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
rho)
Default Initial Parameter Values
lalpha = 4.68512
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
beta_0
beta 1
20.0674
-0.00199124
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
beta_0
beta 1
lalpha
1
-0. 96
0. 019
-0.0016
rho
-0. 96
1
-0.031
0. 015
beta_0
0. 019
-0.031
1
-0. 96
beta_l
-0.0016
0. 015
-0. 96
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
-0.981979
2.11795
18.6205
-0.00168815
Std. Err.
0.982197
0. 401142
3.17872
0.000408035
Lower Conf. Limit
-2.90705
1.33173
12.3903
-0.00248788
Upper Conf. Limit
0.943092
2.90417
24.8507
-0.000888416
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
2670 10
8341 10
22 . 4
11. 4
4 . 54
18 . 6
14 .1
4 . 54
16
7 . 66
3.33
13.5
10.1
3. 04
0. 872
-0.855
-0.00339
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -83.696404 4 175.392808
A2 -73.511830 6 159.023660
A3 -73.530233 5 157.060467
fitted -75.291848 4 158.583695
R -90.294746 2 184.589492
Explanation of Tests
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 33.5658 4 <.0001
Test 2 20.3691 2 <.0001
Test 3 0.0368066 1 0.8479
Test 4 3.52323 1 0.06051
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 8021.29
BMDL = 5183.12
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.2.4. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
35
30
25
20
15
10
Power
BMDL
1000 2000 3000 4000 5000
dose
6000 7000
8000
13:41 11/16 2009
E.2.2.5. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Pwr_Unrest_BMRl_50_s_c.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Pwr_Unrest_BMRl_50_s_c.plt
Mon Nov 16 13:41:56 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(
exp(lalpha + log(mean(i)
rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha
rho
control
slope
power
4 . 68512
0
22 . 3564
-0.381559
0. 42572
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 96
0.34
-0.2
-0.15
rho
-0. 96
1
-0.47
0.23
0.15
control
0.34
-0.47
1
-0. 63
-0.52
slope
-0.2
0.23
-0. 63
1
0. 99
power
-0.15
0.15
-0.52
0. 99
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
-0.708629
1.96142
22.6293
-0.50513
0.396043
Std. Err.
1.298
0.529653
4.48415
0. 841243
0.168878
95.0% Wald Confidence Interval
Lower Conf. Limit
-3.25267
0. 923323
13.8405
-2.15394
0. 0650481
Upper Conf. Limit
1.83541
2.99953
31.4181
1.14368
0.727037
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
2670 10
8341 10
22 . 4
11. 4
4 . 54
22 . 6
11.1
4 . 58
Degrees of freedom for Test A3 vs fitted
16
7 . 66
3.33
0
15
7.46
3.12
-0.0577
0.105
-0.0475
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/N2
Model
Likelihoods of Interest
Log(likelihood)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-26 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
A1 -83.696404 4 175.392808
A2 -73.511830 6 159.023660
A3 -73.530233 5 157.060467
fitted -73.530233 5 157.060467
R -90.294746 2 184.589492
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 33.5658 4 <.0001
Test 2 20.3691 2 <.0001
Test 3 0.0368066 1 0.8479
Test 4 0 0 NA
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
NA - Degrees of freedom for Test 4 are less than or equal to 0. The Chi-Square
test for fit is not valid
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 5186.92
BMDL = 913.947
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-27 DRAFT—DO NOT CITE OR QUOTE
-------
E.2.3. Amin et al. (2000): Saccharin Preference Ratio, Female (0.25%)
E.2.3.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Valuea
x2 Test
Statistic
t2 P-
Valueb
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linear0
1
0.01
9.51
0.00
227.81
9.2E+03
4.4E+03
nonconstant
variance
polynomial
1
0.01
9.51
0.00
227.81
9.2E+03
4.4E+03
nonconstant
variance
power
1
0.01
9.51
0.00
227.81
9.2E+03
4.4E+03
nonconstant
variance, power
restricted >1,
bound hit
power d
1
0.01
1.22
0.27
219.52
8.3E+05
error
nonconstant
variance, power
unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant
variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.3.2.
Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
90
80
70
60
50
40
30
Linear
2000
4000
6000
8000
dose
13:41 11/16 2009
E.2.3.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_25_s_p_f.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_25_s_p_f.plt
Mon Nov 16 13:41:29 2009
Rel Male Thymus wt Tbl 2
The form of the response function is:
Y[dose] = beta 0 + beta l*dose + beta 2*dose/N2 +
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 6.34368
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
beta_0
beta 1
75.4969
-0.00284822
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
beta_0
beta 1
lalpha
1
-1
0.22
-0.31
rho
-1
1
-0.22
0.31
beta_0
0.22
-0.22
1
-0.77
beta_l
-0.31
0.31
-0.77
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
3.02282
0.793523
75.1183
-0.00274398
Std. Err.
9.21151
2.21122
6.74307
0.00127757
Lower Conf. Limit
-15.0314
-3.54039
61.9021
-0.00524797
Upper Conf. Limit
21.077
5.12744
88.3345
-0.000239995
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
2670 10
8341 10
82 .1
58 .1
54 . 9
75.1
67 . 8
52 . 2
13.3
33. 9
19.5
25.2
24 . 2
21. 8
0. 883
-1. 27
0.383
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -108.574798 4 225.149597
A2 -104.269377 6 220.538754
A3 -105.147952 5 220.295903
fitted -109.902600 4 227.805201
R -112.382522 2 228.765045
Explanation of Tests
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 16.2263 4 0.00273
Test 2 8.61084 2 0.0135
Test 3 1.75715 1 0.185
Test 4 9.5093 1 0.002044
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 9167.26
BMDL = 4 3 94.21
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.3.4. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model
90
80
70
60
50
40
30
Power
13:41 11/16 2009
E.2.3.5. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Pwr_Unrest_BMRl_25_s_p_f.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Pwr_Unrest_BMRl_25_s_p_f.plt
Mon Nov 16 13:41:30 2009
Rel Male Thymus wt Tbl 2
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-32 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha
rho
control
slope
power
6.34368
0
82 .1429
-9.98589
0.111278
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -power
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha rho control slope
lalpha 1 -1 -0.26 0.64
rho -1 1 0.27 -0.63
control -0.26 0.27 1 -0.56
slope 0.64 -0.63 -0.56 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
Estimate
Std
Err.
Lower Conf. Limit
Upper Conf. Limit
lalpha
22 . 273
8
00764
6.57832
37.9677
rho
-3.90063
1
89036
-7.60567
-0.195596
control
82 .1429
4
00411
74.295
89.9908
slope
-25.6494
7
11029
-39.5853
-11.7135
power
0
NA
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
2 N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10 82.1 56.5 13.3 26.3 3.09
2670 10 58.1 56.5 33.9 26.3 0.195
8341 10 54.9 56.5 19.5 26.3 -0.195
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-33 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -108.574798 4 225.149597
A2 -104.269377 6 220.538754
A3 -105.147952 5 220.295903
fitted -105.759821 4 219.519641
R -112.382522 2 228.765045
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 16.2263 4 0.00273
Test 2 8.61084 2 0.0135
Test 3 1.75715 1 0.185
Test 4 1.22374 1 0.2686
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Since the power was estimated to be 0, the BMD is infinite.
Setting BMD = 100*(maximum dose).
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 834 064
Warning: optimum may not have been found. Bad completion code in Optimization routine.
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-34 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.4. Amin et al. (2000): Saccharin Preference Ratio, Female (0.50%)
2 E.2.4.1. Summary Table of BMDS Modeling Results
Saccharin preference ratio, female (0.50%) (Amin et al., 2000)
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linearc
1
0.56
2.23
0.14
234.25
6.4E+03
4.0E+03
constant variance
polynomial
1
0.56
2.23
0.14
234.25
6.4E+03
4.0E+03
constant variance
power
1
0.56
2.23
0.14
234.25
6.4E+03
4.0E+03
constant variance,
power restricted >1,
bound hit
power d
0
0.56
0.00
NA
234.02
2.1E+03
1.3E-05
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
3
4
5
6 E.2.4.2. Figure for Selected Model: Linear, Constant Variance
Linear Model with 0.95 Confidence Level
Linear
90
80
70
60
50
40
30
20
BMDL
BMD
10
0
1000
2000
3000
4000
5000
6000
7000
8000
dose
7 13:41 11/16 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-35 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.2.4.3. Output File for Selected Model: Linear, Constant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LinearCV_BMRl_5 0_s_p_f.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\LinearCV_BMRl_50_s_p_f.plt
Mon Nov 16 13:41:03 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose/s2 + ...
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Signs of the polynomial coefficients are not restricted
A constant variance model is fit
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 764.602
rho = 0 Specified
beta_G = 65.8731
beta 1 = -0.00423638
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
alpha
1
beta_0
-4 . 3e-009
beta_l
-3.4e-010
beta_0
beta 1
-4 . 3e-009
-3.4e-010
1
-0.73
-0.73
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
alpha 741.152 191.365 366.084 1116.22
beta_0 65.8731 7.22637 51.7096 80.0365
beta 1 -0.00423638 0.00142921 -0.00703759 -0.00143517
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-36 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0 10 72.7
2670 10 44.5
8341 10 33.8
65.9 24.6
54.6 32.9
30.5 24.6
27.2 0.796
27.2 -1.17
27.2 0.375
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma'"2
Model A2: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -113.009921 4 234.019841
A2 -112.428886 6 236.857773
A3 -113.009921 4 234.019841
fitted -114.123097 3 234.246193
R -117.976057 2 239.952114
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
11.0943
1.16207
1.16207
2 . 22635
0.02552
0.5593
0.5593
0.1357
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-37 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 6426.26
BMDL
4028.71
E.2.4.4. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
100
90
80
70
60
50
40
30
20
10,
B ylDL
13:41 11/16 2009
Power Model with 0.95 Confidence Level
Power
1000
2000 3000
4000 5000
dose
6000 7000
8000
E.2.4.5. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\PwrCV_Unrest_BMRl_50_s_p_f.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\PwrCV_Unrest_BMRl_50_s_p_f.plt
Mon Nov 16 13:41:04 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-38 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
rho
control
slope
power
764.602
0
72.7273
-3.04504
0.282321
Specified
Asymptotic Correlation Matrix of Parameter Estimates
alpha
control
The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
1
-2 . 2e-008
control
-2 . 2e-008
1
slope
3.9e-009
-0.3
power
1.5e-009
-0.22
slope 3.9e-009 -0.3 1 0.99
power 1.5e-009 -0.22 0.99 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
688.142
72.7273
-3.04504
0.282321
Std. Err.
177.677
8.29543
8.78405
0.326249
Lower Conf. Limit
339. 9
56.4686
-20.2615
-0.357114
Upper Conf. Limit
1036.38
88 . 986
14 .1714
0. 921757
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10 72.7 72.7 24.6 26.2 8.48e-008
2670 10 44.5 44.5 32.9 26.2 -1.25e-008
8341 10 33.8 33.8 24.6 26.2 -3.93e-008
Degrees of freedom for Test A3 vs fitted <= 0
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -113.009921 4 234.019841
A2 -112.428886 6 236.857773
A3 -113.009921 4 234.019841
fitted -113.009921 4 234.019841
R -117.976057 2 239.952114
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
11.0943
1.16207
1.16207
0
0.02552
0.5593
0.5593
NA
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
NA - Degrees of freedom for Test 4 are less than or egual to 0. The Chi-Sguare
test for fit is not valid
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 2054.47
BMDL = 1.26421e-005
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-40 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 E.2.5. Bell et al. (2007): Balano-Preputial Separation in Male Pups (10% extra risk)
3 E.2.5.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
X
p-Valuea
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
2
0.69
0.71
112.07
2.5E+03
1.7E+03
power restricted >1, bound
hit
logistic
2
2.10
0.35
113.86
5.3E+03
4.1E+03
log-logistic b
2
0.47
0.79
111.88
2.0E+03
1.2E+03
slope restricted >1,
bound hit
log-logisticc
1
0.44
0.51
113.86
1806
264.4
slope unrestricted
log-probit
1
0.54
0.46
113.96
1.8E+03
3.1E+02
slope restricted >1
multistage, 1-
degree
2
0.69
0.71
112.07
2.5E+03
1.7E+03
betas restricted >0, bound
hit
probit
2
1.96
0.38
113.65
5.0E+03
3.8E+03
Weibull
2
0.69
0.71
112.07
2.5E+03
1.7E+03
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-41 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.5.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
BMDL BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
11:35 11/29 2009
E.2.5.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\Blood\LogLogistic_BMR2_BPS_d49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\Blood\LogLogistic_BMR2_BPS_d4 9.plt
Sun Nov 29 11:35:46 2009
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-42 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Default Initial Parameter Values
background = 0.0333333
intercept = -9.77382
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.48
intercept -0.48 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0371259 * * *
intercept -9.77952 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -53.9377 2 0.460052 2 0.7945
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 111.875
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0371
1.114
1. 000
30
-0.110
1997.8780
0.1349
4.048
5. 000
30
0.509
4539.2839
0.2339
7 . 018
6. 000
30
-0.439
15952.0000
0.4940
14 . 820
15.000
30
0. 066
Chi'" 2 = 0.47 d.f. = 2 P-value = 0.7914
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1963.13
BMDL = 1223.41
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-43 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.5.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
Log-Logistic Model with 0.95 Confidence Level
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Log-Logistic
3MDL
BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
11:35 11/29 2009
E.2.5.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\Blood\LogLogistic_Unrest_BMR2_BPS_d49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\Blood\LogLogistic_Unrest_BMR2_BPS_d4 9.plt
Sun Nov 29 11:35:48 2009
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-44 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Default Initial Parameter Values
background = 0.0333333
intercept = -8.67441
slope = 0.877 628
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.38 0.34
intercept -0.38 1 -1
slope 0.34 -1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0352883 * * *
intercept -9.31114 * * *
slope 0.948644 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -53.928 3 0.440703 1 0.5068
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 113.856
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0353
1. 059
1. 000
30
-0.058
1997.8780
0.1404
4 . 212
5. 000
30
0.414
4539.2839
0.2382
7 .145
6. 000
30
-0.491
15952.0000
0.4861
14.584
15.000
30
0.152
Chi/N2 = 0.44 d.f. = 1 P-value = 0.5076
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 18 06.29
BMDL = 264.35
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-45 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.6. Bell et al. (2007): Balano-Preputial Separation in Male Pups (5% extra risk)
2 E.2.6.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
1 P"
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
2
0.69
0.71
112.07
1.2E+03
8.2E+02
power restricted >1, bound
hit
logistic
2
2.10
0.35
113.86
3.0E+03
2.3E+03
log-logistic b
2
0.47
0.79
111.88
9.3E+02
5.8E+02
slope restricted >1,
bound hit
log-logisticc
1
0.44
0.51
113.86
8.2E+02
4.5E+01
slope unrestricted
log-probit
1
0.54
0.46
113.96
9.5E+02
7.2E+01
slope restricted >1
multistage, 1-
degree
2
0.69
0.71
112.07
1.2E+03
8.2E+02
betas restricted >0, bound
hit
probit
2
1.96
0.38
113.65
2.8E+03
2.1E+03
Weibull
2
0.69
0.71
112.07
1.2E+03
8.2E+02
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
3
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-46 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.6.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Log-Logistic
BMDL BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
11:35 11/29 2009
E.2.6.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\Blood\LogLogistic_BMRl_BPS_d49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\Blood\LogLogistic_BMRl_BPS_d4 9.plt
Sun Nov 29 11:35:45 2009
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-47 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Default Initial Parameter Values
background = 0.0333333
intercept = -9.77382
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.48
intercept -0.48 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0371259 * * *
intercept -9.77952 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -53.9377 2 0.460052 2 0.7945
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 111.875
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0371
1.114
1. 000
30
-0.110
1997.8780
0.1349
4.048
5. 000
30
0.509
4539.2839
0.2339
7 . 018
6. 000
30
-0.439
15952.0000
0.4940
14 . 820
15.000
30
0. 066
Chi'" 2 = 0.47 d.f. = 2 P-value = 0.7914
Benchmark Dose Computation
Specified effect = 0.05
Risk Type = Extra risk
Confidence level = 0.95
BMD = 929.901
BMDL = 57 9.512
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.6.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
Log-Logistic Model with 0.95 Confidence Level
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Log-Logistic
i/IDL BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
11:35 11/29 2009
E.2.6.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\Blood\LogLogistic_Unrest_BMRl_BPS_d49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\Blood\LogLogistic_Unrest_BMRl_BPS_d4 9.plt
Sun Nov 29 11:35:47 2009
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-49 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Default Initial Parameter Values
background = 0.0333333
intercept = -8.67441
slope = 0.877 628
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.38 0.34
intercept -0.38 1 -1
slope 0.34 -1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0352883 * * *
intercept -9.31114 * * *
slope 0.948644 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -53.928 3 0.440703 1 0.5068
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 113.856
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0353
1. 059
1. 000
30
-0.058
1997.8780
0.1404
4 . 212
5. 000
30
0.414
4539.2839
0.2382
7 .145
6. 000
30
-0.491
15952.0000
0.4861
14.584
15.000
30
0.152
Chi/N2 = 0.44 d.f. = 1 P-value = 0.5076
Benchmark Dose Computation
Specified effect = 0.05
Risk Type = Extra risk
Confidence level = 0.95
BMD = 821.69
BMDL = 45.4 953
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-50 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.7. Cantoni et al. (1981): Urinary Copro-Porhyrins
2 E.2.7.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.00
11.91
0.00
32.88
1.8E+04
8.6E+03
nonconstant variance,
power restricted >1
exponential (M3)
2
0.00
11.91
0.00
32.88
1.8E+04
8.6E+03
nonconstant variance,
power restricted >1
exponential (M4)
C
1
0.00
0.48
0.49
23.46
2.9E+02
9.9E+01
nonconstant
variance, power
restricted >1
exponential (M5)
1
0.00
0.48
0.49
23.46
2.9E+02
9.9E+01
nonconstant variance,
power restricted >1
exponential (M5)d
1
0.00
0.48
0.49
23.46
2.9E+02
9.9E+01
nonconstant variance,
power unrestricted
Hill
1
0.00
0.07
0.79
23.05
2.4E+02
error
nonconstant variance,
n restricted >1, bound
hit
Hilld
0
0.00
0.00
NA
24.97
1.4E+02
error
nonconstant variance,
n unrestricted
linear
2
0.00
10.62
0.00
31.59
8.1E+03
1.5E+03
nonconstant variance
polynomial
2
0.00
10.62
0.00
31.59
8.1E+03
1.5E+03
nonconstant variance
power
2
0.00
10.62
0.00
31.59
8.1E+03
1.5E+03
nonconstant variance,
power restricted >1,
bound hit
power d
1
0.00
0.26
0.61
23.23
1.5E+01
2.3E-06
nonconstant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
3
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-51 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.7.2.
Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
EMDL BMD
Exponential
13:43 11/16 2009
5000
10000 15000
dose
20000
25000
E.2.7.3. Output File for Selected Model: Exponential (M4'), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_BMRl_urin_copropor.(d)
Gnuplot Plotting File:
Mon Nov 16 13:43:37 2009
Figurel-UrinaryCoproporphyrin_3months
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-52 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
lnalpha
rho
Model 4
-1.50063
2.60979
0.704303
0.000109864
4 .47268
1
Parameter Estimates
Variable
lnalpha
rho
Model 4
-1.75303
2.63218
0.76122
0.000438426
4 .15614
1
Table of Stats From Input Data
Dose
N
0 4
1018 4
4868 4
2 . 756e + 004
Obs Mean
0.7414
1.807
2 .734
3
Obs Std Dev
0.3475
0.8341
1. 506
2 . 6
Dose
0
1018
4868
2 . 756e + 004
Estimated Values of Interest
Est Mean Est Std Scaled Residual
0.7612
1. 626
2.879
3.164
0.2907
0.7892
1. 674
1. 895
-0.1366
0. 4589
-0.1742
-0.1727
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-53 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Model R: Yij = Mu + e(i
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -12.90166 5 35.80333
A2 -6.203643 8 28.40729
A3 -6.487204 6 24.97441
R -15.73713 2 35.47427
4 -6.729565 5 23.45913
Additive constant for all log-likelihoods = -14.7. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 4 fit the data? (A3 vs 4)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 19.07 6 0.004052
Test 2 13.4 3 0.003854
Test 3 0.5671 2 0.7531
Test 6a 0.4847 1 0.4863
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 2 94.122
BMDL = 9 9.3366
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-54 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.7.4.
Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
EMDL BMD
5000 10000 15000
dose
20000
25000
13:43 11/16 2009
E.2.7.5. Output file for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_Unrest_BMRl_urin_copropor.(d)
Gnuplot Plotting File:
Mon Nov 16 13:43:39 2009
Figurel-UrinaryCoproporphyrin_3months
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-55 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
lnalpha
rho
Model 5
-1.50063
2.60979
0.704303
0.000109864
4 .47268
1
Parameter Estimates
Variable
lnalpha
rho
Model 5
-1.75303
2.63218
0.76122
0.000438426
4 .15614
1
Table of Stats From Input Data
Dose
N
0 4
1018 4
4868 4
2 . 756e + 004
Obs Mean
0.7414
1.807
2 .734
3
Obs Std Dev
0.3475
0.8341
1. 506
2 . 6
Dose
0
1018
4868
2 . 756e + 004
Estimated Values of Interest
Est Mean Est Std Scaled Residual
0.7612
1. 626
2.879
3.164
0.2907
0.7892
1. 674
1. 895
-0.1366
0. 4589
-0.1742
-0.1727
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-56 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Model R: Yij = Mu + e(i
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -12.90166 5 35.80333
A2 -6.203643 8 28.40729
A3 -6.487204 6 24.97441
R -15.73713 2 35.47427
5 -6.729565 5 23.45913
Additive constant for all log-likelihoods = -14.7. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 19.07 6 0.004052
Test 2 13.4 3 0.003854
Test 3 0.5671 2 0.7531
Test 7a 0.4847 1 0.4863
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 2 94.122
BMDL = 9 9.3366
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-57 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.7.6.
Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
Hill
BMD
13:43 11/16 2009
5000
10000 15000
dose
20000
25000
E.2.7.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_urin_copropor.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_urin_copropor.plt
Mon Nov 16 13:43:40 2009
Figurel-UrinaryCoproporphyrin_3months
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 0.90039
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-58 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
intercept
0
0.741372
2.25875
0.0266478
8454.34
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
-0. 62
-0.53
-0.013
0. 027
-0.0092
rho
-0. 62
1
0.43
-0.2
-0.017
-0.051
intercept
-0.53
0.43
1
-0.081
0. 032
0. 011
v
-0.013
-0.2
-0.081
1
-0.88
0. 96
n
0. 027
-0.017
0. 032
-0.88
1
-0. 92
k
-0.0092
-0.051
0. 011
0. 96
-0. 92
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
-1.78758
2.64296
0.759014
3.18202
0.739248
3317.45
Std. Err.
0.616312
0.750855
0.14058
2.82949
0. 896737
9482.63
Lower Conf. Limit
-2.99553
1.17131
0. 483483
-2.36368
-1.01832
-15268.2
Upper Conf. Limit
-0.579633
4 .11461
1.03455
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 4 0.741 0.759 0.348 0.284 -0.124
1018 4 1.81 1.7 0.834 0.822 0.27
4868 4 2.73 2.57 1.51 1.43 0.224
2.7 56e + 004 4 3 3.39 2.6 2.05 -0.38
Warning: Likelihood for fitted model larger than the Likelihood for model A3.
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-59 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -12.901663 5 35.803325
A2 -6.203643 8 28.407287
A3 -6.487204 6 24.974409
fitted -6.487204 6 24.974409
R -15.737135 2 35.474269
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 19.067 6 0.004052
Test 2 13.396 3 0.003854
Test 3 0.567122 2 0.7531
Test 4 -1.9007e-013 0 NA
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
NA - Degrees of freedom for Test 4 are less than or equal to 0. The Chi-Square
test for fit is not valid
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 143.414
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-60 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.7.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
i/IDLBMD
5000
10000 15000
dose
20000
25000
13:43 11/16 2009
E.2.7.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Pwr_Unrest_BMRl_urin_copropor.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Pwr_Unrest_BMRl_urin_copropor.plt
Mon Nov 16 13:43:39 2009
Figurel-UrinaryCoproporphyrin_3months
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(
exp(lalpha + log(mean(i)
rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-61 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha = 0.90039
rho = 0
control = 0.741372
slope = 0.226515
power = 0.224 935
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 62
-0.53
-0. 03
0. 024
rho
-0. 62
1
0.43
0. 052
-0.16
control
-0.53
0.43
1
-0.15
0. 086
slope
-0. 03
0. 052
-0.15
1
-0. 98
power
0. 024
-0.16
0. 086
-0. 98
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
-1.78125
2.64332
0.75678
0.123953
0.304254
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
0.617808 -2.99213 -0.570369
0.744947 1.18325 4.10339
0.139979 0.482426 1.03113
0.145639 -0.161493 0.4094
0.135074 0.0395142 0.568993
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 4
1018 4
4868 4
2 . 756e + 004
0.741
1. 81
2 .73
0.757
1.78
2 . 4
3.54
0.348
0. 834
1. 51
0.284
0 . 877
1. 3
2 .18
-0.109
0. 0705
0.515
-0.493
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma'*-2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -12.901663 5 35.803325
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-62 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
A2 -6.203643 8 28.407287
A3 -6.487204 6 24.974409
fitted -6.617381 5 23.234762
R -15.737135 2 35.474269
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 19.067 6 0.004052
Test 2 13.396 3 0.003854
Test 3 0.567122 2 0.7531
Test 4 0.260353 1 0.6099
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 15.247
BMDL = 2.31222e-006
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-63 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 E.2.8. Cantoni et al. (1981): Urinary Porphyrins
3 E.2.8.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
C
2
<0.0001
16.12
0.00
55.46
2.1E+03
1.5E+03
nonconstant variance,
power restricted >1
exponential (M3)
2
<0.0001
16.12
0.00
55.46
2.1E+03
1.5E+03
nonconstant variance,
power restricted >1
exponential (M4)
1
<0.0001
17.85
<0.0001
59.19
1.4E+02
8.0E+01
nonconstant variance,
power restricted >1
exponential (M5)
0
<0.0001
17.74
N/A
61.08
1.6E+02
8.0E+01
nonconstant variance,
power restricted >1
Hill
0
<.0001
18.86
NA
62.20
3.4E+03
1.8E+03
nonconstant variance, n
restricted >1
linear
2
<.0001
17.85
0.00
57.19
1.4E+02
8.0E+01
nonconstant variance
polynomial
1
<.0001
16.63
<.0001
57.97
1.9E+02
8.9E+01
nonconstant variance
power
1
<.0001
17.74
<.0001
59.08
1.6E+02
8.0E+01
nonconstant variance,
power restricted >1
11 Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-64 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.8.2.
Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
350
300
250
200
150
100
50
0 5000 10000 15000 20000 25000
dose
13:28 11/16 2009
E.2.8.3. Output File for Selected Model: Exponential (M2'), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_BMRl_Urinary_porphyrins.(d)
Gnuplot Plotting File:
Mon Nov 16 13:28:56 2009
Table 1, dose converted to ng per kg per day
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-65 DRAFT—DO NOT CITE OR QUOTE
BMDL BMD
Exponential
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
lnalpha -3.57509
rho 2.23456
a 2.1565
b 3.4 9686e-008
c 91300.7
d 1
Parameter Estimates
Variable
lnalpha
rho
Model 2
-4 . 64559
3.18357
2 . 32146
2 . 5137 2e-00 9
838302
1.04944
Table of
Stats From Input
Data
Dose
N
Obs Mean
Obs Std
0
4
2 . 27
0.49
1018
4
5.55
o
CO
CJI
4868
3
7 . 62
1.79
2 . 756e + 004
3
196. 9
63.14
Dose
0
1018
4868
2 . 756e + 004
Estimated Values of Interest
Est Mean Est Std Scaled Residual
3.579
4 .152
7 .281
199.5
1.262
1.445
2 .411
49.25
-2.074
1. 936
0.2437
- 0.0 9 0 6 9
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-66 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Model R: Yij = Mu + e(i
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -51.42175 5 112.8435
A2 -15.31211 8 46.62422
A3 -15.66963 6 43.33925
R -68.75058 2 141.5012
2 -23.73172 4 55.46344
Additive constant for all log-likelihoods = -12.87. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 106.9 6 < 0.0001
Test 2 72.22 3 < 0.0001
Test 3 0.715 2 0.6994
Test 4 16.12 2 0.0003153
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is less than .1. Model 2 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 2070.13
BMDL = 1521.05
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-67 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.9. Crofton et al. (2005): Serum T4
2 E.2.9.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
8
0.76
44.20
<0.0001
516.36
6.3E+04
3.4E+04
constant variance,
power restricted >1
exponential (M3)
8
0.76
44.20
<0.0001
516.36
6.3E+04
3.4E+04
constant variance,
power restricted >1
exponential (M4)
C
7
0.76
2.29
0.94
476.45
2.9E+03
1.7E+03
constant variance,
power restricted >1
exponential (M5)
6
0.76
2.08
0.91
478.23
3.2E+03
1.7E+03
constant variance,
power restricted >1
exponential (M5)d
6
0.76
2.08
0.91
478.23
3.2E+03
1.7E+03
constant variance,
power unrestricted
Hill
6
0.76
1.29
0.97
477.45
3.2E+03
1.7E+03
constant variance, n
restricted >1
Hilld
6
0.76
1.29
0.97
477.45
3.2E+03
1.7E+03
constant variance, n
unrestricted
linear
8
0.76
50.31
<.0001
522.46
1.3E+05
9.7E+04
constant variance
polynomial
8
0.76
50.31
<.0001
522.46
1.3E+05
9.7E+04
constant variance
power
8
0.76
50.31
<.0001
522.46
1.3E+05
9.7E+04
constant variance,
power restricted >1,
bound hit
power d
7
0.76
16.95
0.02
491.10
1.4E+03
1.8E+02
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
3
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-68 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.9.2.
Figure for Selected Model: Exponential (M4), Constant Variance, Power Restricted
>1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
120
100
EMDL
BMD
0
50000
100000
150000
200000
250000
300000
dose
15:13 11/16 2009
E.2.9.3. Output File for Selected Model: Exponential (M4), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\ExpConstVar_BMRl_SerumT4.(d)
Gnuplot Plotting File:
Mon Nov 16 15:13:33 2009
0
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-69 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 10
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 4
lnalpha
rho(S)
5.47437
0
104.999
1.16502e-005
0.445764
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 4
5.50322
0
99.7846
0. 000149614
0.533127
1.19797
Table of Stats From Input Data
Dose
N
Obs Mean
Obs Std Dev
0
14
100
i—
(ji
11.15
6
96.27
14 . 98
269.2
12
98 . 57
18 .11
763
6
99.76
19. 04
1905
6
93.32
12 .11
5104
6
70. 94
12.74
1. 271e + 004
6
62 . 52
14.75
3.617e+004
6
52 . 68
22 .73
9.965e+004
6
54 . 66
19.71
3.215e+005
4
49.15
i—1
i—1
i—1
CJi
Estimated Values of
Interest
Dose
Est Mean
Est Std
Scaled Residi
0
100.3
15. 69
-0.07 977
11.15
100.3
15. 69
-0.6232
269.2
CO
CJI
CO
15. 69
-0.0008246
763
95.52
15. 69
0.6615
1905
89.21
15. 69
0.6428
5104
76. 04
15. 69
-0.7955
271e+004
60.7
15. 69
0.2839
617e+004
52 . 85
15. 69
-0.02601
965e+004
52 . 53
15. 69
0.3323
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-70 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
3 . 215e + 005
52 . 53
15. 69
-0.432
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'-2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -233.0774 11 488.1549
A2 -230.2028 20 500.4056
A3 -233.0774 11 488.1549
R -268.4038 2 540.8076
4 -234.2238 4 476.4476
Additive constant for all log-likelihoods = -66.16. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2^log(Likelihood Ratio)
76.4
5.749
5.749
2 . 293
p-value
18
9
9
7
< 0.0001
0.7647
0.7647
0.9419
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-71 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from contrui
Confidence Level = 0.950000
EMD = 2 8 60.82
E'.MDL = 1670.13
E.2.9.4. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
120
100
80
60
40
Exponential_beta Model 5 with 0.95 Confidence Level
20
Exponential
EMDL
BMD
50000 100000
150000
dose
200000 250000 300000
15:13 11/16 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-72 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.9.5. Output File for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\ExpConstVar_Unrest_BMRl_SerumT4.(d)
Gnuplot Plotting File:
Mon Nov 16 15:13:40 2009
0
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 10
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
Specified
lnalpha 5.47437
rho(S) 0
a 104.999
b 1.16502e-005
c 0.445764
d 1
Parameter Estimates
Variable Model 5
lnalpha 5.50322
rho 0
a 99.7846
b 0.000149614
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-73 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
C
d
0.533127
1.19797
Table of Stats From Input Data
Dose
0
11.15
269.2
763
1905
5104
1. 271e + 004
3.617e+004
9.965e+004
3.215e+005
14
12
Obs Mean
100
96.27
98 . 57
99.76
93.32
70. 94
62 . 52
52 . 68
54 . 66
49.15
Obs Std Dev
15. 44
14 . 98
18 .11
19. 04
12 .11
12.74
14.75
22 .73
19.71
11.15
Estimated Values of Interest
Dose
Est Mean
Est Std
Scaled Residual
0
11.15
269.2
763
1905
5104
1.271e+004
3.617e+004
9.965e+004
3.215e+005
99.78
99.76
98 . 8
96. 45
90.5
75.78
58 . 58
53.22
53.2
53.2
15. 67
15. 67
15. 67
15. 67
15. 67
15. 67
15. 67
15. 67
15. 67
15. 67
0.0512
-0.5465
-0.05054
0.5173
0.4419
-0.7573
0. 616
-0.08476
0.2291
-0.5174
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -233.0774 11 488.1549
A2 -230.2028 20 500.4056
A3 -233.0774 11 488.1549
R -268.4038 2 540.8076
5 -234.1158 5 478.2316
Additive constant for all log-likelihoods = -66.16. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-74 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
Test 1
Test 2
Test 3
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test
-2*log(Likelihood Ratio)
D. F.
p-value
Test 1
Test 2
Test 3
Test 7a
76.4
5.749
5.749
2 . 077
18
9
9
6
< 0.0001
0.7647
0.7647
0.9125
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD
3175.08
BMDL
1706.36
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-75 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.9.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
120
100
80
60
40
20
EMDL
50000
100000
150000
dose
200000
250000
300000
15:13 11/16 2009
E.2.9.7. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\HillConstVar_Unrest_BMRl_SerumT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\HillConstVar_Unrest_BMRl_SerumT4.plt
Mon Nov 16 15:13:42 2009
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 10
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-76 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
alpha
rho
intercept
276.969
0
9 9 9 9 9
-50.854
1.5549
4585.23
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
1. 9e-009
-2e-008
-1. le-008
1. le-008
intercept
1. 9e-009
1
-0.58
-0.3
-0.2
v
-2e-008
-0.58
1
0.6
-0.36
n
-1. le-008
-0.3
0.6
1
-0.34
k
1. le-008
-0.2
-0.36
-0.34
1
Parameter Estimates
Variable
alpha
intercept
Estimate
242 . 825
99.3375
-46.4797
1. 85655
4564.01
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
40.4708 163.504 322.146
2.66145 94.1212 104.554
5.51009 -57.2793 -35.6801
0.927361 0.0389606 3.67415
1406.15 1808 7320.02
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 14 100 99.3 15.4 15.6 0.159
11.15 6 96.3 99.3 15 15.6 -0.483
269.2 12 98.6 99.1 18.1 15.6 -0.116
763 6 99.8 97.7 19 15.6 0.321
1905 6 93.3 91.7 12.1 15.6 0.26
5104 6 70.9 73.7 12.7 15.6 -0.433
1.271e+004 6 62.5 58.9 14.8 15.6 0.568
3.617e+004 6 52.7 53.8 22.7 15.6 -0.181
9.965e+004 6 54.7 53 19.7 15.6 0.26
3.215e+005 4 49.1 52.9 11.1 15.6 -0.479
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A3 uses any fixed variance parameters that
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-77 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma""'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -233.077445 11 488.154889
A2 -230.202783 20 500.405566
A3 -233.077445 11 488.154889
fitted -233.724271 5 477.448543
R -268.403817 2 540.807634
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
76.4021
5.74932
5.74932
1.29365
18
9
9
<.0001
0.7647
0.7647
0. 972
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 3156.67
BMDL = 1668.07
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-78 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.9.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
120
100
0
50000
100000
150000
200000
250000
300000
dose
15:13 11/16 2009
E.2.9.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\PowerConstVar_Unrest_BMRl_SerumT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\PowerConstVar_Unrest_BMRl_SerumT4.plt
Mon Nov 16 15:13:43 2009
0
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 10
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-79 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
alpha
rho
control
slope
power
276.969
0 Specified
9 9 9 9 9
-0.28302
0. 42875
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
control
slope
power
alpha
1
3e-010
-1. 9e-010
-1. 9e-010
control
3e-010
1
-0.73
-0. 62
slope
-1.9e-010
-0.73
1
0. 98
power
-1. 9e-010
-0. 62
0. 98
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
301.804
103.499
-3.01678
0.242881
Std. Err.
50.3007
3.94867
1.68354
0.0442912
Lower Conf. Limit
203.217
95.76
-6.31645
0.156071
Upper Conf. Limit
400.392
111.239
0.282886
0.32969
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
11.15
269.2
763
1905
5104
1.271e+004
3.617e+004
9.965e+004
3.215e+005
14
100
103
15. 4
17
4
-0.754
6
96.3
CO
1—1
15
17
4
-0.256
12
<£>
CD
O
1—1
CO
i—1
CO
1—1
17
4
1. 36
6
99.8
CO
CO
19
17
4
1. 6
6
93.3
00
12 . 1
17
4
1. 23
6
70. 9
79.5
12 . 7
17
4
<—i
C\]
\—i
62 . 5
52 . 7
54 . 7
49.1
73.6
64 . 9
54 .1
37 . 9
14 . 8
22 . 7
19.7
11 . 1
17 .
17 .
17 .
17 .
-1 . 56
-1 .72
0 . 077
1 . 3
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R:
Yi = Mu + e(i)
Var{e(i)}
Sigma/N2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-80 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -233.077445 11 488.154889
A2 -230.202783 20 500.405566
A3 -233.077445 11 488.154889
fitted -241.552045 4 491.104090
R -268.403817 2 540.807634
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 76.4021 18 <.0001
Test 2 5.74932 9 0.7647
Test 3 5.74932 9 0.7647
Test 4 16.9492 7 0.01773
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1350.28
BMDL = 182.329
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-81 DRAFT—DO NOT CITE OR QUOTE
-------
2 E.2.10. Hojo et al. (2002): DRL Reinforce Per Min
3 E.2.10.1. Summary Table of BMDS Modeling Results
Model3
Degrees
of
Freedom
2
1 P"
Valueb
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
Hill
0
NA
6.465
1.320E+03
4.017E-04
linear
2
0.009
9.126
1.070E+04
4.762E+03
polynomial
2
0.009
9.126
1.070E+04
4.762E+03
power
2
0.009
9.126
1.070E+04
4.762E+03
power bound hit
exponential (M2)
2
0.007
9.614
1.284E+04
6.859E+03
exponential (M3)
1
0.001
12.870
2.720E+08
1.522E+05
exponential (M4)c
1
0.054
5.490
1.041E+03
4.944E+00
exponential (M5)
0
N/A
6.465
1.367E+03
1.245E+01
a Constant variance model selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model, BMDS output presented in this appendix
4
5
6
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-82 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.10.2. Figure for Selected Model: Exponental (M4)
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
BMD
0 1000 2000 3000 4000 5000 6000 7000 8000
dose
10:23 01/12 2010
Hojo et al., 2002: DRL reinforce per min
E.2.10.3. Output File for Selected Model: Exponential (M4)
Hojo et al., 2002: DRL reinforce per min
Exponential Model. (Version: 1.61; Date: 7/24/2009)
Input Data File: C:\l\Blood\21_Hojo_2002_DRL_rein_min_exp_ExpCV_l.(d)
Gnuplot Plotting File:
Tue Jan 12 10:23:58 2010
Table 5, values adjusted by a constant to allow exponential model
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign
exp{sign
[c—(c—1)
[c—(c—1)
b * dose}
(b * dose)^d}
exp{-b * dose}]
exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *In(Y[dose]))
rho is set to 0.
A constant variance model is fit.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-83 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 4
lnalpha
rho(S)
-1.29672
0
0.0817
0.000197777
16.3733
1
(S)
Specified
Parameter Estimates
Variable Model 4
lnalpha -1.11954
rho 0
a 0.054752
b 0.000895762
c 18.2107
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 5 0.086 0.448
1285 5 0.536 0.821
3295 6 1.274 0.54
8465 5 0.737 0.443
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 0.05475 0.5713 0.1223
1285 0.6991 0.5713 -0.6381
3295 0.9478 0.5713 1.398
8465 0.9966 0.5713 -1.016
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-84 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 3.11555 5 3.7689
A2 4.489557 8 7.020886
A3 3.11555 5 3.7689
R -2.435087 2 8.870174
4 1.255168 4 5.489665
Additive constant for all log-likelihoods = -19.3. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Test
Does Model 4 fit the data? (A3 vs 4)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 13.85 6 0.03137
Test 2 2.748 3 0.4321
Test 3 2.748 3 0.4321
Test 6a 3.721 1 0.05374
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is less than .1. Model 4 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1040.67
BMDL = 4.94408
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-85 DRAFT—DO NOT CITE OR QUOTE
-------
2 E.2.11. Hojo et al. (2002): DRL Response Per Min
3 E.2.11.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
J2 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.30
1.13
0.57
122.98
1.1E+03
error
constant variance,
power restricted >1
exponential (M3)
2
0.30
1.13
0.57
122.98
1.1E+03
error
constant variance,
power restricted >1
exponential (M4)
C
1
0.30
0.50
0.48
124.36
8.4E+02
5.6E+01
constant variance,
power restricted >1
exponential (M5)
0
0.30
0.50
N/A
126.35
2.0E+03
4.9E+01
constant variance,
power restricted >1
Hill
0
0.30
0.50
NA
126.35
2.9E+03
3.2E-11
constant variance, n
restricted >1
linear
2
0.30
11.00
0.00
132.86
3.7E+04
1.7E+04
constant variance
polynomial
2
0.30
11.00
0.00
132.86
3.7E+04
1.7E+04
constant variance
power
2
0.30
11.00
0.00
132.86
3.7E+04
1.7E+04
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-86 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.11.2. Figure for Selected Model: Exponential (M4)
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
30
0
-10
BMDL BMD
0
1000 2000 3000 4000 5000 6000 7000 8000
dose
10:25 01/12 2010
Hojo et al., 2002: DRL response per min
E.2.11.3. Output File for Selected Model: Exponential (M4)
Hojo et al., 2002: DRL response per min
Exponential Model. (Version: 1.61; Date: 7/24/2009)
Input Data File: C:\l\Blood\23_Hojo_2002_DRL_resp_min_exp_ExpCV_l.(d)
Gnuplot Plotting File:
Table 5, values adjusted by a constant to allow exponential model
The form of the response function by Model:
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *In(Y[dose]))
rho is set to 0.
Tue Jan 12 10:25:21 2010
Model 2
Model 3
Model 4
Model 5
Y[dose] = a * exp{sign * b * dose}
Y[dose] = a * exp{sign * (b * dose)^d}
Y[dose] = a * [c-(c-l) * exp{-b * dose}]
Y[dose] = a * [c-(c-l) * exp{-(b * dose)^d}]
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-87 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 4
lnalpha
rho(S)
4.51689
0
24.6362
0.00047963
0.0184785
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 4
4 . 54096
0
23.4674
0. 00203802
0.101322
1
Table of Stats From Input Data
Dose
0
1285
3295
8465
Obs Mean
23. 46
4 . 013
0.478
4 .594
Obs Std Dev
7 . 986
10. 96
7.194
15.23
Estimated Values of Interest
Dose
0
1285
3295
8465
Est Mean
23. 47
3. 914
2 . 403
2 . 378
Est Std
9. 684
9. 684
9. 684
9. 684
Scaled Residual
-0.001011
0.02275
-0.487
0.5117
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-88 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
Model R: Yij = Mu + e(i
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -57.92733 5 125.8547
A2 -56.09669 8 128.1934
A3 -57.92733 5 125.8547
R -64.49611 2 132.9922
4 -58.18012 4 124.3602
Additive constant for all log-likelihoods = -19.3. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 4 fit the data? (A3 vs 4)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 16.8 6 0.01005
Test 2 3.661 3 0.3004
Test 3 3.661 3 0.3004
Test 6a 0.5056 1 0.4771
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 301.607
BMDL = 7.54 952
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-89 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.12. Kattainen et al. (2001): 3rd Molar Mesio-Distal Length (Molar Development)
2 E.2.12.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
<0.0001
38.96
<0.0001
122.90
6.4E+04
3.8E+04
nonconstant variance,
power restricted >1
exponential (M3)
3
<0.0001
38.96
<0.0001
122.90
6.4E+04
3.8E+04
nonconstant variance,
power restricted >1
exponential (M4)
2
<0.0001
79.12
<0.0001
-80.75
error
error
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
13.81
0.00
146.06
8.5E+02
5.1E+02
nonconstant variance,
power restricted >1
Hillc
2
<0001
8.72
0.01
151.15
6.3E+02
3.4E+02
nonconstant variance,
n restricted >1, bound
hit
Hilld
1
<.0001
2.92
0.09
154.95
3.0E+00
2.9E-02
nonconstant variance, n
unrestricted
linear
3
<.0001
39.59
<.0001
122.28
7.4E+04
4.7E+04
nonconstant variance
polynomial
2
<.0001
36.61
<.0001
123.26
3.0E+04
1.4E+04
nonconstant variance
power
3
<.0001
39.59
<.0001
122.28
7.4E+04
4.7E+04
nonconstant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
3
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-90 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.12.2. Figure for Selected Model: Hill, Nonconstant Variance, n Restricted >1, Bound Hit
1.9
1.7
1.6
1.5
1.4
1.3
1.2
1.1
Hill Model with 0.95 Confidence Level
Hill
i/IDLBMD
0 20000 40000 60000 80000 100000 120000 140000 160000
dose
13:14 11/16 2009
E.2.12.3. Output File for Selected Model: Hill, Nonconstant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_BMRl_3rd_molar.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Hill_BMRl_3rd_molar.plt
Mon Nov 16 13:14:09 2009
Figure 3 female only
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter restricted to be greater than 1
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-91 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha
rho
intercept
-2 . 37155
0
1. 85591
-0.507874
0. 825979
4284.51
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha
rho
intercept
lalpha
1
-0. 98
-0.16
0.84
-0.37
rho
-0. 98
1
0.2
-0.79
0.39
intercept
-0.16
0.2
1
-0.31
-0.11
v
0.84
-0.79
-0.31
1
-0.48
k
-0.37
0.39
-0.11
-0.48
1
Parameter Estimates
Variable
lalpha
rho
intercept
k
Estimate
3.34591
-14 . 3329
1. 8548
-0.441028
1
3764.75
Std. Err.
1.40451
2.62142
0.0159016
0.0588146
NA
1228.49
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
0.593124
-19.4708
1.82364
-0.556302
1356.95
6.0987
-9.19505
1.88597
-0.325753
6172.54
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 16 1.86 1.85 0.0661 0.0637 0.0692
4703 17 1.58 1.61 0.185 0.176 -0.767
1.568e+004 15 1.6 1.5 0.265 0.293 1.28
4.7 25e + 004 12 1.5 1.45 0.221 0.378 0.527
1.585e+005 19 1.35 1.42 0.515 0.423 -0.783
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-92 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 56.758717 6 -101.517434
A2 85.856450 10 -151.712901
A3 84.934314 7 -155.868628
fitted 80.574735 5 -151.149469
R 45.373551 2 -86.747101
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 80.9658 8 <.0001
Test 2 58.1955 4 <.0001
Test 3 1.84427 3 0.6053
Test 4 8.71916 2 0.01278
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 634.985
BMDL = 335.87 9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-93 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.12.4. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
1.9
1.7
1.6
1.5
1.4
1.3
1.2
1.1
Hill
B^IDLBMD
0 20000 40000 60000 80000 100000 120000 140000 160000
dose
13:14 11/16 2009
E.2.12.5. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_3rd_molar.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_3rd_molar.plt
Mon Nov 16 13:14:09 2009
Figure 3 female only
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(
exp(lalpha + rho
In(mean(i)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = -2.37155
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-94 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
intercept
0
1. 85591
-0.507874
0. 825979
4284.51
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
-0. 98
-0.18
0.18
-0.28
-0.011
rho
-0. 98
1
0.22
-0.18
0.29
0. 012
intercept
-0.18
0.22
1
-0.026
-0.058
0.0019
v
0.18
-0.18
-0.026
1
0.52
-0. 96
n
-0.28
0.29
-0.058
0.52
1
-0.71
k
-0.011
0. 012
0.0019
-0. 96
-0.71
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
3.21867
-14.0858
1.85564
-2.44363
0.196843
2.75627e+008
Std. Err.
1.42224
2 . 68326
0.016023
2 . 81206
0. 0500689
2.27948e+009
Lower Conf. Limit
0.431136
-19.3449
1.82424
-7.95516
0.0987096
-4.19207e+009
Upper Conf. Limit
6.00621
-8.82671
1.88705
3.06791
0.294976
4 . 74332e + 009
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 16
4703 17
1.568e+004
4.725e+004
1.585e+005
1.86
1. 58
15 1.6
12 1.5
19 1.35
1.86
0.0661
1.6 0.185
1.54 0.265
1.48 0.221
1.4 0.515
0.0643
0.18
0.234
0.316
0.471
0.0163
-0.597
0. 856
0.259
-0.465
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-95 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Model Log(likelihood) # Param's AIC
A1 56.758717 6 -101.517434
A2 85.856450 10 -151.712901
A3 84.934314 7 -155.868628
fitted 83.472636 6 -154.945271
R 45.373551 2 -86.747101
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 80.9658 8 <.0001
Test 2 58.1955 4 <.0001
Test 3 1.84427 3 0.6053
Test 4 2.92336 1 0.08731
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 2.96451
BMDL = 0.0287389
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-96 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 E.2.13. Kattainen et al. (2001): Females 3rd Molar Eruption
3 E.2.13.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
2
1 P"
Value3
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
logistic
3
0.360
88.508
7.290E+03
5.273E+03
log-logistic b
3
0.982
85.227
1.896E+03
1.050E+03
slope bound hit
log-probit,
unrestricted
2
0.941
87.181
1.641E+03
1.895E+02
slope unrestricted
probit
3
0.379
88.352
6.958E+03
5.177E+03
multistage, 4-
degree
3
0.781
86.155
3.195E+03
2.076E+03
final B=0
log-logistic,
unrestrictedc
2
0.949
87.162
1.527E+03
1.456E+02
slope unrestricted
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model, BMDS output presented in this appendix
c Alternate model, BMDS output also presented in this appendix
4
5
6
7 E.2.13.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
0.8
0.6
0.4
0.2
0
BMDL BMD
0
5000
10000
15000
20000
25000
30000
35000
dose
8 10:26 01/12 2010
9
10 Kattainen et al., 2001: 3rd molar eruption in pups
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-97 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.13.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Kattainen et al., 2001: 3rd molar eruption in pups
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\l\Blood\24_Katt_2001_3molar_erup_LogLogistic_BMRl.(d)
Gnuplot Plotting File: C:\l\Blood\24_Katt_2001_3molar_erup_LogLogistic_BMRl.plt
_Tue Jan 12 10:26706 2010
Figure 2
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0.0625
intercept = -9.748
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.53
intercept -0.53 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0699182 * * *
intercept -9.74484 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-98 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Full model
Fitted model
Reduced model
-40.5286
-40.6136
-50.7341
85.2273
0.170098
20.411
0.9823
0.0004142
Goodness of Fit
Scaled
Dose
Est
. Prob.
Expected
Observed
Size
Residual
0.0000
0.
06 9 9
1.119
1.
000
16
-0.116
1763.4151
0.
1570
2 . 669
3.
000
17
0. 220
4943.6112
0.
2788
4 .182
4 .
000
15
-0.105
12712.0000
0.
4670
5. 604
6.
000
12
0. 229
37039.0000
0.
7066
13.426
13.
000
19
-0.215
Chi ^2 = 0.17
d.f. =
3 P
-value
= 0.9820
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
0.1
Extra risk
0. 95
1896.22
1049.96
E.2.13.4. Figure for Unrestricted Model: Log-Logistic, Unrestricted
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
BMDL BMD
0 5000 10000 15000 20000 25000 30000 35000
dose
10:26 01/12 2010
Kattainen et al., 2001: 3rd molar eruption in pups
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-99 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.13.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Kattainen et al., 2001: 3rd molar eruption in pups
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\l\Blood\24_Katt_2001_3molar_erup_LogLogistic_Unrest_BMRl.(d)
Gnuplot Plotting File: C:\l\Blood\24_Katt_2001_3molar_erup_LogLogistic_Unrest_BMRl.plt
_Tue Jan 12 10:26T07 2010
Figure 2
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0.0625
intercept = -8.7855
slope = 0.902051
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.43 0.38
intercept -0.43 1 -0.99
slope 0.38 -0.99 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0630017 * * *
intercept -8.87185 * * *
slope 0.910471 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -40.5286 5
Fitted model -40.5812 3 0.105153 2 0.9488
Reduced model -50.7341 1 20.411 4 0.0004142
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-100 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
AIC: 87.1623
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0630
1. 008
1
000
16
-0
008
1763.4151
0.1684
2 .862
3
000
17
0
089
4943.6112
0.2922
4 . 383
4
000
15
-0
218
12712.0000
0.4692
5. 630
6
000
12
0
214
37039.0000
0.6903
13.117
13
000
19
-0
058
Chi'"2 = 0.10 d.f. = 2 P-value = 0.94 91
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1526.84
BMDL = 14 5.591
E.2.14. Kattainen et al., 2001: 3rd molar length in pups
E.2.14.1. Summary Table of BMDS modeling results
Model3
Degrees
of
Freedom
2
X p-
Value"
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
exponential (M2)
3
<0.0001
124.869
1.319E+04
7.850E+03
exponential (M3)
3
<0.0001
124.869
1.319E+04
7.850E+03
power bound hit
exponential (M4)
2
0.002
147.122
3.351E+02
2.001E+02
exponential (M5)
2
0.002
147.122
3.351E+02
2.001E+02
power bound hit
Hillc
2
0.022
152.241
2.477E+02
1.328E+02
n lower bound hit
linear
3
<.0001
124.026
1.567E+04
1.009E+04
polynomial
4
<.0001
-84.747
error
error
power
3
<.0001
124.026
1.567E+04
1.009E+04
power bound hit
Hill, unrestricted d
1
<.0001
-78.747
2.007E+05
error
n unrestricted
a Non-constant variance model selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model, BMDS output presented in this appendix
d Alternate model, BMDS output also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-101 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.14.2. Figure for selected model: Hill
Hill Model with 0.95 Confidence Level
15000 20000
dose
10:26 01/12 2010
Kattainen et al., 2001: 3rd molar length in pups
E.2.14.3. Output for selected model: Hill
Kattainen et al., 2001: 3rd molar length in pups
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\l\Blood\25_Katt_2001_3molar_length_Hill_l.(d)
Gnuplot Plotting File: C:\l\Blood\25_Katt_2 001_3molar_length_Hill_l.plt
Tue Jan 12 10:26:49 2010
Figure 3 female only
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter restricted to be greater than 1
The variance is to be modeled as Var(i) = exp(lalpha
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
+ rho
In(mean(i)
Default Initial Parameter Values
lalpha = -2.37155
rho = 0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-102 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
intercept
1. 85591
-0.507874
0. 845971
1606.5
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha
rho
intercept
lalpha
1
-0. 98
-0.16
0.84
-0.38
rho
-0. 98
1
0.2
-0.79
0.4
intercept
-0.16
0.2
1
-0.3
-0.11
v
0.84
-0.79
-0.3
1
-0.52
k
-0.38
0.4
-0.11
-0.52
1
Parameter Estimates
Variable
lalpha
rho
intercept
k
Estimate
3.31075
-14 . 2656
1. 85483
-0.45369
1
1512.49
Std. Err.
1.40399
2 . 6274
0.0159478
0.0620284
NA
494.187
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
0.558982
-19.4152
1.82357
-0.575263
543.903
6.06253
-9.11596
1.88609
-0.332116
2481.08
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
16
i—1
CD
i—1
CO
(jn
0.0661
0.0639
0.0674
1763
17
1—1
CJ1
CO
1. 61
0.185
0.175
-0.789
4 944
15
1 . 6
1 . 51
0.265
0.28
1. 22
1.271e+004 12 1.5 1.45 0.221 0.371 0.51
3.7 04e + 004 19 1.35 1.42 0.515 0.432 -0.716
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-103 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 56.758717 6 -101.517434
A2 85.856450 10 -151.712901
A3 84.934314 7 -155.868628
fitted 81.120729 5 -152.241459
R 45.373551 2 -86.747101
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 80.9658 8 <.0001
Test 2 58.1955 4 <.0001
Test 3 1.84427 3 0.6053
Test 4 7.62717 2 0.02207
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 247.728
BMDL = 132.818
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-104 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.14.4. Figure for additional model presented: Hill, unrestricted
5000 10000 15000 20000 25000 30000 35000
10:26 01/12 2010
Kattainen et al., 2001: 3rd molar length in pups
E.2.14.5. Output for additional model presented: Hill, unrestricted
Kattainen et al., 2001: 3rd molar length in pups
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\l\Blood\25_Katt_2001_3molar_length_Hill_Unrest_l.(d)
Gnuplot Plotting File: C:\l\Blood\25_Katt_2 001_3molar_length_Hill_Unrest_l.plt
Tue Jan 12 10:26:49 2010
Figure 3 female only
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = -2.37155
rho = 0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-105 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
intercept
1. 85591
-0.507874
0. 845971
1606.5
Asymptotic
Correlation Matrix of Parameter
Estimates
lalph
a rho
intercept
V
n
k
lalpha NA
NA
NA
NA
NA
NA
rho NA
NA
NA
NA
NA
NA
;ercept NA
NA
1
NA
0.00038
0.00013
v NA
NA
NA
NA
NA
NA
n NA
NA
0. 00038
NA
1
-1.1
k NA
NA
0. 00013
NA
-1
. 1
1
Parameter Estimates
95.0% Wald Confidence
Interval
Variable
Estimate
Std. Err.
Lower Conf.
Limit
Upper
Conf.
Limit
lalpha
7.01946
NA
NA
NA
rho
-20.2971
NA
NA
NA
intercept
1.57098
NA
NA
NA
V
4 . 02956
NA
NA
NA
n
13.2039
NA
NA
NA
k
240356
NA
NA
NA
.east some variance
estimates are
negative.
3 USUALLY MEANS THE
MODEL HAS NOT
CONVERGED!
again from another
starting point
Table of Data and Estimated Values of Interest
Dose
N Obs
Mean
Est Mean
Obs Std Dev
Est Std Dev
Scaled Res.
0 16
1.
86
1. 57
0.0661
0.342
3.34
1763 17
1.
58
1. 57
0.185
0.342
0.0747
4944 15
1
. 6
1. 57
0.265
0.342
0.284
1.271e+004
12
1. 5
1.
57
0.221
0.342
CO
O
3.7 04e + 004
19
1. 35
1.
57
0.515
0.342
-2 . 85
Model Descriptions
for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R:
Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-106 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 56.758717 6 -101.517434
A2 85.856450 10 -151.712901
A3 84.934314 7 -155.868628
fitted 45.373551 6 -78.747101
R 45.373551 2 -86.747101
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.'
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 80.9658 8 <.0001
Test 2 58.1955 4 <.0001
Test 3 1.84427 3 0.6053
Test 4 79.1215 1 <.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 200720
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-107 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.15. Keller et al. (2006): Missing Mandibular Molars in CBA J Mice
2 E.2.15.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
2
1 P"
Value3
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
gamma
1
0.105
52.510
1.844E+03
4.959E+02
logistic
2
0.334
49.984
1.692E+03
1.220E+03
log-logistic
1
0.105
52.524
2.210E+03
1.330E+03
log-probit,
unrestricted
1
0.105
52.524
2.119E+03
1.336E+03
slope unrestricted
multistage, 1-
degree b
3
0.255
50.434
6.014E+02
4.203E+02
multistage, 2-
degree
1
0.122
51.394
1.057E+03
5.324E+02
multistage, 3-
degree
1
0.150
50.855
9.452E+02
5.285E+02
probit
2
0.342
49.905
1.614E+03
1.132E+03
Weibull
1
0.108
52.221
1.514E+03
5.160E+02
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model, BMDS output presented in this appendix
3
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-108 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.15.2. Figure for Selected Model: Multistage, 1-Degree
Multistage Model with 0.95 Confidence Level
2 0.4
BMDLBMD
10:27 01/12 2010
Multistage
10000
dose
Keller et al., 2007: Missing molars
E.2.15.3. Output File for Selected Model: Multistage, 1-Degree
Keller et al., 2007: Missing molars
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\l\Blood\26_Keller_2007_mand_molars_Multil_l.(d)
Gnuplot Plotting File: C:\l\Blood\26_Keller_2007_mand_molars_Multil_l.plt
Tue Jan 12 10:27:33 2010
Table 1 using mandibular molars only
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-109 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 5.517 35e + G15
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 0.000175192 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -21.5798 4
Fitted model -24.2169 1 5.27424 3 0.152^
Reduced model -71.326 1 99.4926 3 <.0001
AIC: 50.4338
Goodness of Fit
Scaled
Est. Prob. Expected Observed Size Residual
0.0000
296.0903
2364.8010
18764.0000
0.0000
0.0506
0.3392
0.9626
0. 000
1.163
37
79
9
0. 000
2 . 000
6. 000
30.000
29
23
29
30
0. 000
0 . 7 97
-1.505
1.079
Chi'" 2
4 . 0^
d.f.
P-value
0.2547
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 601.401
BMDL = 4 2 0.296
BMDU = 862.599
Taken together, (420.296, 862.599) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 E.2.16. Kociba et al. (1978): Urinary Coproporphyrins, Females (Table 2)
3 E.2.16.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.03
18.65
<0.0001
82.98
1.3E+04
7.4E+03
nonconstant variance,
power restricted >1
exponential (M3)
2
0.03
18.65
<0.0001
82.98
1.3E+04
7.4E+03
nonconstant variance,
power restricted >1
exponential (M4)
1
0.03
7.49
0.01
73.82
8.6E+02
4.0E+02
nonconstant variance,
power restricted >1
exponential (M5)
0
0.03
0.72
N/A
69.05
3.4E+03
8.7E+02
nonconstant variance,
power restricted >1
exponential (M5)
0
0.03
0.72
N/A
69.05
3.4E+03
8.7E+02
nonconstant variance,
power unrestricted
Hill
0
0.03
0.72
NA
69.05
3.0E+03
error
nonconstant variance, n
restricted >1
Hill
0
0.03
0.72
NA
69.05
3.0E+03
error
nonconstant variance, n
unrestricted
linear
2
0.03
17.90
0.00
82.23
9.9E+03
2.1E+03
nonconstant variance
polynomial
2
0.03
17.90
0.00
82.23
9.9E+03
2.1E+03
nonconstant variance
power
2
0.03
17.90
0.00
82.23
9.9E+03
2.1E+03
nonconstant variance,
power restricted >1,
bound hit
power
1
0.03
12.36
0.00
78.69
6.3E+02
5.6E-06
nonconstant variance,
power unrestricted
exponential (M2)
2
0.03
11.60
0.00
81.00
1.4E+04
9.5E+03
constant variance, power
restricted >1
exponential (M3)
2
0.03
11.60
0.00
81.00
1.4E+04
9.5E+03
constant variance, power
restricted >1
exponential (M4)c
1
0.03
4.05
0.04
75.44
1.4E+03
6.5E+02
constant variance, power
restricted >1
exponential (M5)
0
0.03
0.41
N/A
73.80
3.5E+03
8.8E+02
constant variance, power
restricted >1
exponential (M5)d
0
0.03
0.41
N/A
73.80
3.5E+03
8.8E+02
constant variance, power
unrestricted
Hill
0
0.03
0.41
NA
73.80
3.3E+03
error
constant variance, n
restricted >1
Hilld
0
0.03
0.41
NA
73.80
3.3E+03
error
constant variance, n
unrestricted
linear
2
0.03
11.02
0.00
80.41
1.1E+04
7.3E+03
constant variance
polynomial
2
0.03
11.02
0.00
80.41
1.1E+04
7.3E+03
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 11 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
power
2
0.03
11.02
0.00
80.41
1.1E+04
7.3E+03
constant variance, power
restricted >1, bound hit
power d
1
0.03
7.99
0.00
79.38
2.5E+03
1.9E+02
constant variance, power
unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meetBMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 12 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.17. Kociba et al. (1978): Uroporphyrin per Creatinine, Females
2 E.2.17.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.49
1.09
0.58
-93.46
7.6E+03
5.3E+03
nonconstant variance,
power restricted >1
exponential (M3)
2
0.49
1.09
0.58
-93.46
7.6E+03
5.3E+03
nonconstant variance,
power restricted >1
exponential (M4)
1
0.49
0.97
0.32
-91.57
5.7E+03
1.9E+03
nonconstant variance,
power restricted >1
exponential (M5)
0
0.49
0.51
N/A
-90.03
4.0E+03
2.0E+03
nonconstant variance,
power restricted >1
Hill
0
0.49
0.51
NA
-90.03
4.1E+03
2.0E+03
nonconstant variance, n
restricted >1
linear
2
0.49
0.98
0.61
-93.57
5.9E+03
3.7E+03
nonconstant variance
polynomial
2
0.49
0.98
0.61
-93.57
5.9E+03
3.7E+03
nonconstant variance
power
1
0.49
0.97
0.33
-91.58
6.3E+03
3.7E+03
nonconstant variance,
power restricted >1
exponential (M2)
2
0.49
0.56
0.75
-93.83
9.0E+03
6.9E+03
constant variance,
power restricted >1
exponential (M3)
2
0.49
0.56
0.75
-93.83
9.0E+03
6.9E+03
constant variance,
power restricted >1
exponential (M4)
1
0.49
0.46
0.50
-91.93
6.7E+03
2.2E+03
constant variance,
power restricted >1
exponential (M5)
0
0.49
0.20
N/A
-90.19
4.2E+03
2.3E+03
constant variance,
power restricted >1
linearc
2
0.49
0.46
0.79
-93.93
7.2E+03
5.1E+03
constant variance
polynomial
2
0.49
0.46
0.79
-93.93
7.2E+03
5.1E+03
constant variance
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.17.2. Figure for Selected Model: Linear, Constant Variance
Linear Model with 0.95 Confidence Level
Linear
0.4
0.35
0.3
0.25
0.2
0.15
BMDI
BMD
0
5000
10000
15000
20000
dose
13:40 11/16 2009
E.2.17.3. Output File for Selected Model: Linear, Constant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File:
C:\USEPA\BMDS21\AD\Blood\LinearConstVar_BMRl_Females_uroporphyrin_per_creatinine.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\LinearConstVar_BMRl_Females_uroporphyrin_per_creatinine.pit
Mon Nov 16 13:40:10 2009
Table 2
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose^2 + ...
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Signs of the polynomial coefficients are not restricted
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha = 0.0030385
rho = 0 Specified
beta_0 = 0.149139
beta 1 = 6.92935e-006
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha beta 0 beta 1
alpha 1 -5.8e-012 2.2e-011
beta_0 -5.8e-012 1 -0.6
beta 1 2. 2e-011 -0.6 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
beta_0
beta 1
Estimate
0. 00248773
0.149139
5. 92 935e-006
Std. Err.
0. 000786688
0.0139684
1. 29185e-006
Lower Conf. Limit
0.000945846
0.121762
4.39737e-006
Upper Conf. Limit
0. 00402961
0.176517
9.46132e-006
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 5
852.5 5
3942 5
2.125e+004
0.157
0.143
0.181
5 0.296
0.149
0.155
0.176
0. 05
0. 037
0. 053
0.296
0. 074
0. 0499
0. 0499
0. 0499
0. 0499
0.352
-0.54
0.204
-0.0161
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma'" 2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 50.195349 5 -90.390697
A2 51.400051 8 -86.800103
A3 50.195349 5 -90.390697
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
fitted
R
49.963861
41.049755
-93.927722
-78.099510
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio)
Test df
p-value
Test 1
Test 2
Test 3
Test 4
20.7006
2.40941
2 .40941
0. 462975
0.002076
0.4919
0.4919
0.7934
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 7197.95
BMDL = 5116.98
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 16 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.18. Latchoumycandane and Mathur (2002): Daily Sperm Production
2 E.2.18.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.85
19.80
<0.0001
94.90
1.2E+04
5.7E+03
nonconstant variance,
power restricted >1
exponential (M3)
2
0.85
19.80
<0.0001
94.90
1.2E+04
5.7E+03
nonconstant variance,
power restricted >1
exponential (M4)
1
0.85
0.16
0.69
77.26
1.0E+02
3.9E+01
nonconstant variance,
power restricted >1
exponential (M5)
1
0.85
0.16
0.69
77.26
1.0E+02
3.9E+01
nonconstant variance,
power restricted >1
Hill
1
0.85
0.00
0.95
77.10
6.3E+01
6.2E+00
nonconstant variance, n
restricted >1, bound hit
Hill
0
0.85
0.00
NA
79.10
5.1E+01
1.7E-05
nonconstant variance, n
unrestricted
linear
2
0.85
20.13
<.0001
95.23
1.3E+04
7.3E+03
nonconstant variance
polynomial
1
0.85
9.62
0.00
86.72
1.4E+03
7.9E+02
nonconstant variance
power
2
0.85
20.13
<.0001
95.23
1.3E+04
7.3E+03
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.85
20.71
<0.0001
93.82
9.6E+03
5.2E+03
constant variance, power
restricted >1
exponential (M3)
2
0.85
20.71
<0.0001
93.82
9.6E+03
5.2E+03
constant variance, power
restricted >1
exponential (M4)d
1
0.85
0.15
0.70
75.26
1.1E+02
4.4E+01
constant variance, power
restricted >1
exponential (M5)
0
0.85
0.15
N/A
77.26
1.6E+02
4.4E+01
constant variance, power
restricted >1
Hill, rextrictedc
1
0.85
0.00
0.98
118.11
3.40E+02
1.51E-02
constant variance, n
restricted >1, bound
hit
Hill, unrestricted d
0
0.85
0.00
NA
120.11
3.32E+02
8.77E-03
constant variance, n
unrestricted
linear
2
0.85
21.13
<.0001
94.24
1.1E+04
6.7E+03
constant variance
polynomial
1
0.85
11.01
0.00
86.13
1.1E+03
7.1E+02
constant variance
power
2
0.85
21.13
<.0001
94.24
1.1E+04
6.7E+03
constant variance, power
restricted >1, bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.18.2. Figure for Selected Model: Hill, Constant Variance, n Restricted >1, Bound Hit
Hill Model with 0.95 Confidence Level
Hill
25
20
5
VIDL BMD
0 2000 4000 6000 8000 10000 12000 14000
dose
10:15 11/27 2009
E.2.18.3. Output File for Selected Model: Hill, Constant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\Usepa\Bmds2\Data\HilTCDSet.(d)
Gnuplot Plotting File: C:\Usepa\Bmds2\Data\HilTCDSet.plt
Fri Nov 27 10:15:04 2009
BMDS Model Run
The form of the response function is:
Y[dose] = intercept + v*doseAn/(k^n + doseAn)
Dependent variable = m_sperm
Independent variable = DOSE
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 25 0
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
Default Initial Parameter Values
alpha = 43.3822
rho = 0 Specified
intercept = 22.19
v = -9.09
n = 1.93174
k = 304.417
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by
the user,
alpha
intercept
v
k
and do not appear in the correlation matrix )
alpha
1
6.6e-010
-7.3e-008
6.3e-008
intercept
6.6e-010
1
-0.75
-0.23
-7 . 3e-008
-0.75
1
-0.24
6. 3e-008
-0.23
-0.24
1
Interval
Variable
Limit
alpha
56.6072
intercept
27.0005
v
2.88049
n
k
767.569
Estimate
36.1524
22.1894
-9.16864
1
178.32
Parameter Estimates
Std. Err.
10.4363
2.45468
3.2083
NA
300.643
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
95.0% Wald Confidence
Lower Conf. Limit Upper Conf.
15.6976
17.3783
-15.4568
-410.929
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 6
436.7 6
2579 6
1.509e+004
22 .2
15.7
13.7
6 13.1
22.2
15.7
13. 6
6.54
6.49
5.36
6.01
6.01
6.01
13.1
7.74
6. 01
0. 000252
-0.00371
0.0148
-0.0113
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-l 19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = SigmaA2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/S2
Likelihoods of Interest
Model
A1
A2
A3
fitted
R
Log(likelihood)
-55.052739
-54.653533
-55.052739
-55.052919
-58.755106
# Param's
5
8
5
4
2
AIC
120.105478
125 .307067
120.105478
118.105839
121.510213
Explanation of Tests
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Test
1:
Test
2 :
Test
3:
Test
4 :
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test
Test
Test
Test
8.20315
0.798411
0.798411
0.000361116
0.2236
0.8498
0.8498
0.9848
The p-value for Test 1 is greater than .05. There may not be a
diffence between responses and/or variances among the dose levels
Modelling the data with a dose/response curve may not be appropriate
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-120 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 339.732
BMDL = 0.015111
E.2.18.4. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
20000 30000
dose
07:51 11/27 2009
E.2.18.5. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\Usepa\Bmds2\Data\LogTcdSet.(d)
Gnuplot Plotting File: C:\Usepa\Bmds2\Data\LogTcdSet.plt
Fri Nov 27 07:51:12 2009
BMDS Model Run
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = r_skin
Independent variable = DOSE
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-121 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 25 0
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0
intercept = -4.78342
slope = 0.469549
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -background
have been estimated at a boundary point, or have been specified by
the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.98
slope -0.98 1
Parameter Estimates
Interval
Variable
Limit
background
intercept
slope
Estimate
-4.84059
0.475472
Std. Err.
95.0% Wald Confidence
Lower Conf. Limit Upper Conf.
* - Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
Analysis of Deviance Table
#
Log(likelihood)
-71.5177
-71.5376
-95.8498
Param's
4
2
1
Deviance Test d.f.
0. 0398444
48 . 6642
P-value
0.9803
<.0001
AIC:
147.075
Goodness of Fit
Scaled
Dose Est._Prob. Expected Observed Size Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-122 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
0.0000 0.0000 0.000 0.000 38 0.000
316.0000 0.1087 4.784 5.000 44 0.105
4714.0000 0.3060 13.464 13.000 44 -0.152
50105.0000 0.5756 24.753 25.000 43 0.076
Chi^2 = 0.04 d.f. = 2 P-value = 0.9803
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 259.682
BMDL = 31.788
E.2.19. Li et al. (1997): Follicle-Stimulating Hormone
E.2.19.1. Summary Table of BMDS Modeling Results
Model3
Degrees
of
Freedom
2
1 P"
Value"
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
exponential (M2)
8
<0.0001
1095.433
2.898E+05
2.286E+05
exponential (M3)
8
<0.0001
1095.433
2.898E+05
2.286E+05
power bound hit
exponential (M4)
7
<0.0001
1059.480
1.891E+04
5.471E+03
exponential (M5)
6
<0.0001
1066.195
6.118E+04
4.729E+02
Hill
7
<.0001
1056.455
2.993E+03
1.081E+03
n lower bound hit
linear
8
<.0001
1077.819
1.109E+05
7.503E+04
polynomial
9
<.0001
1155.670
error
error
powerc
8
<.0001
1077.819
1.109E+05
7.503E+04
power bound hit
Hill, unrestricted
6
0.001
1039.476
1.206E+02
error
n unrestricted
power, unrestricted
d
7
0.002
1037.471
1.078E+02
1.353E+01
power unrestricted
a Non-constant variance model selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model, BMDS output presented in this appendix
d Alternate model, BMDS output also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-123 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.19.2. Figure for Selected Model: Power
Power Model with 0.95 Confidence Level
Power
700
600
500
400
300
200
100
0
BMDL BMD
0
100000
200000
300000
400000
500000
600000
dose
10:32 01/12 2010
Li et al., 1997: FSH
E.2.19.3. Output for Selected Model: Power
Li et al., 1997: FSH
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\l\Blood\72_Li_l997_FSH_Power_BMRl.(d)
Gnuplot Plotting File: C:\l\Blood\7 2_Li_19 97_FSH_Power_BMRl.plt
Tue Jan 12 10:32:04 2010
Figure 3: FSH in female S-D rats 24hr after dosing, 22 day old rats
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is restricted to be greater than or egual to 1
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 10
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 9.8191
rho = 0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-124 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
control =
slope =
power =
22 .1591
8.17907
0.293959
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
The model parameter(s) -power
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha
1
-0. 99
-0.29
-0.035
rho
-0. 99
1
0.2
0. 035
control
-0.29
0.2
1
-0.36
slope
-0.035
0. 035
-0.36
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
3. 49167
1. 27289
87.5089
0. 000889717
1
Std. Err.
1. 22596
0.242042
12.9454
0.000166742
NA
Lower Conf. Limit
1.08884
0.798492
62 .1364
0.000562908
Upper Conf. Limit
5. 89451
1.74728
112.881
0. 00121653
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
10
23. 9
87
5
29.6
98 . 7
-2 . 04
146.5
10
22 2
87
6
48.5
98 . 8
-2 .1
440.1
10
85.2
87
9
94 . 3
99
-0.0854
1156
10
73.3
88
5
48.5
99.4
-0.485
3232
10
126
90
4
159
101
1.12
8266
10
132
94
9
116
104
1.13
2.388e+004
6.60 8 e + 0 0 4
2.127e+005
6.4 97e + 005
10
10
10
10
117
304
347
455
109
146
277
6 6 6
51. 2
154
151
286
113
137
205
359
0.224
3. 65
1. 08
-1. 85
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-125 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Model R: Yi = Mu + e(i)
Var{e(i) ) = Sigma'"2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -535.687163 11 1093.374327
A2 -496.367061 20 1032.734122
A3 -502.709623 12 1029.419246
fitted -534.909723 4 1077.819445
R -574.835246 2 1153.670492
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2^log(Likelihood Ratio) Test df p-value
Test 1 156.936 18 <.0001
Test 2 78.6402 9 <.0001
Test 3 12.6851 8 0.1232
Test 4 64.4002 8 <.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 110907
BMDL = 75025.9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-126 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.19.4. Figure for Unrestricted Model: Power, Unrestricted
Power Model with 0.95 Confidence Level
Power
WDLBMD
100000 200000 300000 400000 500000 600000
dose
10:32 01/12 2010
Li et al„ 1997: FSH
E.2.19.5. Output for Unrestricted Model: Power, Unrestricted
Li et al., 1997: FSH
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\l\Blood\72_Li_l997_FSH_Power_Unrest_BMRl.(d)
Gnuplot Plotting File: C:\l\Blood\7 2_Li_19 97_FSH_Power_Unrest_BMRl.plt
Tue Jan 12 10:32:11 2010
Figure 3: FSH in female S-D rats 24hr after dosing, 22 day old rats
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 10
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 9.8191
rho = 0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-127 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
control
slope
power
22 .1591
8.17907
0.293959
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 99
-0.69
-0.17
0.26
rho
-0. 99
1
0. 65
0.13
-0.23
control
-0.69
0. 65
1
-0.12
0. 029
slope
-0.17
0.13
-0.12
1
-0. 97
power
0.26
-0.23
0. 029
-0. 97
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
3. 6735
1.17908
15.8235
7.68345
0.30464
Std. Err.
1.12114
0.221492
6.8753
2. 904 99
0. 0336473
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
1.4761
0.744961
2.34812
1. 98976
0.238692
5. 8709
1. 61319
29.2988
13.3771
0.370587
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
146.5
440.1
1156
3232
8266
2.388e+004
6.60 8 e + 0 0 4
2.127e+005
6.4 97e + 005
10
10
10
10
10
10
10
10
10
10
23. 9
22 . 2
85.2
73.3
126
132
117
304
347
455
15. 8
50. 9
64 . 9
81. 7
106
136
181
242
338
469
29.6
48.5
94 . 3
48.5
159
116
51. 2
154
151
286
32
63.7
73.5
84 .1
98 .1
114
135
160
194
236
0.795
-1.43
0. 875
-0.315
0. 652
-0.102
-1.
1.
52
24
0.139
-0.187
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma^2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-128 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Model Log(likelihood) # Param's AIC
A1 -535.687163 11 1093.374327
A2 -496.367061 20 1032.734122
A3 -502.709623 12 1029.419246
fitted -513.735602 5 1037.471204
R -574.835246 2 1153.670492
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 156.936 18 <.0001
Test 2 78.6402 9 <.0001
Test 3 12.6851 8 0.1232
Test 4 22.052 7 0.002489
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 107.761
BMDL = 13.5336
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-129 DRAFT—DO NOT CITE OR QUOTE
-------
2 E.2.20. Li et al. (2006): Hormone Levels (Estradiol)
3 E.2.20.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.44
4.95
0.08
271.02
7.7E+03
2.8E+03
nonconstant variance,
power restricted >1
exponential (M3)
2
0.44
4.95
0.08
271.02
7.7E+03
2.8E+03
nonconstant variance,
power restricted >1
exponential (M4)
1
0.44
0.34
0.56
268.41
error
error
nonconstant variance,
power restricted >1
exponential (M5)
0
0.44
0.34
N/A
270.41
error
error
nonconstant variance,
power restricted >1
exponential (M5)
0
0.44
0.34
N/A
270.41
error
error
nonconstant variance,
power unrestricted
Hill
1
0.44
0.34
0.56
268.41
error
error
nonconstant variance, n
restricted >1
linear
2
0.44
4.87
0.09
270.95
8.7E+03
2.7E+03
nonconstant variance
polynomial
2
0.44
4.87
0.09
270.95
8.7E+03
2.7E+03
nonconstant variance
power
2
0.44
4.87
0.09
270.95
8.7E+03
2.7E+03
nonconstant variance,
power restricted >1,
bound hit
power
2
0.44
0.34
0.84
266.41
2.8E+05
error
nonconstant variance,
power unrestricted
exponential (M2)
2
0.44
3.72
0.16
269.03
7.8E+03
3.1E+03
constant variance, power
restricted >1
exponential (M3)
2
0.44
3.72
0.16
269.03
7.8E+03
3.1E+03
constant variance, power
restricted >1
exponential (M4)
1
0.44
0.91
0.34
268.21
error
error
constant variance, power
restricted >1
exponential (M5)
0
0.44
0.91
N/A
270.21
error
error
constant variance, power
restricted >1
exponential (M5)d
0
0.44
0.91
N/A
270.21
error
error
constant variance, power
unrestricted
Hill
0
0.44
0.91
NA
270.21
error
error
constant variance, n
restricted >1
Hilld
0
0.44
0.96
NA
270.26
5.1E+15
5.1E+15
constant variance, n
unrestricted
linearc
2
0.44
3.65
0.16
268.95
8.8E+03
3.0E+03
constant variance
polynomial
2
0.44
3.65
0.16
268.95
8.8E+03
3.0E+03
constant variance
power
2
0.44
3.65
0.16
268.95
8.8E+03
3.0E+03
constant variance, power
restricted >1, bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-130 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
power d
1
0.44
0.96
0.33
268.27
5.2E+13
error
constant variance, power
unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
E.2.20.2. Figure for Selected Model: Linear, Constant Variance
Linear Model with 0.95 Confidence Level
Linear
35
30
25
20
15
10
5
0
BMDL
BMC i
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
dose
13:50 11/16 2009
E.2.20.3. Output File for Unrestricted Model: Linear, Constant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LinearConst_BMRl_Li_Estradiol_3d.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\LinearConst_BMRl_Li_Estradiol_3d.plt
Mon Nov 16 13:50:03 2009
Figure 3, 3-day estradiol
The form of the response function is:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-131 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Y[dose] = beta_0 + beta_l*dose + beta_2*dose^2 + ...
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Signs of the polynomial coefficients are not restricted
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 267.211
rho = 0 Specified
beta_0 = 16.1706
beta 1 = 0.00183421
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha beta 0 beta 1
alpha 1 2.7e-011 6.4e-013
beta_0 2.7e-011 1 -0.69
beta 1 6.4e-013 -0.69 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
beta_0
beta 1
Estimate
263.435
16.1706
0.00183421
Std. Err.
58.9058
3.55948
0.00220486
Lower Conf. Limit
147.981
9.19411
-0.00248724
Upper Conf. Limit
378.888
23.147
0.00615566
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10 10.2 16.2 12.2 16.2 -1.17
87.49 10 19.9 16.3 20 16.2 0.697
1564 10 24.7 19 14.6 16.2 1.11
2823 10 18.1 21.3 17.6 16.2 -0.635
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-132 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'"2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -129.653527 5 269.307054
A2 -128.294657 8 272.589314
A3 -129.653527 5 269.307054
fitted -131.476105 3 268.952210
R -131.819169 2 267.638338
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 7.04902 6 0.3163
Test 2 2.71774 3 0.4372
Test 3 2.71774 3 0.4372
Test 4 3.64516 2 0.1616
The p-value for Test 1 is greater than .05. There may not be a
diffence between responses and/or variances among the dose levels
Modelling the data with a dose/response curve may not be appropriate
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 884 8.86
BMDL = 2963.62
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-133 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.20.4. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5
Exponential
13:50 11/16 2009
E.2.20.5. Output File for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\ExpConst_Unrest_BMRl_Li_Estradiol_3d.(d)
Gnuplot Plotting File:
Mon Nov 16 13:50:07 2009
Figure 3, 3-day estradiol
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-134 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 5
lnalpha
rho(S)
5.48268
0
9.65979
0.000592388
2.68754
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 5
5.50531
0
10.1682
0. 0192802
2.10526
1.3399
NC
No Convergence
Table of Stats From Input Data
Dose
0
87 . 49
1564
2823
10
10
10
10
Obs Mean
10.17
19. 91
24.72
18.09
Obs Std Dev
12 .18
19. 97
14 . 55
17 . 6
Dose
0
87 . 49
1564
2823
Estimated Values of Interest
Est Mean Est Std Scaled Residual
10.17
19. 91
21.41
21.41
15. 68
15. 68
15. 68
15. 68
2 . 254e-007
-2.355e-007
0.669
-0.6 6 9
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-135 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma"2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -129.6535 5 269.3071
A2 -128.2947 8 272.5893
A3 -129.6535 5 269.3071
R -131.8192 2 267.6383
5 -130.1062 5 270.2123
Additive constant for all log-likelihoods = -36.76. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 7.049 6 0.3163
Test 2 2.718 3 0.4372
Test 3 2.718 3 0.4372
Test 7a 0.9053 0 N/A
The p-value for Test 1 is greater than .05. There may not be a
diffence between responses and/or variances among the dose levels
Modelling the data with a dose/response curve may not be appropriate.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
Degrees of freedom for Test 7a are less than or egual to 0.
The Chi-Sguare test for fit is not valid.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = Not_Computed
BMDL = 0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-136 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.20.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
BMDL
1e+015
2e+015
3e+015
4e+015
5e+015
dose
13:50 11/16 2009
E.2.20.7. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\HillConst_Unrest_BMRl_Li_Estradiol_3d.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Hi1IConst_Unrest_BMRl_Li_Estradiol_3d.pit
Mon Nov 16 13:50:08 2009
Figure 3, 3-day estradiol
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-137 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha = 267.211
rho = 0 Specified
intercept = 10.1682
v = 14.5566
n = 0.0272301
k = 109.605
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha intercept v n k
alpha 1 9.3e-007 NA 0.00038 NA
intercept 9.3e-007 1 NA 0.047 NA
v NA NA NA NA NA
0.00038
NA
0. 047
NA
NA
NA
NA - This parameter's variance has been estimated as zero or less.
THE MODEL HAS PROBABLY NOT CONVERGED!!!
Parameter Estimates
95.0% Wald Confidence Interval
Variable
Estimate
Std. Err.
Lower Conf. Limit
Upper Conf. Limi'
alpha
246.316
NA
NA
NA
intercept
10.168
NA
NA
NA
V
23.0562
NA
NA
NA
n
0. 030228
NA
NA
NA
k
68005.7
NA
NA
NA
At least some variance
THIS USUALLY MEANS THE
Try again from another
estimates are negative.
MODEL HAS NOT CONVERGED!
starting point.
Table of Data and Estimated Values of Interest
2 N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
3 7.49
1564
2823
10
10
10
10
10.2
19. 9
24 . 7
18 .1
10.2
20.5
21
21.1
12 . 2
20
14 . 6
17 . 6
15.7
15.7
15.7
15.7
4 . 22e-005
-0.127
0.743
-0.615
Degrees of freedom for Test A3 vs fitted <= 0
Model Descriptions for
Model A1: Yi j =
Var{e(ij)} =
likelihoods calculated
Mu(i) + e(ij)
Sigma'" 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-138 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)~2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -129.653527 5 269.307054
A2 -128.294657 8 272.589314
A3 -129.653527 5 269.307054
fitted -130.132269 5 270.264537
R -131.819169 2 267.638338
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
7.04902
2.71774
2.71774
0.957483
0.3163
0.4372
0.4372
NA
The p-value for Test 1 is greater than .05. There may not be a
diffence between responses and/or variances among the dose levels
Modelling the data with a dose/response curve may not be appropriate
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
NA - Degrees of freedom for Test 4 are less than or egual to 0. The Chi-Sguare
test for fit is not valid
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 5.11313e+015
BMDL = 5.11313e+015
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-139 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.20.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model
40
35
30
25
20
15
10
Power
13:50 11/16 2009
E.2.20.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\PowerConst_Unrest_BMRl_Li_Estradiol_3d.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\PowerConst_Unrest_BMRl_Li_Estradiol_3d.pit
Mon Nov 16 13:50:08 2009
Figure 3, 3-day estradiol
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-140 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha
rho
control
slope
power
267.211
0
10.1682
10.1311
0. 00388985
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
control
slope
power
alpha
1
3 . 9e-009
-6.4e-009
1. le-008
control
3.9e-009
1
-0.4
0. 038
slope
). 4e-009
-0.4
1
-0. 91
power
1. le-008
0. 038
-0. 91
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
246.319
10.1675
9.71449
0. 0151875
Std. Err.
55.0786
4.96274
12.3808
0.171197
Lower Conf. Limit
138.367
0. 440676
-14.5514
-0.320352
Upper Conf. Limit
354.271
19.8943
33.9803
0.350727
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
87 . 4 9 10
1564 10
2823 10
10.2
19. 9
24 . 7
18 .1
10.2
20.6
21
21.1
12 . 2
20
14 . 6
17 . 6
15.7
15.7
15.7
15.7
0.000148
-0.132
0.744
-0.612
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-141 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Model Log(likelihood) # Param's AIC
A1 -129.653527 5 269.307054
A2 -128.294657 8 272.589314
A3 -129.653527 5 269.307054
fitted -130.132565 4 268.265130
R -131.819169 2 267.638338
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 7.04902 6 0.3163
Test 2 2.71774 3 0.4372
Test 3 2.71774 3 0.4372
Test 4 0.958076 1 0.3277
The p-value for Test 1 is greater than .05. There may not be a
diffence between responses and/or variances among the dose levels
Modelling the data with a dose/response curve may not be appropriate
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 5.21395e+013
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-142 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.21. Li et al. (2006): Hormone Levels (Progesterone)
2 E.2.21.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
C
2
0.00
14.72
0.00
327.86
2.0E+03
7.1E+02
nonconstant variance,
power restricted >1
exponential (M3)
2
0.00
14.72
0.00
327.86
2.0E+03
7.1E+02
nonconstant variance,
power restricted >1
exponential (M4)
1
0.00
0.60
0.44
315.74
8.3E+00
1.4E-02
nonconstant variance,
power restricted >1
exponential (M5)
0
0.00
0.60
N/A
317.74
2.0E+01
3.5E-02
nonconstant variance,
power restricted >1
exponential (M5)d
0
0.00
0.60
N/A
317.74
2.0E+01
3.5E-02
nonconstant variance,
power unrestricted
Hill
1
0.00
0.60
0.44
315.73
9.0E-01
6.3E-03
nonconstant variance, n
restricted >1, bound hit
Hilld
0
0.00
0.62
NA
317.75
1.9E-01
error
nonconstant variance, n
unrestricted
linear
2
0.00
15.21
0.00
328.35
2.4E+03
1.3E+03
nonconstant variance
polynomial
2
0.00
15.21
0.00
328.35
2.4E+03
1.3E+03
nonconstant variance
power
2
0.00
15.21
0.00
328.35
2.4E+03
1.4E+03
nonconstant variance,
power restricted >1,
bound hit
power d
1
0.00
0.55
0.46
315.69
1.4E-39
1.4E-39
nonconstant variance,
power unrestricted
exponential (M2)
2
0.00
2.22
0.33
327.49
2.8E+03
1.1E+03
constant variance,
power restricted >1
exponential (M3)
2
0.00
2.22
0.33
327.49
2.8E+03
1.1E+03
constant variance,
power restricted >1
exponential (M4)
1
0.00
0.02
0.88
327.29
2.0E+02
8.3E-01
constant variance,
power restricted >1
exponential (M5)
1
0.00
0.02
0.88
327.29
2.0E+02
7.8E-01
constant variance,
power restricted >1
exponential (M5)
1
0.00
0.02
0.88
327.29
2.0E+02
7.8E-01
constant variance,
power unrestricted
Hill
0
0.00
0.02
NA
329.29
1.3E+02
1.6E-09
constant variance, n
restricted >1
Hill
0
0.00
0.00
NA
329.27
5.5E+02
1.0E-03
constant variance, n
unrestricted
linear
2
0.00
2.72
0.26
327.99
2.9E+03
1.7E+03
constant variance
polynomial
2
0.00
2.72
0.26
327.99
2.9E+03
1.7E+03
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-143 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
power
2
0.00
2.72
0.26
327.99
2.9E+03
1.7E+03
constant variance,
power restricted >1,
bound hit
power
1
0.00
0.02
0.90
327.28
8.1E+02
2.8E-12
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
1
2
3 E.2.21.2. Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
4 Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
80
Exponential
70
60
50
40
30
20
10
0
-10
BMDL
BMD
0
500
1000
1500
2000
2500
dose
5 13:48 11/16 2009
6
7
8 E.2.21.3. Output File for Selected Model: Exponential (M2'), Nonconstant Variance, Power
9 Restricted >1
10
11
12
13
14
15
16
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-144 DRAFT—DO NOT CITE OR QUOTE
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_BMRl_Li_Progesterone_3d.(d)
Gnuplot Plotting File:
Mon Nov 16 13:48:35 2009
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Figure 4, 3-day progesterone
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)'""d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
lnalpha 15.2703
rho -2.36741
a 68.5132
b 0.00136853
c 0.350182
d 1
Parameter Estimates
Variable Model 2
lnalpha 19.957 2
rho -3.64 854
a 65.2616
b 0.0274418
c 0.490738
d 1.59344
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 10 65.25 11.1
37.49 10 43.36 40.48
1564 10 27.46 33.3
2823 10 25.19 43.75
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-145 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dose
0
87 . 49
1564
2823
Estimated Values of Interest
Est Mean Est Std Scaled Residual
55.31
53.54
30. 93
19.36
28 . 9
29.21
34 . 87
40.57
1. 088
-1.102
-0.314
0.4542
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -159.6327 5 329.2653
A2 -151.8128 8 319.6255
A3 -152.5679 6 317.1358
R -163.9025 2 331.805
2 -159.928 4 327.856
Additive constant for all log-likelihoods = -36.76. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Test
4 :
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Test
Test 1
Test 2
Test 3
Test 4
Tests of Interest
-2*log(Likelihood Ratio)
24.18
15. 64
1. 51
14.72
p-value
0.000484
0.001344
0.4699
0.0006361
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-146 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
The p-value for Test 4 is less than .1. Model 2 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1988.62
BMDL = 712.505
E.2.21.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
80
70
60
50
40
30
20
10
"10 B i/lDL BMD
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
500
1000
1500
dose
2000
2500
13:48 11/16 2009
E.2.21.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_Unrest_BMRl_Li_Progesterone_3d.(d)
Gnuplot Plotting File:
Mon Nov 16 13:48:36 2009
Figure 4, 3-day progesterone
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-147 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'M}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha 15.2703
rho -2.36741
a 68.5132
b 0.00136853
c 0.350182
d 1
Parameter Estimates
Variable Model 5
lnalpha 19.957 2
rho -3.64 854
a 65.2616
b 0.0274418
c 0.490738
d 1.59344
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 10 65.25 11.1
3 7.49 10 43.36 40.48
1564 10 27.46 33.3
2823 10 25.19 43.75
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-148 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0 65.26 10.55 -0.003266
87.49 32.61 37.4 0.909
1564 32.03 38.65 -0.3733
2823 32.03 38.65 -0.5591
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + log(mean(i)) 'k rho)
Model R: Yij = Mu + e(1)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -159.6327 5 329.2653
A2 -151.8128 8 319.6255
A3 -152.5679 6 317.1358
R -163.9025 2 331.805
5 -152.8697 6 317.7393
Additive constant for all log-likelihoods = -36.76. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
24.18
15. 64
1. 51
0.6035
D. F.
p-value
0.000484
0.001344
0.4 6 9 9
N/A
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
Degrees of freedom for Test 7a are less than or egual to 0.
The Chi-Sguare test for fit is not valid.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-149 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 19.9163
BMDL = 0.034 8 9
E.2.21.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
13:48 11/16 2009
E.2.21.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_Li_Progesterone_3d.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_Li_Progesterone_3d.pit
Mon Nov 16 13:48:37 2009
Figure 4, 3-day progesterone
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-150 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i)
exp(lalpha + rho * ln(mean(i)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha
rho
intercept
7 . 08699
0
65.2507
-40.059
4 .4725
80.0627
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
-1
-0.17
0.84
6e-008
1. le-008
rho
-1
1
0.19
-0. 82
-5.6e-008
-le-008
intercept
-0.17
0.19
1
-0.43
le-008
1.9e-009
v
0.84
-0. 82
-0.43
1
1. 4e-009
2 . 6e-010
n
6e-008
-5.6e-008
le-008
1. 4e-009
1
1.1
k
1. le-008
-le-008
1. 9e-009
2.6e-010
1.1
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
19.8437
-3.62235
65.2507
-33.2448
5. 43075
0.22398
Std. Err.
5. 41703
1. 35086
3.33016
7 .73875
5 . 32553e + 006
1. 45115e + 006
Lower Conf. Limit
9.22649
-6.27
58 .7237
-48.4125
-1. 04378e + 007
-2 . 84421e + 006
Upper Conf. Limit
30.4609
-0.974711
71.7777
-18.0772
1.04378e+007
2 . 84421e + 006
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
87 . 49
1564
2823
10
10
10
10
65.3
43.4
27 . 5
25.2
65.3
32
32
32
11.1
40.5
33.3
43.7
10.5
38 . 3
38 . 3
38 . 3
-7.47e-007
0. 939
-0.375
-0.563
Degrees of freedom for Test A3 vs fitted <= 0
Model Descriptions for likelihoods calculated
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-151 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model A1: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma^2
Model A2 : Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) ) = Sigma'-2
Likelihoods of Interest
Model
A1
A2
A3
fitted
R
Log(likelihood)
-159.632675
-151.812765
-152.567898
-152.876553
-163.902499
Param1
5
AIC
329.265349
319.625529
317.135795
317.753105
331.804998
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
24.1795
15.6398
1.51027
0.61731
0.000484
0.001344
0.4 6 9 9
NA
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
NA - Degrees of freedom for Test 4 are less than or egual to 0. The Chi-Sguare
test for fit is not valid
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.19442
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-152 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
BMDL computation failed.
E.2.21.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
80
70
60
50
40
30
20
10
-10 ,
Power
i/IDLBMD
500
1000
1500
dose
2000
2500
13:48 11/16 2009
E.2.21.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Power_Unrest_BMRl_Li_Progesterone_3d.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Power_Unrest_BMRl_Li_Progesterone_3d.pit
Mon Nov 16 13:48:37 2009
Figure 4, 3-day progesterone
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i)
exp(lalpha + log(mean(i)
rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-153 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 7.08699
rho = 0
control = 65.2507
slope = -9.66956
power = 0.178886
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-1
-0.17
0.57
0.15
rho
-1
1
0.19
-0.55
-0.13
control
-0.17
0.19
1
-0.22
0. 02
slope
0.57
-0.55
-0.22
1
0.84
power
0.15
-0.13
0. 02
0.84
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
20.0647
-3.67315
65.2739
-30.3669
0. 0117985
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
5.5864 9.11557 31.0139
1.39112 -6.39969 -0.946614
3.34327 58.7212 71.8266
13.1525 -56.1453 -4.58852
0.0472043 -0.0807202 0.104317
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
87 . 4 9 10
1564 10
2823 10
65.3
43.4
27 . 5
25.2
65.3
33.3
32 . 2
31. 9
11.1
40.5
33.3
43.7
10.6
36.5
38 . 8
39.3
-0.00695
0. 876
-0.382
-0.541
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma'*-2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-154 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -159.632675 5 329.265349
A2 -151.812765 8 319.625529
A3 -152.567898 6 317.135795
fitted -152.844599 5 315.689197
R -163.902499 2 331.804998
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 24.1795 6 0.000484
Test 2 15.6398 3 0.001344
Test 3 1.51027 2 0.4699
Test 4 0.553402 1 0.4569
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1.42955e-039
BMDL = 1.42955e-039
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-155 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.22. Markowski et al. (2001): FRIO Run Opportunities
2 E.2.22.1. Summary Table of BMDS Modeling Results
Model3
Degrees
of
Freedom
2
1 P"
Valueb
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
exponential (M2)c
2
0.304
117.151
6.769E+03
2.281E+03
exponential (M3)
2
0.304
117.151
6.769E+03
2.281E+03
power bound hit
exponential (M4)
1
0.370
117.574
2.732E+03
1.151E+01
exponential (M5)
0
N/A
118.918
1.834E+03
9.541E-03
Hill
0
NA
118.918
1.428E+03
1.932E-04
linear
2
0.226
117.744
8.734E+03
4.535E+03
polynomial
2
0.226
117.744
8.734E+03
4.535E+03
power
2
0.226
117.744
8.734E+03
4.535E+03
power bound hit
a Non-constant variance model selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model, BMDS output presented in this appendix
3
4 E.2.22.2. Figure for Selected Model: Exponential (M2)
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
20
BMDL
BMD
0
1000
2000
3000
4000
5000
6000
7000
8000
dose
5 10:28 01/12 2010
6 Markowski et al., 2001: FR10 run opportunities
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-156 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.22.3. Output File for Selected Model: Exponential (M2)
Markowski et al., 2001: FRIO run opportunities
Exponential Model. (Version: 1.61; Date: 7/24/2009)
Input Data File: C:\l\Blood\33_Markowski_2001_FR10_run_opp_ExpCV_BMRl.(d)
Gnuplot Plotting File:
Tue Jan 12 10:28:14 2010
Table 3
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
lnalpha 3.5321
rho(S) 0
a 6.7793
b 7.3662 9e-005
c 0
d 1
Specified
Parameter Estimates
Variable Model 2
lnalpha 3.63129
rho 0
a 12.2912
b 0.00010238
c 0
d 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-157 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Stats From Input Data
Obs Mean
Obs Std Dev
0
1234
3184
8152
7
13.29
11. 25
5.75
7
8 . 65
5.56
3.53
6. 01
Dose
0
1234
3184
8152
Estimated Values of Interest
Est Mean Est Std Scaled Residual
12.29
10. 83
8 . 872
5.335
5.145
5.145
5.145
5.145
0.43
0.1359
-1.245
0.7168
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -54.38526 5 118.7705
A2 -51.88568 8 119.7714
A3 -54.38526 5 118.7705
R -57.45429 2 118.9086
2 -55.57543 3 117.1509
Additive constant for all log-likelihoods = -22.05. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Test
4 :
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Test
Test 1
Test 2
Test 3
Test 4
Tests of Interest
-2*log(Likelihood Ratio)
11.14
4.999
4.999
2 . 38
D. F.
p-value
0.08423
0.1719
0.1719
0.3042
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-158 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
The p-value for Test 1 is greater than .05. There may not be a
diffence between responses and/or variances among the dose levels
Modelling the data with a dose/response curve may not be appropriate.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is greater than .1. Model 2 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 6769.45
BMDL = 2280.85
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-159 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.23. Markowski et al. (2001): FR2 Revolutions
2 E.2.23.1. Summary Table of BMDS Modeling Results
Model3
Degrees
of
Freedom
2
1 P"
Value"
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
power, unrestricted
1
0.053
216.124
1.570E+04
3.350E+02
power unrestricted
exponential (M2)
2
0.236
217.220
6.704E+03
2.553E+03
exponential (M3)
2
0.236
217.220
6.704E+03
2.553E+03
power bound hit
exponential (M4)
1
0.262
217.588
2.702E+03
1.655E+01
exponential (M5)c
0
N/A
218.532
1.922E+03
7.384E+02
Hill
1
0.654
216.532
1.458E+03
4.757E+02
n lower bound hit
linear
2
0.180
217.765
8.361E+03
4.426E+03
polynomial
2
0.180
217.765
8.361E+03
4.426E+03
Hill, unrestricted
1
0.654
216.532
1.458E+03
error
n unrestricted
gower, unrestricted
1
0.161
218.297
4.538E+03
8.152E-12
power unrestricted
a Constant variance model selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model, BMDS output presented in this appendix
d Alternate model, BMDS output also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-160 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.23.2. Figure for Selected Model: Exponential (M5)
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
T
BMD
1000 2000 3000 4000 5000 6000 7000 8000
dose
14:52 01/13 2010
Markowski et al„ 2001: FR2 revolutions
E.2.23.3. Output File for Selected Model: Exponential (M5)
Markowski et al., 2001: FR2 revolutions
Exponential Model. (Version: 1.61; Date: 7/24/2009)
Input Data File: C:\l\Blood\34_Markowski_2001_FR2_rev_ExpCV_l.(d)
Gnuplot Plotting File:
Wed Jan 13 14:52:52 2010
Table 3
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *In(Y[dose]))
rho is set to 0.
A constant variance model is fit.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-161 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 5
lnalpha
rho(S)
7.68046
0
125.255
0.000305547
0.429602
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 5
7.68885
0
119.29
0.000585299
0.526177
4.76993
Table of Stats From Input Data
Dose
0
1234
3184
8152
Obs Mean
119.3
108 . 5
56.5
68 .14
Obs Std Dev
6 9.9
61
31. 21
33.23
Estimated Values of Interest
Dose
0
1234
3184
8152
Est Mean
119.3
108 . 5
62 .77
62 .77
Est Std
46.73
46.73
46.73
46.73
Scaled Residual
-1.267e-006
2.704e-006
-0.3285
0.3042
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-162 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -104.1655 5 218.331
A2 -101.1402 8 218.2803
A3 -104.1655 5 218.331
R -107.5993 2 219.1985
5 -104.2662 5 218.5323
Additive constant for all log-likelihoods = -22.05. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 12.92 6 0.04435
Test 2 6.051 3 0.1092
Test 3 6.051 3 0.1092
Test 7a 0.2013 0 N/A
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
Degrees of freedom for Test 7a are less than or egual to 0.
The Chi-Sguare test for fit is not valid.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1921.95
BMDL = 7 38.412
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-163 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.23.4. Figure for Unrestricted Model: Power, Unrestricted
Power Model with 0.95 Confidence Level
Power
200
150
100
50
BMD
0
1000
2000
3000
4000
5000
6000
7000
8000
dose
14:52 01/13 2010
Markowski et al., 2001: FR2 revolutions
E.2.23.5. Output for Unrestricted Model: Power, Unrestricted
Markowski et al., 2001: FR2 revolutions
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\l\Blood\34_Markowski_2001_FR2_rev_PowerCV_Unrest_l.(d)
Gnuplot Plotting File: C:\l\Blood\34_Markowski_2 001_FR2_rev_PowerCV_Unrest_l.plt
Wed Jan 13 14:52:55 2010
Table 3
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 2598.74
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-164 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho = 0 Specified
control = 119.29
slope = -0.0418736
power = 0.825655
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
control
slope
power
alpha
1
3 . 2e-009
-4 . 2e-009
-2 . 8e-009
control
3 . 2e-009
1
-0.39
-0.28
slope
-4 . 2e-009
-0.39
1
0. 99
power
-2 . 8e-009
-0.28
0. 99
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
2350.46
120.079
-3.33162
0.318007
Std. Err.
678.52
18.0799
10.4368
0.351246
Lower Conf. Limit
1020.59
84.6433
-23.7875
-0.370423
Upper Conf. Limit
3680.33
155.515
17 .1242
1.00644
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 7
1234 4
3184 6
8152 7
119
109
56.5
68 .1
120
88
76.8
61. 7
6 9.9
61
31. 2
33.2
48.5
48.5
48.5
48.5
-0.0431
0.844
-1. 02
0.353
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -104.165520 5 218.331040
A2 -101.140174 8 218.280349
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-165 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
A3 -104.165520 5 218.331040
fitted -105.148400 4 218.296799
R -107.599268 2 219.198536
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 12.9182 6 0.04435
Test 2 6.05069 3 0.1092
Test 3 6.05069 3 0.1092
Test 4 1.96576 1 0.1609
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 4538.4
BMDL = 8.1517 3e-012
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-166 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.24. Markowski et al. (2001): FR5 Run Opp
2 E.2.24.1. Summary Table of BMDS Modeling Results
Modela
Degrees
of
Freedom
2
1 P"
Valueb
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
exponential (M2)
2
0.205
133.194
4.012E+03
1.927E+03
exponential (M3)
2
0.205
133.194
4.012E+03
1.927E+03
power bound hit
exponential (M4)
1
0.253
133.335
1.710E+03
5.425E+02
exponential (M5)
1
0.212
133.587
1.757E+03
5.030E+02
power bound hit
Hillc
1
0.939
132.032
1.366E+03
7.212E+02
n lower bound hit
linear
2
0.122
134.230
5.715E+03
3.500E+03
polynomial
2
0.122
134.230
5.715E+03
3.500E+03
power
2
0.122
134.230
5.715E+03
3.500E+03
power bound hit
Hill, unrestricted
1
0.939
132.032
1.366E+03
6.598E+02
n unrestricted
gower, unrestricted
1
0.134
134.272
2.109E+03
8.152E-12
power unrestricted
a Constant variance model selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model, BMDS output presented in this appendix
d Alternate model, BMDS output also presented in this appendix
3
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-167 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.24.2. Figure for Selected Model: Hill
Hill Model with 0.95 Confidence Level
T
BMDL BMD
0 1000 2000 3000 4000 5000 6000 7000 8000
dose
10:29 01/12 2010
Markowski et al., 2001: FR5 run opportunities
E.2.24.3. Output File for Selected Model: Hill
Markowski et al., 2001: FR5 run opportunities
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\l\Blood\35_Markowski_2001_FR5_run_opp_HillCV_BMRl.(d)
Gnuplot Plotting File: C:\l\Blood\35_Markowski_2 001_FR5_run_opp_HillCV_BMRl.plt
Tue Jan 12 10:29:45 2010
Table 3
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 77.4849
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-168 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
intercept
0 Specified
26.14
-13.34
2 .78062
1968.39
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
-1. 9e-010
1. 7e-008
1. 8e-008
intercept
-1. 9e-010
1
-0. 81
-0.51
v
1. 7e-008
-0. 81
1
0.36
k
1. 8e-008
-0.51
0.36
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
intercept
k
Estimate
64 . 5863
26.14
-13.1569
18
1332.5
Std. Err.
18 . 6445
3. 03753
3.7676
NA
165.441
Lower Conf. Limit
28.0438
20.1865
-20.5413
1008.24
Upper Conf. Limit
101.129
32 .0935
-5.77257
1656.76
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
1234
3184
8152
26.1
23.5
12 . 8
13.1
26.1
23.5
13
13
12 . 3
7 . 04
6.17
7 .14
1. 04
S. 04
J. 04
5. 04
-3.13e-008
-1. 71e-008
-0.0558
0.0517
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-169 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -62.013133 5 134.026266
A2 -59.839035 8 135.678070
A3 -62.013133 5 134.026266
fitted -62.016025 4 132.032049
R -67.530040 2 139.060081
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 15.382 6 0.01748
Test 2 4.3482 3 0.2262
Test 3 4.3482 3 0.2262
Test 4 0.00578335 1 0.9394
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1366.2 9
BMDL = 7 21.238
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-170 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.24.4. Figure for Unrestricted Model: Power, Unrestricted
Power Model with 0.95 Confidence Level
40
Power
35
30
25
20
BMD
0
1000
2000
3000
4000
5000
6000
7000
8000
dose
10:29 01/12 2010
Markowski et al., 2001: FR5 run opportunities
E.2.24.5. Output File for Unrestricted Model: Power, Unrestricted
Markowski et al., 2001: FR5 run opportunities
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\l\Blood\35_Markowski_2001_FR5_run_opp_PowerCV_Unrest_BMRl.(d)
Gnuplot Plotting File:
C:\l\Blood\35_Markowski_2 001_FR5_run_opp_PowerCV_Unrest_BMRl.pit
Tue Jan 12 10:29:46 2010
Table 3
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-171 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
alpha = 77.4849
rho = 0 Specified
control = 2 6.14
slope = -0.00843066
power = 0.845567
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
control
slope
power
alpha
1
-2e-008
-le-008
-1. 3e-008
control
-2e-008
1
-0.43
-0.34
slope
-le-008
-0.43
1
0. 99
power
-1. 3e-008
-0.34
0. 99
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
70.905
26.3577
-0.41863
0.392134
Std. Err.
20.4685
3.12942
1.06088
0.282163
Lower Conf. Limit
30.7875
20.2242
-2.49792
-0.160895
Upper Conf. Limit
111.023
32.4913
1.66066
0. 945164
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
1234
3184
8152
26.1
23.5
12 . 8
13.1
26.4
19.5
16.5
12 .1
12 . 3
7 . 04
6.17
7 .14
3.42
S .42
J .42
] .42
-0.0684
0. 942
-1. 07
0.342
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
Likelihoods of Interest
Model
A1
Log(likelihood)
-62.013133
Param's
5
AIC
134.026266
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-172 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
A2 -59.839035 8 135.678070
A3 -62.013133 5 134.026266
fitted -63.136095 4 134.272189
R -67.530040 2 139.060081
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 15.382 6 0.01748
Test 2 4.3482 3 0.2262
Test 3 4.3482 3 0.2262
Test 4 2.24592 1 0.134
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 210 9.29
BMDL = 8.1517 5e-012
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-173 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.25. Mietinnin et al. (2006): Cariogenic Lesions in Pups
2 E.2.25.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
2
1 P"
Value3
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
gamma
3
0.410
162.281
2.689E+03
1.494E+03
power bound hit
logistic
3
0.371
162.518
3.248E+03
1.937E+03
log-logistic b
3
0.603
161.291
1.129E+03
4.091E+02
slope bound hit
log-probit,
unrestricted
2
0.732
161.972
5.141E+01
error
slope unrestricted
multistage, 4-
degree
3
0.410
162.281
2.689E+03
1.494E+03
final B=0
probit
3
0.350
162.656
3.596E+03
2.284E+03
Weibull
3
0.410
162.281
2.689E+03
1.494E+03
power bound hit
log-logistic,
unrestrictedc
2
0.728
161.983
3.912E+01
error
slope unrestricted
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model, BMDS output presented in this appendix
c Alternate model, BMDS output also presented in this appendix
3
4 E.2.25.2. Figure for Selected Model: Log-Logistic
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
0.7
0.5
0.4 ejmdiJJbmd
0
5000
10000
15000
20000
25000
30000
35000
dose
5 10:30 01/12 2010
6 Mietinnen et al., 2006: Cariogenic lesions in pups
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-174 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.25.3. Output File for Selected Model: Log-Logistic
Mietinnen et al., 2006: Cariogenic lesions in pups
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\l\Blood\36 Miet 06 carc lesions LogLogistic 1.(d)
Gnuplot Plotting File: C:\l\Blood\36 Miet 06 carc lesions LogLogistic l.plt
~Tue Jan_12 10:30:32 2010
Table 2 converting the percentage into the number of animals, and control is Control II from the
study. Dose is in ng per kg and is from Table 1
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0.595238
intercept = -9.1668
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background
intercept
background
1
-0. 66
intercept
-0. 66
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.644146 * * *
intercept -9.22611 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-175 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
Full model -77.6769 5
Fitted model -78.6457 2 1.93773 3 0.5854
Reduced model -83.2067 1 11.0597 4 0.0259
AIC: 161.291
Goodness of Fit
Scaled
Dose
Est
. Prob.
Expected
Observed
Size
Residual
0
. 0000
0.
6441
27.054
25.000
42
-0.662
1755
. 6399
0.
6 966
20.201
23.000
29
1.131
4922
. 4989
0.
7603
19.007
19.000
25
-0.003
12657
. 0000
0.
8416
20.197
20.000
24
-0.110
36874
. 0000
0.
9231
29.540
29.000
32
-0.359
Chi^
2 = 1.8
6
d.f. =
3 P
-value = 0.6025
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1128.77
BMDL = 409.065
E.2.25.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
Log-Logistic Model
Log-Logistic
5000 10000 15000 20000 25000 30000 35000
dose
10:30 01/12 2010
Mietinnen et al., 2006: Cariogenic lesions in pups
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-176 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.25.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Mietinnen et al., 2006: Cariogenic lesions in pups
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\l\Blood\36 Miet 06 carc lesions LogLogistic Unrest 1.(d)
Gnuplot Plotting File: C:\l\Blood\36 Miet 06 carc lesions LogLogistic Unrest l.plt
Tue Jan 12 10:30:33 2010
Table 2 converting the percentage into the number of animals, and control is Control II from the
study. Dose is in ng per kg and is from Table 1
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0.595238
intercept = -3.69546
slope = 0.44 2957
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.34 0.24
intercept -0.34 1 -0.99
slope 0.24 -0.99 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.597745 * * *
intercept -3.90353 * * *
slope 0.465358 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -77.6769 5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-177 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Fitted model -77.9913 3 0.62887 2 0.7302
Reduced model -83.2067 1 11.0597 4 0.0259
AIC: 161.983
Dose
Est
. Prob.
Goodness of Fit
Expected Observed
Size
Scale
Residu
0.0000
0.
5977
25.105
25.000
42
-0.033
1755.6399
0.
7566
21.941
23.000
29
0. 458
4922.4989
0.
8042
20.104
19.000
25
-0.557
12657.0000
0.
8474
20.338
20.000
24
-0.192
36874.0000
0.
8910
28.512
29.000
32
0 . 277
Chi ^2 = 0
63
d. f. =
2 P
-value = 0.7282
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 39.1207
Benchmark dose computation failed. Lower limit includes zero.
E.2.26. National Toxicology Program (1982): Male Mice, Toxic Hepatitis
E.2.26.1. Summary Table of BMDS Modeling Results
Model
Degrees of
Freedom
X1 Test
Statistic
2
X
p-Valuea
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model
Notes
gamma
1
4.92
0.03
113.10
2.1E+03
1.1E+03
power
restricted
>1
logistic b
2
A.11
0.09
110.35
1.7E+03
1.3E+03
log-logistic
1
4.93
0.03
113.09
2.1E+03
1.2E+03
slope
restricted
>1
log-probit
1
4.89
0.03
113.11
1.9E+03
1.3E+03
slope
restricted
>1
multistage 2-
degree
1
6.04
0.01
113.71
1.3E+03
7.0E+02
betas
restricted
>0
probit
2
4.99
0.08
110.51
1.5E+03
1.2E+03
Weibull
1
5.00
0.03
113.04
2.2E+03
9.3E+02
power
restricted
>1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-178 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
E.2.26.2. Figure for Selected Model: Multistage, 2nd Degree
~o 0.6
13:20 11/16 2009
Multistage Model with 0.95 Confidence Level
0 1000 2000 3000 4000 5000 6000
dose
E.2.26.3. Output File for Selected Model: Multistage, 2nd Degree
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\Blood\Multistage_BMR2_Toxic_hepatitis.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\Blood\Multistage_BMR2_Toxic_hepatitis.plt
Sun Nov 29 13:16:19 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2 ) ]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
BMDL
Multistage
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-179 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
Background = 0.0298369
Beta(1) = 0
Beta(2) = 5.57954e-QQ8
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(l) Beta(2)
Background
Beta (1)
Beta (2)
1
-0.8
0.74
-0.8
1
-0. 97
0.74
-0. 97
1
Variable
Background
Beta(1)
Beta(2)
Parameter Estimates
Estimate
0.0286224
1.97711e-005
5.00241e-008
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood) # Param'
-51.0633 4
-53.8523 3
-121.743 1
Deviance Test d.f.
5.57784
141.358
P-value
0.01819
<.0001
AIC:
113.705
Dose
Goodness of Fit
Est. Prob.
Expected
Observed
Scaled
Residual
0.0000
420.0366
1239.6134
6117.5662
0.0286
0.0451
0.1223
0.8676
2.089
2 . 211
5. 991
43.381
1.
5.
3.
44 .
000
000
000
000
73
49
49
50
-0.765
1. 920
-1.304
0.258
Chi'" 2
. 04
d.f.
P-value
0.0140
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1267.05
BMDL = 698.659
BMDU = 1628.68
Taken together, (698.659, 1628.68) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-180 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 E.2.27. National Toxicology Program (2006): Alveolar Metaplasia
3 E.2.27.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
13.37
0.01
320.09
2.7E+02
2.3E+02
power restricted >1, bound
hit
gamma
4
13.37
0.01
320.09
5.4E+02
4.6E+02
power restricted >1, bound
hit
logistic
4
33.08
0.00
343.28
6.8E+02
5.8E+02
logistic
4
33.08
0.00
343.28
1.3E+03
1.1E+03
log-logistic
3
1.32
0.72
312.56
1.8E+02
9.8E+01
slope restricted >1
log-logistic b
3
1.32
0.72
312.56
3.6E+02
2.1E+02
slope restricted >1
log-probit
3
1.44
0.70
312.68
2.2E+02
7.4E+01
slope restricted >1
log-probit
3
1.44
0.70
312.68
3.8E+02
1.5E+02
slope restricted >1
multistage, 2-
degree
4
13.37
0.01
320.09
2.7E+02
2.3E+02
betas restricted >0, bound
hit
multistage, 2-
degree
4
13.37
0.01
320.09
5.4E+02
4.6E+02
betas restricted >0, bound
hit
probit
4
35.22
0.00
347.07
7.2E+02
6.2E+02
probit
4
35.22
0.00
347.07
1.4E+03
1.2E+03
Weibull
4
13.37
0.01
320.09
2.7E+02
2.3E+02
power restricted >1, bound
hit
Weibull
4
13.37
0.01
320.09
5.4E+02
4.6E+02
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-181 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.27.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
0.6
0.4
0.2
KMDL BMD
2000
4000
6000
8000
dose
10000 12000 14000 16000
13:20 11/16 2009
E.2.27.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LogLogistic_BMR2_Alveolar_metaplasia.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\LogLogistic_BMR2_Alveolar_metaplasia.pit
Mon Nov 16 13:20:58 2009
0
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-182 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Default Initial Parameter Values
background = 0.0377358
intercept = -8.78161
slope = 1.1228
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.13 0.1
intercept -0.13 1 -1
slope 0.1 -1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0373474 * * *
intercept -8.85134 * * *
slope 1.13159 * * -k
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -152.615 6
Fitted model -153.279 3 1.32714 3 0.7227
Reduced model -216.802 1 128.374 5 <.0001
AIC: 312.558
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0
0373
1. 979
2 . 000
53
0
015
1408.4504
0
3682
19.881
19.000
54
-0
249
3137.0446
0
5807
30.777
33.000
53
0
619
5392.9593
0
7162
37 . 244
35.000
52
-0
690
9128.8027
0
8197
43.445
45.000
53
0
556
16361.0000
0
8976
46.674
46.000
52
-0
308
Chi/S2 = 1.32
d.f.
= 3 P
-value = 0.7232
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 357.926
BMDL = 206.635
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-183 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.28. National Toxicology Program (2006): Gingival Hyperplasia Squamous, 2 Years
2 E.2.28.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
X
p-Valuea
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
10.30
0.04
314.99
4.3E+03
2.8E+03
power restricted >1, bound
hit
logistic
4
12.16
0.02
318.60
7.7E+03
5.8E+03
log-logistic b
4
9.26
0.06
313.35
3.2E+03
2.1E+03
slope restricted >1,
bound hit
log-logisticc
3
1.62
0.66
307.51
3.9E+02
6.9E-03
slope unrestricted
log-probit
3
1.56
0.67
307.44
4.6E+02
2.6E-02
slope restricted >1
multistage, 1-
degree
4
10.30
0.04
314.99
4.3E+03
2.8E+03
betas restricted >0, bound
hit
probit
4
11.97
0.02
318.24
7.3E+03
5.5E+03
Weibull
4
10.30
0.04
314.99
4.3E+03
2.8E+03
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-184 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.28.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
0.4
0.3
0.2
0.1
0
11:36 11/29 2009
Log-Log istic
BMDL
10000
12000
14000
16000
E.2.28.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\Blood\LogLogistic_BMR2_Ging_Hyp_2yr.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\Blood\LogLogistic_BMR2_Ging_Hyp_2yr.plt
Sun Nov 29 11:36:25 2009
[insert study notes]
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-185 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Default Initial Parameter Values
background = 0.0188679
intercept = -10.0647
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.79
intercept -0.79 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0671889 * * *
intercept -10.2754 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -149.95 6
Fitted model -154.675 2 9.45083 4 0.05077
Reduced model -162.631 1 25.3627 5 0.0001186
AIC: 313.351
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0672
3.561
1.
000
53
-1. 405
1408.4504
0.1104
5. 961
7 .
000
54
0. 451
3137.0446
0.1582
8 . 386
14 .
000
53
2 .113
5392.9593
0.2134
11.311
13.
000
53
0.566
9128.8027
0.2905
15.395
15.
000
53
-0.119
16361.0000
0.4036
21.389
16.
000
53
-1.509
Chi'" 2 = 9.26
d.f.
= 4 P
-value
= 0.0550
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 3223.25
BMDL = 2054.88
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-186 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.28.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
0.4
0
B <1DL BMD
i
0
2000 4000 6000 8000 10000 12000 14000 16000
dose
11:36 11/29 2009
E.2.28.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\Blood\LogLogistic_Unrest_BMR2_Ging_Hyp_2yr.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\Nov2 9\Blood\LogLogistic_Unrest_BMR2_Ging_Hyp_2yr.pit
Sun Nov 29 11:36:27 2009
[insert study notes]
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-187 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Default Initial Parameter Values
background = 0.0188679
intercept = -4.87817
slope = 0.424322
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.16 0.11
intercept -0.16 1 -0.99
slope 0.11 -0.99 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0185138 * * *
intercept -4.42531 * * *
slope 0.373718 * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -149.95 6
Fitted model -150.753 3 1.60686 3 0.6578
Reduced model -162.631 1 25.3627 5 0.0001186
AIC: 307.507
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0185
0
981
1. 000
53
0. 019
1408.4504
0.1681
9
078
7 . 000
54
-0.756
3137.0446
0.2101
11
136
14.000
53
0.966
5392.9593
0.2433
12
893
13.000
53
0. 034
9128.8027
0.2792
14
795
15.000
53
0. 063
16361.0000
0.3230
17
117
16.000
53
-0.328
Chi/S2 = 1.62
d.f.
= 3
P
-value = 0.6555
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 388.363
BMDL = 0.00694785
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-188 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.29. National Toxicology Program (2006): Heart, Cardiomyopathy
2 E.2.29.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
2.65
0.62
394.49
1.5E+03
1.2E+03
power restricted >1, bound
hit
logistic
4
6.73
0.15
398.64
2.7E+03
2.2E+03
log-logistic
3
1.32
0.72
395.20
1.2E+03
7.3E+02
slope restricted >1
log-probit
3
1.11
0.78
394.98
1.3E+03
4.9E+02
slope restricted >1
multistage, 2-
degree b
4
2.65
0.62
394.49
1.5E+03
1.2E+03
betas restricted >0,
bound hit
probit
4
6.71
0.15
398.61
2.6E+03
2.2E+03
Weibull
4
2.65
0.62
394.49
1.5E+03
1.2E+03
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
3
4
5 E.2.29.2. Figure for Selected Model: Multistage, 2-Degree, Betas Restricted >0, Bound Hit
Multistage Model with 0.95 Confidence Level
Multistage
0.7
0.6
0.5
0.4
0.3
0.2
BMDL BMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
6 13:37 11/16 2009
7
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-189 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.29.3. Output File for Selected Model: Multistage, 2-Degree, Betas Restricted >0, Bound
Hit
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Multistage_BMR2_Cardiomyopathy.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Multistage_BMR2_Cardiomyopathy.plt
Mon Nov 16 13:37:37 2009
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/xl-beta2*dose/x2 ) ]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.234028
Beta(l) = 6.088Q3e-QQ5
Beta ( 2) = 0
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Beta(2)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Background Beta(1)
Background 1 -0.69
Beta(1) -0.69 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.196221 * * *
Beta(l) 6.98634e-005 * * *
Beta(2) 0 * * *
* - Indicates that this value is not calculated.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-190 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -193.93 6
Fitted model -195.247 2 2.63378 4 0.6209
Reduced model -216.802 1 45.7449 5 <.0001
AIC: 394.493
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.1962
10.400
10.000
53
-0.138
1408.4504
0.2715
14.663
12.000
54
-0.815
3137.0446
0.3544
18.784
22.000
53
0. 924
5392.9593
0.4485
23.325
25.000
52
0. 467
9128.8027
0.5752
30.487
32.000
53
0.420
16361.0000
0.7437
38.673
36.000
52
-0.849
Chi ^2 = 2.65
d.f.
= 4 P-
-value = 0.617 6
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1508.09
BMDL = 117 0.08
BMDU = 2325.84
Taken together, (1170.08, 2325.84) is a 90 % two-sided confidence
interval for the BMD
E.2.30. National Toxicology Program (2006): Hepatocyte Hypertrophy, 2 Years
E.2.30.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
5
12.03
0.03
273.88
2.4E+02
2.1E+02
power restricted >1, bound
hit
gamma
5
12.03
0.03
273.88
5.0E+02
4.3E+02
power restricted >1, bound
hit
logistic
4
26.14
0.00
297.90
7.4E+02
6.2E+02
logistic
4
26.14
0.00
297.90
1.4E+03
1.2E+03
log-logistic
4
14.32
0.01
279.21
3.7E+02
1.8E+02
slope restricted >1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-191 DRAFT—DO NOT CITE OR QUOTE
-------
log-logistic
4
14.32
0.01
279.21
6.3E+02
3.6E+02
slope restricted >1
log-probit
4
13.74
0.01
278.36
4.3E+02
2.3E+02
slope restricted >1
log-probit
4
13.74
0.01
278.36
6.6E+02
3.9E+02
slope restricted >1
multistage, 2-
degree
5
12.03
0.03
273.88
2.4E+02
2.1E+02
betas restricted >0, bound
hit
multistage, 2-
degree b
5
12.03
0.03
273.88
5.0E+02
4.3E+02
betas restricted >0,
bound hit
probit
4
28.00
0.00
299.73
7.2E+02
6.2E+02
probit
4
28.00
0.00
299.73
1.4E+03
1.2E+03
Weibull
5
12.03
0.03
273.88
2.4E+02
2.1E+02
power restricted >1, bound
hit
Weibull
5
12.03
0.03
273.88
5.0E+02
4.3E+02
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
E.2.30.2. Figure for Selected Model: Multistage, 2-Degree, Betas Restricted >0, Bound Hit
Multistage Model with 0.95 Confidence Level
Multistage
1
0.8
0.6
0.4
0.2
0
BMDLBMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
13:27 11/16 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-192 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.30.3. Output File for Selected Model: Multistage, 2-Degree, Betas Restricted >0, Bound
Hit
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File:
C:\USEPA\BMDS21\AD\Blood\Multistage_BMR2_Hepatocyte_hypertrophy_2years.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Multistage_BMR2_Hepatocyte_hypertrophy_2years.pit
Mon Nov 16 13:27:47 2009
[insert study notes]
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l-beta2',t:'dose^2 ) ]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.117028
Beta (1) = 0.000142077
Beta ( 2) = 5.42278e-009
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background -Beta(2)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta (1) 0.00021035 * * *
Beta(2) 0 * * *
- Indicates that this value is not calculated.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-193 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
Analysis of Deviance Table
Model
Log(likelihood) #
Param's Deviance Test d.f. P-va
Full model
-
129.986
6
Fitted model
-
135.938
1 11.9043 5 0
Reduced model
-219.97
1 179.968 5 <
AIC:
273.876
Goodness of Fit
Scaled
Dose Est
. Prob.
Expected
Observed Size Residual
0.0000 0.
0000
0. 000
0.000 53 0.000
1408.4504 0.
2564
13.846
19.000 54 1.606
3137.0446 0.
4831
25.604
19.000 53 -1.815
5392.9593 0.
6784
35.955
42.000 53 1.778
9128.8027 0.
8534
45.232
41.000 53 -1.643
16361.0000 0.
9680
51. 303
52.000 53 0.544
Chi ^2 = 12.03
d. f.
= 5 P-
value = 0.034 4
Benchmark Dose
Computation
Specified effect
=
0.1
Risk Type
=
Extra risk
Confidence level
=
0. 95
BMD
=
500.882
BMDL
=
433.488
BMDU
=
637.074
Taken together, (
433.488
, 637.074) is
a 90 % two-sided confidence
interval for the
BMD
E.2.31. National Toxicology Program (2006): Liver, Eosinophilic Focus, Multiple
E.2.31.1. Summary Table of BMDS Modeling Results
Model
Degrees of
Freedom
X Test
Statistic
XP-
Value3
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
gamma
3
3.72
0.29
331.90
2.0E+03
1.2E+03
power restricted >1
logistic
4
4.01
0.40
330.40
3.3E+03
2.8E+03
log-logistic
3
5.29
0.15
333.52
2.3E+03
1.1E+03
slope restricted >1
log-probit
3
5.90
0.12
334.15
2.3E+03
1.2E+03
slope restricted >1
multistage, 2-degree
3
2.69
0.44
330.82
2.0E+03
1.3E+03
betas restricted >0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-194 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
probitb
4
3.62
0.46
329.94
3.1E+03
2.7E+03
Weibull
3
3.47
0.32
331.63
2.1E+03
1.2E+03
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
E.2.31.2. Figure for Selected Model: Probit
Probit Model with 0.95 Confidence Level
Probit
.8
0.6
0.4
0.2
0
BMDL BMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
13:31 11/16 2009
E.2.31.3. Output File for Selected Model: Probit
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Probit_BMR2_liver_eosin_focus.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Probit_BMR2_liver_eosin_focus.plt
Mon Nov 16 13:31:29 2009
0
The form of the probability function is:
P[response] = CumNorm(Intercept+Slope*Dose) ,
where CumNorm(.) is the cumulative normal distribution function
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-195 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial (and Specified) Parameter Values
background = 0 Specified
intercept = -1.28017
slope = 0.000129308
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.77
slope -0.77 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
intercept -1.23453 0.125131 -1.47978 -0.98928
slope 0.000124995 1.49436e-005 9.57063e-005 0.000154284
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -161.07 6
Fitted model -162.972 2 3.80457 4 0.4331
Reduced model -202.816 1 83.4925 5 <.0001
AIC: 329.944
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0
1085
5.751
3
000
53
-1
215
1408.4504
0
1449
7 . 826
8
000
54
0
067
3137.0446
0
1998
10.588
14
000
53
1
172
5392.9593
0
2876
15.242
17
000
53
0
533
9128.8027
0
4628
24.526
22
000
53
-0
6 9 6
16361.0000
0
7912
41.932
42
000
53
0
023
Chi'A2 = 3.62 d.f. = 4 P-value = 0.4593
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-196 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Confidence level = 0.95
3
4 BMD = 3076.08
5
6 BMDL = 267 9.85
7
8
9
10 E.2.32. National Toxicology Program (2006): Liver, Fatty Change, Diffuse
11 E.2.32.1. Summary Table of BMDS Modeling Results
Model
Degrees of
Freedom
X Test
Statistic
XP-
Value3
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
gamma
4
2.42
0.66
252.35
2.2E+03
1.6E+03
power restricted >1
logistic
4
9.22
0.06
262.13
3.2E+03
2.8E+03
log-logistic
4
4.36
0.36
254.41
2.3E+03
1.8E+03
slope restricted >1
log-probit
4
4.30
0.37
254.43
2.3E+03
1.8E+03
slope restricted >1
multistage, 2-degree
4
2.03
0.73
252.07
2.0E+03
1.4E+03
betas restricted >0
probit
4
8.50
0.07
260.92
3.1E+03
2.6E+03
Weibull"
4
2.06
0.72
251.99
2.2E+03
1.6E+03
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
12
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-197 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.32.2. Figure for Selected Model: Weibull, Power Restricted >1
Weibull Model with 0.95 Confidence Level
0.6
0.4
0.2
Weibull
BMDU BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
13:31 11/16 2009
E.2.32.3. Output File for Selected Model: Weibull, Power Restricted >1
Weibull Model using Weibull Model (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Weibull_BMR2_liver_fatty_change_diff.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Weibull_BMR2_liver_fatty_change_diff.pit
Mon Nov 16 13:31:55 2009
NTP_liver_fatty_change_di ffuse
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(-slope*doseApower)]
Dependent variable = DichEff
Independent variable = Dose
Power parameter is restricted as power >=1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-198 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial (and Specified) Parameter Values
Background = 0.00925926
Slope = 1.61086e-007
Power = 1.69678
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Slope Power
Slope 1 -1
Power -1 1
Parameter Estimates
Variable
Background
Slope
Power
Estimate
0
1.01566e-006
1. 50443
Std. Err.
NA
1. 5567 2e-006
0.168998
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-2.03545e-006
1.1732
4.0667 8e-00£
1. 83566
NA - Indicates that this parameter has hit a bound
implied by some ineguality constraint and thus
has no standard error.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -122.992 6
Fitted model -123.994 2 2.00421 4 0.735
Reduced model -204.846 1 163.708 5 <.0001
AIC: 251.989
Dose
Est
. Prob.
Goodness of Fit
Expected Observed
Size
Scale
Residu
0.0000
0.
0000
0. 000
0. 000
53
0. 000
1408.4504
0.
0539
2 . 912
2 . 000
54
-0.550
3137.0446
0.
1688
8. 949
12.000
53
1.119
5392.9593
0.
3415
18 .102
17.000
53
-0.319
9128.8027
0.
6024
31.929
30.000
53
-0.542
16361.0000
0.
8913
47 . 238
48 . 000
53
0.336
Chi ^2 = 2.06
d.f. =
4 P
-value = 0.7243
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2158.24
BMDL = 1573.34
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-199 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3 E.2.33. National Toxicology Program (2006): Liver Necrosis
4 E.2.33.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
1 P"
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
0.80
0.94
234.40
4.8E+03
3.5E+03
power restricted >1, bound
hit
logistic
4
2.75
0.60
236.74
8.2E+03
6.8E+03
log-logistic
4
0.77
0.94
234.38
4.4E+03
3.1E+03
slope restricted >1, bound
hit
log-logistic
3
0.75
0.86
236.38
4.3E+03
1.9E+03
slope unrestricted
log-probitb
3
0.99
0.80
236.60
4.1E+03
1.9E+03
slope unrestricted
multistage, 2-
degree
4
0.80
0.94
234.40
4.8E+03
3.5E+03
betas restricted >0, bound
hit
probit
4
2.38
0.67
236.29
7.7E+03
6.4E+03
Weibull
4
0.80
0.94
234.40
4.8E+03
3.5E+03
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
5
6
7 E.2.33.2. Figure for Selected Model: Log-Probit
LogProbit Model with 0.95 Confidence Level
i'' 1111 ¦ '¦ i
LogProbit
0.5
0.4
0.3
0.2
BMDL
BMD
0
2000
4000
6000
8000
10000
12000
14000
16000
8 12:27 11/29 2009
9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-200 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.33.3. Output File for Selected Model: Log-Probit
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\Blood\LogProbit_BMR2_liver_necrosis.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\Blood\LogProbit_BMR2_liver_necrosis.plt
Sun Nov 29 12:27:13 2009
NTP liver necrosis
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope*Log(Dose) ) ,
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial (and Specified) Parameter Values
background = 0.0188679
intercept = -5.04893
slope = 0.457364
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.59 0.55
intercept -0.59 1 -1
slope 0.55 -1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0221159 0.0221444 -0.0212863 0.0655182
intercept -5.58721 1.71363 -8.94586 -2.22855
slope 0.517092 0.185108 0.154287 0.879898
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -114.813 6
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-201 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
Fitted model -115.299 3 0.972296 3 0.80E
Reduced model -127.98 1 26.3331 5 <.0001
AIC: 236.598
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0221
1.172
1.
000
53
-0.161
1408.4504
0.0544
2 . 938
4 .
000
54
0. 637
3137.0446
0.0976
5.174
4 .
000
53
-0.543
5392.9593
0.1457
7 .720
8 .
000
53
0.109
9128.8027
0.2096
11.106
10.
000
53
-0.373
16361.0000
0.3002
15.908
17 .
000
53
0.327
Chi ^2 = 0.99
d. f.
= 3 P
-value
= 0.8048
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 4132.6
BMDL = 1930.47
E.2.34. National Toxicology Program (2006): Liver, Pigmentation
E.2.34.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X Test
Statistic
X2P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
3
2.10
0.55
196.97
1.2E+03
8.2E+02
power restricted >1
logistic
4
5.42
0.25
197.07
1.0E+03
8.4E+02
log-logistic
3
0.16
0.98
195.53
1.4E+03
1.1E+03
slope restricted >1
log-probitb
3
0.29
0.96
195.53
1.4E+03
1.0E+03
slope restricted >1
multistage, 2-
degree
3
7.47
0.06
199.96
1.0E+03
5.5E+02
betas restricted >0
probit
4
15.44
0.00
200.50
9.4E+02
7.9E+02
Weibull
3
4.42
0.22
199.01
9.7E+02
6.6E+02
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-202 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.34.2. Figure for Selected Model: Log-Probit, Slope Restricted >1
LogProbit Model with 0.95 Confidence Level
0.6
0.4
0.2
LogProbit
BMDL BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
13:38 11/16 2009
E.2.34.3. Output File for Selected Model: Log-Probit, Slope Restricted >1
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LogProbit_BMR2_Pigmentation.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\LogProbit_BMR2_Pigmentation.plt
Mon Nov 16 13:38:02 2009
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope*Log(Dose)
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-203 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
User has chosen the log transformed model
Default Initial (and Specified) Parameter Values
background = 0.0754717
intercept = -12.1574
slope = 1.53218
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.35 0.33
intercept -0.35 1 -1
slope 0.33 -1 1
Parameter Estimates
Variable
background
intercept
slope
Estimate
0.0725493
-14 . 4 941
1. 83177
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
0.0338874 0.00613127 0.138967
2.03052 -18.4738 -10.5144
0.246866 1.34792 2.31562
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-94.6177
-94.7632
-210.717
Param's Deviance Test d.f.
0.290885
232.198
P-value
0.9617
C.0001
195.526
Est. Prob.
Goodness of Fit
Expected Observed Size
Scaled
Residual
0.0000
0
0725
3.845
4 .
000
53
0. 082
1408.4504
0
1769
9.552
9.
000
54
-0.197
3137.0446
0
6291
33.342
34 .
000
53
0.187
5392.9593
0
9013
47 .771
48 .
000
53
0.105
9128.8027
0
9874
52.334
52 .
000
53
-0.412
16361.0000
0
9 9 95
52.974
53.
000
53
0.160
Chi/S2 = 0.29
d.f.
= 3 P
-value
= 0.9624
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1356.93
BMDL = 1041.17
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-204 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.35. National Toxicology Program (2006): Liver, Toxic Hepatopathy
2 E.2.35.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
1.90
0.75
185.76
2.4E+03
1.9E+03
power restricted >1
logistic
4
6.59
0.16
191.14
2.7E+03
2.2E+03
log-logistic
3
3.01
0.39
189.58
2.6E+03
2.1E+03
slope restricted >1
log-probit
3
2.99
0.39
189.58
2.7E+03
2.1E+03
slope restricted >1
multistage, 2-
degree b
5
2.28
0.81
184.08
2.1E+03
1.7E+03
betas restricted >0,
bound hit
probit
4
5.60
0.23
189.82
2.5E+03
2.1E+03
Weibull
4
2.11
0.72
185.79
2.3E+03
1.8E+03
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
3
4
5 E.2.35.2. Figure for Selected Model: Multistage, 2-Degree, Betas Restricted >0, Bound Hit
Multistage Model with 0.95 Confidence Level
Multistage
1
0.8
0.6
0.4
0.2
0
BMDL BMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
6 11:44 11/19 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-205 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
E.2.35.3. Output File for Selected Model: Multistage, 2-Degree, Betas Restricted >0, Bound
Hit
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Multistage_BMR2_Toxic_hepatopathy.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Multistage_BMR2_Toxic_hepatopathy.plt
Thu Nov 19 11:44722 2009
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/Nl-beta2*dose/N2 ) ]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0
Beta(2) = 3.75131e+011
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background -Beta(l)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(2)
Beta(2) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 0 * * *
Beta (2) 2.3767e-008 * * *
* - Indicates that this value is not calculated.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-206 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
AIC:
Log(likelihood)
-89.8076
-91.0417
-218.207
184 . 083
Param's Deviance Test d.f.
2 .46809
256.799
P-value
0.7813
<.0001
Dose
Est. Prob.
Goodness of Fit
Expected Observed Size
Scaled
Residual
0.0000
0
0000
0. 000
0.
000
53
0. 000
1408.4504
0
0461
2.487
2 .
000
54
-0.316
3137.0446
0
2086
11.053
8 .
000
53
-1.032
5392.9593
0
4 990
26.449
30.
000
53
0. 975
9128.8027
0
8620
45.687
45.
000
53
-0.274
16361.0000
0
9983
52.909
53.
000
53
0.303
Chi ^2 = 2.28
d.f.
= 5 P
-value
= 0.8087
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2105.48
BMDL = 1698.91
BMDU = 2318.05
Taken together, (1698.91, 2318.05) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-207 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.35.4. Figure for Unrestricted Model: Multistage, 2-Degree, Betas Unrestricted
0.6
0.4
0.2
Multistage Model with 0.95 Confidence Level
Multistage
BMDL BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
11:22 11/19 2009
E.2.35.5. Output File for Unrestricted Model: Multistage, 2-Degree, Betas Unrestricted
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Multistage_Unrest_BMR2_Toxic_hepatopathy.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Multistage_Unrest_BMR2_Toxic_hepatopathy.pit
Thu Nov 19 11:22:30 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2 ) ]
The parameter betas are not restricted
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
Maximum number of iterations = 250
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-208 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 1
Beta (1) = -6.1241e + G15
Beta ( 2) = 7.17596e + Gll
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(1) Beta(2)
Beta(1) 1 -0.92
Beta(2) -0.92 1
Parameter Estimates
Variable
Background
Beta (1)
Beta (2)
Estimate
0
-1.36642e-005
2.62877e-008
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-89.8076
-90.8336
-218.207
Param's Deviance Test d.f.
2.05202
256.799
P-value
0.7262
<.0001
AIC:
185.667
Goodness of Fit
Scaled
Dose
Est
. Prob.
Expected
Observed
Size
Residual
0.0000
0.
0000
0. 000
0
000
53
0
000
1408.4504
0.
0324
1.748
2
000
54
0
194
3137.0446
0.
1941
10.289
8
000
53
-0
795
5392.9593
0.
4 989
26.439
30
000
53
0
978
9128.8027
0.
8733
46.285
45
000
53
-0
531
16361.0000
0.
9989
52 942
53
000
53
0
241
Chi ^2
1. 97
d.f.
P-value
0.7420
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 227 8.69
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-209 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 BMDL = 1743.86
3
4 BMDU = 2713.68
5
6 Taken together, (1743.86, 2713.68) is a 90 % two-sided confidence
7 interval for the BMD
8
9
10
11 E.2.36. National Toxicology Program (2006): Lung, Alveolar to Bronchiolar Epithelial
12 Metaplasia (Alveolar Epithelium, Metaplasia, Bronchiolar)
13 E.2.36.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
13.37
0.01
320.09
5.4E+02
4.6E+02
power restricted >1, bound
hit
logistic
4
33.08
0.00
343.28
1.3E+03
1.1E+03
log-logistic b
3
1.32
0.72
312.56
3.6E+02
2.1E+02
slope restricted >1
log-probit
3
1.44
0.70
312.68
3.8E+02
1.5E+02
slope restricted >1
multistage, 2-
degree
4
13.37
0.01
320.09
5.4E+02
4.6E+02
betas restricted >0, bound
hit
probit
4
35.22
0.00
347.07
1.4E+03
1.2E+03
Weibull
4
13.37
0.01
320.09
5.4E+02
4.6E+02
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
14
15
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-210 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.36.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
0.6
0.4
0.2
KMDL BMD
2000
4000
6000
8000
dose
10000 12000 14000 16000
13:38 11/16 2009
E.2.36.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LogLogistic_BMR2_Alv_bronch_epith_metapl.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\LogLogistic_BMR2_Alv_bronch_epith_metapl.pit
Mon Nov 16 13:38:53 2009
0
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope <= 1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-211 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Default Initial Parameter Values
background = 0.0377358
intercept = -8.78161
slope = 1.1228
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.13 0.1
intercept -0.13 1 -1
slope 0.1 -1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0373474 * * *
intercept -8.85134 * * *
slope 1.13159 * * -k
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -152.615 6
Fitted model -153.279 3 1.32714 3 0.7227
Reduced model -216.802 1 128.374 5 <.0001
AIC: 312.558
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0
0373
1. 979
2 . 000
53
0
015
1408.4504
0
3682
19.881
19.000
54
-0
249
3137.0446
0
5807
30.777
33.000
53
0
619
5392.9593
0
7162
37 . 244
35.000
52
-0
690
9128.8027
0
8197
43.445
45.000
53
0
556
16361.0000
0
8976
46.674
46.000
52
-0
308
Chi/S2 = 1.32
d.f.
= 3 P
-value = 0.7232
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 357.926
BMDL = 206.635
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-212 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.37. National Toxicology Program (2006): Oval Cell Hyperplasia, 2 Years
2 E.2.37.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
3
6.94
0.07
199.47
3.7E+03
2.8E+03
power restricted >1
logistic
4
6.40
0.17
196.80
3.3E+03
2.8E+03
log-logistic
3
8.21
0.04
201.66
3.8E+03
3.1E+03
slope restricted >1
log-probit
3
7.00
0.07
200.12
3.9E+03
3.3E+03
slope restricted >1
multistage, 2-
degree
4
7.05
0.13
197.13
2.5E+03
2.0E+03
betas restricted >0
probitb
4
5.64
0.23
195.45
3.1E+03
2.6E+03
Weibull
3
6.85
0.08
198.38
3.2E+03
2.3E+03
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
3
4
5 E.2.37.2. Figure for Selected Model: Probit
Probit Model with 0.95 Confidence Level
Probit
1
0.8
0.6
0.4
0.2
0
BMDL BMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
6 13:21 11/16 2009
7
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-213 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.37.3. Output File for Selected Model: Probit
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Probit_BMR2_Oval_cell_hyperplasia.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Probit_BMR2_Oval_cell_hyperplasia.plt
Mon Nov 16 13:21:57 2009
The form of the probability function is:
P[response] = CumNorm(Intercept+Slope* Dose),
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial (and Specified) Parameter Values
background = 0 Specified
intercept = -2.29925
slope = 0.000307725
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.87
slope -0.87 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
intercept -2.18988 0.208022 -2.59759 -1.78216
slope 0.000313001 3.3114e-005 0.000248098 0.000377903
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -92.4898 6
Fitted model -95.7243 2 6.46898 4 0.1668
Reduced model -210.191 1 235.402 5 <.0001
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-214 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
AIC:
195.449
Goodness of Fit
Dose
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
0.0143
0.756
0.
000
53
-0
876
1408.4504
0.0401
2 .168
4 .
000
54
1
270
3137.0446
0.1135
6. 017
3.
000
53
-1
306
5392 9593
0.3079
16.317
20.
000
53
1
096
9128 8027
0.7478
39.631
38 .
000
53
-0
516
16361 0000
0.9983
52.911
53.
000
53
0
299
Chi ^2 = 5.
64 d.f. =
4 P
-value
= 0.2274
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 3125.6
BMDL = 2 64 0.9 9
E.2.38. National Toxicology Program (2006): Toxic Hepatopathy
E.2.38.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
1.90
0.75
185.76
2.4E+03
1.9E+03
power restricted >1
logistic
4
6.59
0.16
191.14
2.7E+03
2.2E+03
log-logistic
3
3.01
0.39
189.58
2.6E+03
2.1E+03
slope restricted >1
log-probit
3
2.99
0.39
189.58
2.7E+03
2.1E+03
slope restricted >1
multistage, 2-
degree b
5
2.28
0.81
184.08
2.1E+03
1.7E+03
betas restricted >0,
bound hit
probit
4
5.60
0.23
189.82
2.5E+03
2.1E+03
Weibull
4
2.11
0.72
185.79
2.3E+03
1.8E+03
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-215 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.38.2. Figure for Selected Model: Multistage, 2-Degree, Betas Restricted >0, Bound Hit
Multistage Model with 0.95 Confidence Level
0.6
0.4
0.2
Multistage
BMDU BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
11:44 11/19 2009
E.2.38.3. Output File for Selected Model: Multistage, 2-Degree, Betas Restricted >0, Bound
Hit
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Multistage_BMR2_Toxic_hepatopathy.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Multistage_BMR2_Toxic_hepatopathy.plt
Thu Nov 19 11:44:22 2009
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/s2 ) ]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-216 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0
Beta(2) = 3.75131e+011
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background -Beta(l)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(2)
Beta(2) 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Estimate
0
0
2 . 37 67e-008
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-89.8076
-91.0417
-218.207
Param's Deviance Test d.f.
P-value
2 .46809
256.799
0.7813
<.0001
AIC:
184 . 083
Dose
Est. Prob.
Goodness of Fit
Expected Observed Size
Scaled
Residual
0.0000
0
0000
0. 000
0.
000
53
0. 000
1408.4504
0
0461
2.487
2 .
000
54
-0.316
3137.0446
0
2086
11.053
8 .
000
53
-1.032
5392.9593
0
4 990
26.449
30.
000
53
0. 975
9128.8027
0
8620
45.687
45.
000
53
-0.274
16361.0000
0
9983
52.909
53.
000
53
0.303
Chi ^2 = 2.28
d.f.
= 5 P
-value
= 0.8087
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2105.48
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-217 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E'.MDL = 1698.91
E'.MDU = 2 318.05
Taken together, (1698.91, 2318.05)
interval for the EMD
is a 90 % two'-sided confidenoe
E.2.38.4. Figure for Unrestricted Model: Multistage, 2-Degree, Betas Unrestricted
0.6
0.4
0.2
Multistage Model with 0.95 Confidence Level
Multistage
BMDL BMD
0 2000 4000
6000 8000 10000 12000 14000 16000
dose
11:44 11/19 2009
E.2.38.5. Output File for Unrestricted Model: Multistage, 2-Degree, Betas Unrestricted
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File:
C:\USEPA\BMDS21\AD\Blood\Multistage_Unrest_BMR2_2nd_Toxic_hepatopathy.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Multistage_Unrest_BMR2_2nd_Toxic_hepatopathy.pit
Thu Nov 19 11:44:23 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2 ) ]
The parameter betas are not restricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-218 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 1
Beta(1) = -6.1241e+015
Beta(2) = 7.17596e+011
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(1) Beta(2)
Beta(1) 1 -0.92
Beta(2) -0.92 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Estimate
0
-1. 36642e-005
2 . 62877e-008
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-89.8076
-90.8336
-218.207
Param's Deviance Test d.f.
2.05202
256.799
P-value
0.7262
<.0001
AIC:
185.667
Goodness of Fit
Scaled
Dose
Est
. Prob.
Expected
Observed
Size
Residual
0.0000
0.
0000
0. 000
0
000
53
0
000
1408.4504
0.
0324
1.748
2
000
54
0
194
3137.0446
0.
1941
10.289
8
000
53
-0
795
5392.9593
0.
4 989
26.439
30
000
53
0
978
9128.8027
0.
8733
46.285
45
000
53
-0
531
16361.0000
0.
9989
52 942
53
000
53
0
241
Chi'" 2
1. 97
d.f.
P-value
0.7420
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-219 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 227 8.69
BMDL = 17 4 3.8 6
BMDU = 2713.68
Taken together, (1743.86, 2713.68) is a 90 % two-sided confidence
interval for the BMD
E.2.39. Ohsako et al. (2001): Anogenital Distance in Male Pups
E.2.39.1. Summary Table of BMDS Modeling Results
Model3
Degrees
of
Freedom
2
1 P"
Valueb
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
exponential (M2)
3
0.092
185.349
2.358E+04
1.529E+04
exponential (M3)
3
0.092
185.349
2.358E+04
1.529E+04
power bound hit
exponential (M4)
2
0.190
184.217
2.617E+03
8.029E+02
exponential (M5)
1
0.092
185.741
2.204E+03
8.487E+02
Hillc
2
0.261
183.587
3.628E+03
8.053E+02
n lower bound hit
linear
3
0.086
185.490
2.436E+04
1.638E+04
polynomial
3
0.086
185.490
2.436E+04
1.638E+04
power
3
0.086
185.490
2.436E+04
1.638E+04
power bound hit
Hill, unrestricted d
1
0.106
185.515
4.741E+03
4.517E+02
n unrestricted
a Constant variance model selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model, BMDS output presented in this appendix
d Alternate model, BMDS output also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-220 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.39.2. Figure for Selected Model: Hill
Hill Model with 0.95 Confidence Level
32
Hill
30
28
26
24
22
BMI
BMD
0
5000
10000
15000
20000
25000
30000
dose
10:31 01/12 2010
Ohsako et al., 2001: Ano-genital distance in male pups
E.2.39.3. Output File for Selected Model: Hill
Ohsako et al., 2001: Ano-genital distance in male pups
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\l\Blood\56_Ohsako_2001_anogenital_HillCV_l.(d)
Gnuplot Plotting File: C:\l\Blood\56_Ohsako_2 001_anogenital_HillCV_l.plt
Tue Jan 12 10:31:18 2010
Figure 7
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 9.96434
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-221 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
intercept
0 Specified
28.9146
-5.04512
1.64399
2013.5
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
4 . le-008
7 . 2e-008
-le-007
intercept
4 . le-008
1
-0.53
-0.53
v
7 . 2e-008
-0.53
1
-0.27
k
-le-007
-0.53
-0.27
1
Parameter Estimates
Variable
alpha
intercept
k
Estimate
9.50299
28.988
-5.03805
1
2301.52
Std. Err.
1. 82885
0. 868025
1. 23954
NA
2261.96
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
95.0% Wald Confidence Interval
Lower Conf. Limit
5.91851
27.2867
-7 .4675
-2131.83
Upper Conf. Limit
13.0875
30.6893
-2.6086
6734.88
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 12 28.9
845.3 10 28.1
2763 10 25.3
9022 10 26.1
3. 05e + 004 12 23.9
29
27 . 6
26.2
25
3.54
2 . 52
3.59
3.59
3. 08
3. 08
3. 08
3. 08
24 . 3
2 . 36
3. 08
-0.0824
0. 455
-0.953
1.12
-0.488
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R:
Yi = Mu + e(i)
Var{e(i)} = Sigma/N2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-222 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
Likelihoods of Interest
Model
Log(likelihood)
# Param's
AIC
A1
-86.449919
6
184.899838
A2
-84 . 654549
10
189.309098
A3
-86.449919
6
184.899838
fitted
-87.793369
4
183.586738
R
-95.473923
2
194.947846
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
21. 6387
3.59074
3.59074
2 . 6869
0.005631
0. 4642
0. 4642
0.2609
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 3628.44
BMDL = 805.33
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-223 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
E.2.39.4. Figure for Unrestricted Model: Hill, Unrestricted
Hill Model with 0.95 Confidence Level
32
Hill
30
BMD
0
10:31 01/12 2010
5000
10000
15000
dose
20000
25000
30000
Ohsako et al., 2001: Ano-genital distance in male pups
E.2.39.5. Output File for Unrestricted Model: Hill, Unrestricted
Ohsako et al., 2001: Ano-genital distance in male pups
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\l\Blood\56_Ohsako_2001_anogenital_HillCV_Unrest_l.(d)
Gnuplot Plotting File: C:\l\Blood\56_Ohsako_2 001_anogenital_HillCV_Unrest_l.plt
Figure 7
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Tue Jan 12 10:31:19 2010
Default Initial Parameter Values
alpha
9.96434
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-224 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
intercept
0 Specified
28.9146
-5.04512
1.64399
2013.5
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
1. 7e-008
7 . 5e-008
7.3e-008
-7 . 2e-008
intercept
1. 7e-008
1
-0.0053
-0.0089
-0.14
v
7.5e-008
-0.0053
1
0. 98
-0. 99
n
7 . 3e-008
-0.0089
0. 98
1
-0. 96
k
-7 . 2e-008
-0.14
-0. 99
-0. 96
1
Parameter Estimates
Variable
alpha
intercept
Estimate
9. 4 9042
28.9785
-6.77236
0. 615459
6361.67
Std. Err.
1.82643
0.871908
12.034
1.15558
43105.4
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
5. 91068
27 .2696
-30.3585
-1.64943
-78123.4
13.0702
30.6874
16.8138
2 .88035
90846.7
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 12 28.9
845.3 10 28.1
2763 10 25.3
9022 10 26.1
3. 05e + 004 12 23.9
29
27 . 5
26.4
25.2
3.54
2 . 52
3.59
3.59
3. 08
3. 08
3. 08
3. 08
24 .1
2 . 36
3. 08
-0.0718
0. 633
-1.16
0. 861
-0.231
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-225 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Likelihoods of Interest
Model
Log(likelihood)
# Param's
AIC
A1
-86.449919
6
184.899838
A2
-84 . 654549
10
189.309098
A3
-86.449919
6
184.899838
fitted
-87 .757640
5
185.515280
R
-95.473923
2
194.947846
Test
1
Test
2
Test
3
Test
4
(Note:
Explanation of Tests
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
21. 6387
3.59074
3.59074
2.61544
0.005631
0. 4642
0. 4642
0.1058
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 4741.19
BMDL = 451.715
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-226 DRAFT—DO NOT CITE OR QUOTE
-------
2 E.2.40. Schantz et al. (1996): Maze Errors Per Block, Female
3 E.2.40.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linear
1
0.71
2.38
0.12
19.76
5.1E+03
2.9E+03
nonconstant variance
polynomial
1
0.71
2.38
0.12
19.76
5.1E+03
2.9E+03
nonconstant variance
power
1
0.71
2.38
0.12
19.76
5.1E+03
2.9E+03
nonconstant variance,
power restricted >1,
bound hit
power
0
0.71
0.00
NA
19.38
1.2E+03
5.4E-08
nonconstant variance,
power unrestricted
linearc
1
0.71
1.99
0.16
17.95
5.5E+03
3.6E+03
constant variance
polynomial
1
0.71
1.99
0.16
17.95
5.5E+03
3.6E+03
constant variance
power
1
0.71
1.99
0.16
17.95
5.5E+03
3.6E+03
constant variance,
power restricted >1,
bound hit
power d
0
0.71
0.00
NA
17.95
2.0E+03
8.1E-06
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-227 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.40.2. Figure for Selected Model: Linear, Constant Variance
Linear Model with 0.95 Confidence Level
Linear
4
3.5
3
2.5
2
BMDL
BMD
0 1000 2000 3000 4000 5000 6000 7000 8000
dose
13:42 11/16 2009
E.2.40.3. Output File for Selected Model: Linear, Constant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LinearConstVar_BMR4_maze_errors.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\LinearConstVar_BMR4_maze_errors.plt
Mon Nov 16 13:42:46 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta 2*doseA2 + ...
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Signs of the polynomial coefficients are not restricted
A constant variance model is fit
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum, number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-228 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
alpha = 0.569565
rho = 0 Specified
beta_0 = 3.37789
beta 1 = -0.000133906
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha beta 0 beta 1
alpha 1 1.5e-010 7.3e-012
beta_0 1.5e-010 1 -0.73
beta 1 7 . 3e-012 -0.73 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
beta_0
beta 1
Estimate
0.547839
3.37789
-0.000133906
Std. Err.
0.141451
0.196469
3. 88571e-005
Lower Conf. Limit
0.270599
2 . 99282
-0.000210064
Upper Conf. Limit
0. 825079
3.76296
-5.77472e-005
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
2670 10
8341 10
3.55
2.76
2 . 34
3.38
3. 02
2 .26
0. 639
0. 806
0. 806
0.74
0.74
0.74
0.755
-1.11
0.355
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -4.976366 4 17.952732
A2 -4.638353 6 21.276707
A3 -4.976366 4 17.952732
fitted -5.973388 3 17.946777
R -10.975997 2 25.951993
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-229 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
Explanation of Tests
Test 1:
Test 2
Test 3
Test 4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio)
Test df
p-value
Test 1
Test 2
Test 3
Test 4
12 . 6753
0. 676025
0. 676025
1.99405
0.01298
0.7132
0.7132
0.1579
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 5527.48
BMDL = 3627.8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-230 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
K.2.40,4. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
E.2.40.5. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\PwrConstVar_Unrest_BMR6_maze_errors.(d)
Gnuplot Plotting File:
G:\USEPA\BMDS21\AD\Blood\PwrConstVar_Unrest_BMR6_maze_errors.pit
Mon Nov 16 13:42:47 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-231 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha
rho
control
slope
power
0.569565
0
3.55459
-0.0428676
0.369985
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
control
slope
power
alpha
1
-6.8e-011
-1. 4e-012
-1. 6e-013
control
-6.8e-011
1
-0.35
-0.28
slope
-1. 4e-012
-0.35
1
1
power
-1. 6e-013
-0.28
1
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
0.512609
3.55459
-0.0428676
0.369985
Std. Err.
0.132355
0.226409
0.119074
0.311491
Lower Conf. Limit
0.253198
3.11084
-0.276249
-0.240526
Upper Conf. Limit
0.77202
3.99834
0.190514
0.980496
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
2670 10
8341 10
3.55
2.76
2 . 34
3.55
2.76
2 . 34
0. 639
0. 806
0. 806
0.716
0.716
0.716
2.62e-010
3.09e-010
3.32e-010
Degrees of freedom for Test A3 vs fitted <= 0
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-232 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Model Log(likelihood) # Param's AIC
A1 -4.976366 4 17.952732
A2 -4.638353 6 21.276707
A3 -4.976366 4 17.952732
fitted -4.976366 4 17.952732
R -10.975997 2 25.951993
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2^log(Likelihood Ratio) Test df p-value
Test 1 12.6753 4 0.01298
Test 2 0.676025 2 0.7132
Test 3 0.676025 2 0.7132
Test 4 1.77636e-015 0 NA
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
NA - Degrees of freedom for Test 4 are less than or equal to 0. The Chi-Square
test for fit is not valid
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 2017 . 9
BMDL = 8.10578e-006
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-233 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.41. Shi et al. (2007): Estradiol
2 E.2.41.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
0.05
11.41
0.01
391.64
3.8E+03
2.1E+03
nonconstant variance,
power restricted >1
exponential (M3)
3
0.05
11.41
0.01
391.64
3.8E+03
2.1E+03
nonconstant variance,
power restricted >1
exponential (M4)
C
2
0.05
0.74
0.69
382.97
4.4E+02
2.0E+02
nonconstant variance,
power restricted >1
exponential (M5)
2
0.05
0.74
0.69
382.97
4.4E+02
2.0E+02
nonconstant variance,
power restricted >1
exponential (M5)d
2
0.05
0.74
0.69
382.97
4.4E+02
2.0E+02
nonconstant variance,
power unrestricted
Hill
2
0.05
0.05
0.97
382.28
4.0E+02
error
nonconstant variance, n
restricted >1, bound hit
Hilld
1
0.05
0.02
0.90
384.24
3.9E+02
error
nonconstant variance, n
unrestricted
linear
3
0.05
14.08
0.00
394.31
5.4E+03
3.7E+03
nonconstant variance
polynomial
2
0.05
5.06
0.08
387.29
1.8E+03
1.2E+03
nonconstant variance
power
3
0.05
14.08
0.00
394.31
5.4E+03
3.7E+03
nonconstant variance,
power restricted >1,
bound hit
power d
2
0.05
1.36
0.51
383.59
3.5E+02
1.8E+01
nonconstant variance,
power unrestricted
exponential (M2)
3
0.05
9.37
0.02
392.09
2.8E+03
1.6E+03
constant variance,
power restricted >1
exponential (M3)
3
0.05
9.37
0.02
392.09
2.8E+03
1.6E+03
constant variance,
power restricted >1
exponential (M4)
2
0.05
0.61
0.74
385.34
3.3E+02
1.5E+02
constant variance,
power restricted >1
exponential (M5)
2
0.05
0.61
0.74
385.34
3.3E+02
1.5E+02
constant variance,
power restricted >1
exponential (M5)
2
0.05
0.61
0.74
385.34
3.3E+02
1.5E+02
constant variance,
power unrestricted
Hill
1
0.05
0.26
0.61
386.98
3.1E+02
1.2E+02
constant variance, n
restricted >1
Hill
1
0.05
0.26
0.61
386.98
3.1E+02
4.0E+01
constant variance, n
unrestricted
linear
3
0.05
12.21
0.01
394.93
4.4E+03
3.2E+03
constant variance
polynomial
2
0.05
5.39
0.07
390.12
1.4E+03
9.3E+02
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-234 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
power
3
0.05
12.21
0.01
394.93
4.4E+03
3.2E+03
constant variance,
power restricted >1,
bound hit
power
2
0.05
1.66
0.44
386.38
2.3E+02
1.2E+01
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
E.2.41.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
140
Exponential
120
100
BMDL BMD
0
1000
2000
3000
4000
5000
6000
7000
8000
dose
13:45 11/16 2009
E.2.41.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_BMRl_Shi_estradiol_17B_conc_PE9.(d)
Gnuplot Plotting File:
Mon Nov 16 13:45:19 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-235 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Figure 4 PE9 only
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign 'k b 'k dose}
exp{sign 'k (b 'k dose)Ad}
[c-(c-l) 'k exp{-b 'k dose}]
[c-(c-l) * exp{-(b * dose)Ad}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i) ) 'k rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha 2.65881
rho 0.913414
a 108
b 0.000503911
c 0.340136
d 1
Parameter Estimates
Variable Model 4
lnalpha 1.66777
rho 1.15313
a 103.145
b 0.00182735
c 0.418744
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 10 102.9 41.41
188.3 10 86.19 19.58
592.1 10 63.33 29.36
2882 10 48.1 18.82
7665 10 38.57 22.59
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-236 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Estimated Values of Interest
Dose
0
188 . 3
592 .1
2882
7665
Est Mean
103.1
85. 69
63.51
43.5
43.19
Est Std
33.35
29. 96
25.21
20.27
20.19
Scaled Residual
-0.02732
0.05287
-0.02235
0.7167
-0.1231
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -188.3615 6 388.7231
A2 -183.667 10 387.3339
A3 -186.1132 7 386.2263
R -203.3606 2 410.7211
4 -186.4844 5 382.9688
Additive constant for all log-likelihoods = -45.95. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
39.39
9.389
4 . 892
0.7425
D. F.
p-value
< 0.0001
0.05208
0.1798
0.6899
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-237 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 444.551
BMDL = 195.24 9
E.2.41.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
140
Exponential
120
100
80
60
40
20
BMDL BMD
0
1000
2000
3000
4000
5000
6000
7000
8000
dose
13:45 11/16 2009
E.2.41.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File:
C:\USEPA\BMDS21\AD\Blood\Exp_Unrest_BMRl_Shi_estradiol_17B_conc_PE9.(d)
Gnuplot Plotting File:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-238 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Mon Nov 16 13:45:21 2009
Figure 4 PE9 only
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha 2.65881
rho 0.913414
a 108
b 0.000503911
c 0.340136
d 1
Parameter Estimates
Variable Model 5
lnalpha 1.66777
rho 1.15313
a 103.145
b 0.00182735
c 0.418744
d 1
Table of
Stats From Input
Data
Dose
N
Obs Mean
Obs Std Dev
0
10
102 . 9
41.41
188 . 3
10
86.19
19.58
592 .1
10
63.33
29.36
2882
10
48.1
18 . 82
7665
10
38 . 57
22 .59
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-239 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Estimated Values of Interest
Dose
0
188 . 3
592 .1
2882
7665
Est Mean
103.1
85. 69
63.51
43.5
43.19
Est Std
33.35
29. 96
25.21
20.27
20.19
Scaled Residual
-0.02732
0.05287
-0.02235
0.7167
-0.7237
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -188.3615 6 388.7231
A2 -183.667 10 387.3339
A3 -186.1132 7 386.2263
R -203.3606 2 410.7211
5 -186.4844 5 382.9688
Additive constant for all log-likelihoods = -45.95. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
39.39
9.389
4 . 892
0.7425
D. F.
p-value
< 0.0001
0.05208
0.1798
0.6899
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-240 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 444.551
BMDL = 195.24 9
E.2.41.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
140
Hill
120
100
BMD
0
1000
2000
3000
4000
5000
6000
7000
8000
dose
13:45 11/16 2009
E.2.41.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File:
C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_Shi_estradiol_17B_conc_PE9.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_Shi_estradiol_17B_conc_PE9.pit
Mon Nov 16 13:45:22 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-241 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Figure 4 PE9 only
The form of the response function is:
Y[dose] = intercept + v^doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho ^ ln(mean(i)))
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 6.63982
rho = 0
intercept = 102.857
v = -64.2856
n = 1.33525
k = 461.707
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -1 0.064 -0.095 0.073 0.085
rho -1 1 -0.075 0.096 -0.074 -0.085
intercept 0.064 -0.075 1 -0.61 -0.22 -0.37
v -0.095 0.096 -0.61 1 0.83 -0.4
n 0.073 -0.074 -0.22 0.83 1 -0.52
k 0.085 -0.085 -0.37 -0.4 -0.52 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
1.54702
1.17907
105.265
-70.2058
0.875252
426.676
Std. Err.
2.36086
0.564621
10.3805
20.1009
0.64467
337.186
Lower Conf. Limit
-3.08017
0.0724289
84.9191
-109.603
-0.388278
-234.197
Upper Conf. Limit
6.17422
2.2857
125.61
-30.8086
2.13878
1087.55
Table of Data and Estimated Values of Interest
Dose
0
188 . 3
592 .1
N
Obs Mean
Est Mean
Obs Std Dev
Est Std Dev
Scaled Res
10
103
105
41. 4
33.7
-0.226
10
86.2
82 . 2
19.6
29.2
0. 431
10
63.3
65.2
29.4
25. 4
-0.227
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-242 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
2882 10 48.1 46.2 18.8 20.8 0.294
7665 10 38.6 40.2 22.6 19.1 -0.277
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma'"2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) ) = Sigma'"2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -188.361545 6 388.723090
A2 -183.666974 10 387.333947
A3 -186.113162 7 386.226325
fitted -186.121461 6 384.242922
R -203.360558 2 410.721116
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
39.3872
9.38914
4.89238
0.0165976
<.0001
0.05208
0.1798
0.8975
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-243 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Confidence level =
BMD =
0. 95
390.413
BMDL computation failed.
E.2.41.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
¦
140 - Power
120
100
80
60
40
20
¦
S/IDL BMD
1000 2000 3000 4000 5000 6000 7000 8000
dose
13:45 11/16 2009
E.2.41.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File:
C:\USEPA\BMDS21\AD\Blood\Power_Unrest_BMRl_Shi_estradiol_17B_conc_PE9.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Power_Unrest_BMRl_Shi_estradiol_17B_conc_PE9.pit
Mon Nov 16 13:45:22 2009
Figure 4 PE9 only
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-244 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i)
exp(lalpha + log(mean(i))
rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 6.63982
rho = 0
control = 102.857
slope = -2.986
power = 0.343163
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-1
0. 048
0.17
0.25
rho
-1
1
-0.059
-0.17
-0.25
control
0. 048
-0.059
1
-0.74
-0.59
slope
0.17
-0.17
-0.74
1
0. 98
power
0.25
-0.25
-0.59
0. 98
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
1.5482
1.1846
106.216
-9.40933
0.221631
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
2.39188 -3.13979 6.23619
0.571778 0.0639365 2.30526
10.4574 85.7201 126.712
6.9801 -23.0901 4.27142
0.0746081 0.0754014 0.36786
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10 103 106 41.4 34.4 -0.309
188.3 10 86.2 76.2 19.6 28.2 1.12
592.1 10 63.3 67.5 29.4 26.3 -0.5
2882 10 48.1 51.2 18.8 22.3 -0.443
7665 10 38.6 37.9 22.6 18.7 0.113
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)/S2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-245 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -188.361545 6 388.723090
A2 -183.666974 10 387.333947
A3 -186.113162 7 386.226325
fitted -186.795167 5 383.590334
R -203.360558 2 410.721116
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels:
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
39.3872
9.38914
4.89238
1.36401
<.0001
0.05208
0.1798
0.5056
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 346.016
BMDL = 18.2028
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-246 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.42. Smialowicz et al. (2008): PFC per 10A6 Cells
2 E.2.42.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
<0.0001
13.24
0.00
892.22
5.8E+03
3.9E+03
nonconstant variance,
power restricted >1
exponential (M3)
3
<0.0001
639.80
<0.0001
1518.75
6.4E+03
error
nonconstant variance,
power restricted >1
exponential (M4)
3
<0.0001
13.24
0.00
892.22
5.8E+03
3.9E+03
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
10.69
0.00
891.67
8.4E+03
5.1E+03
nonconstant variance,
power restricted >1
exponential (M5)d
2
<0.0001
10.69
0.00
891.67
8.4E+03
5.1E+03
nonconstant variance,
power unrestricted
Hill
2
<.0001
9.23
0.01
890.21
8.2E+03
error
nonconstant variance, n
restricted >1, bound hit
Hilld
1
<.0001
8.09
0.00
891.07
6.0E+03
error
nonconstant variance, n
unrestricted
linearc
3
<0001
9.68
0.02
888.66
9.7E+03
7.8E+03
nonconstant variance
polynomial
2
<.0001
9.28
0.01
890.26
8.4E+03
5.4E+03
nonconstant variance
power
3
<.0001
9.68
0.02
888.66
9.7E+03
7.8E+03
nonconstant variance,
power restricted >1,
bound hit
power d
2
<.0001
7.86
0.02
888.84
5.9E+03
1.6E+03
nonconstant variance,
power unrestricted
exponential (M2)
3
<0.0001
6.23
0.10
901.90
4.6E+03
2.8E+03
constant variance,
power restricted >1
exponential (M3)
3
<0.0001
6.23
0.10
901.90
4.6E+03
2.8E+03
constant variance,
power restricted >1
exponential (M4)
2
<0.0001
6.23
0.04
903.90
4.6E+03
8.1E+02
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
6.23
0.04
903.90
4.6E+03
8.1E+02
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
6.23
0.04
903.90
4.6E+03
8.1E+02
constant variance,
power unrestricted
Hill
2
<.0001
5.53
0.06
903.19
2.0E+03
3.8E+02
constant variance, n
restricted >1, bound hit
Hill
1
<.0001
1.55
0.21
901.22
1.1E+03
1.2E+02
constant variance, n
unrestricted
linear
3
<.0001
7.92
0.05
903.59
7.6E+03
5.8E+03
constant variance
polynomial
2
<.0001
6.55
0.04
904.22
5.3E+03
3.3E+03
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-247 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
power
3
<.0001
7.92
0.05
903.59
7.6E+03
5.8E+03
constant variance,
power restricted >1,
bound hit
power
2
<.0001
1.46
0.48
899.13
1.0E+03
1.2E+02
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
E.2.42.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
2000
1500
1000
500
Linear
BMDL
2000
4000
6000
8000 10000
dose
12000 14000 16000 18000
13:47 11/16 2009
E.2.42.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_PFC_per_cells.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_PFC_per_cells.plt
Mon Nov 16 13:47:58 2009
Anti Response to SRBCs, PFC per 10^6 cells, Table 4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-248 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose,A2 + ...
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 12.3562
rho = 0
beta_G = 1213.22
beta 1 = -0.0629452
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho beta 0 beta 1
lalpha 1 -1 0.081 -0.15
rho -1 1 -0.08 0.15
beta_0 0.081 -0.08 1 -0.9
beta 1 -0.15 0.15 -0.9 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
1.72142
1.55211
1192.68
-0.0597519
Std. Err.
1.91282
0.2835
79.6002
0.00532318
Lower Conf. Limit
-2.02764
0.9 9 6 4 6
1036.66
-0.0701851
Upper Conf. Limit
5. 47047
2 .10776
1348.69
-0.0493186
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
241. 3
1358
7385
1.744e+004 8 161 151 117 116 0.251
15
1. 4 9e + 003
1.19e+003
716
577
2
14
1.13e + 003
1.18e+003
171
572
-0.322
15
945
1.lle+003
516
547
i—1
i—1
CO
15
677
751
465
403
-0.715
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-249 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model
A1
A2
A3
fitted
R
Log(likelihood)
-444 . 832859
-425.402825
-435.489363
-440.330158
-463.753685
Param's
6
10
7
4
2
AIC
901.665718
870.805651
884 . 978727
888.660316
931.507371
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
76.7017
38 .8601
20.1731
9. 68159
<.0001
<.0001
0.0001563
0.02148
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 9660.4 8
BMDL = 7 7 55.63
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-250 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.42.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
2000
1500
1000
500
BMDL
BMD
0
2000
4000
6000
8000 10000 12000 14000 16000 18000
dose
13:48 11/16 2009
E.2.42.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_Unrest_BMRl_PFC_per_cells.(d)
Gnuplot Plotting File:
Mon Nov 16 13:48:00 2009
Anti Response to SRBCs, PFC per 10^6 cells, Table 4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-251 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha 3.29848
rho 1.2578
a 1565.55
b 0.000129358
c 0.000102839
d 1
Parameter Estimates
Variable
lnalpha
rho
Model 5
1.88041
1.53102
1200.9
9.15015e-005
0
1.53838
Table of Stats From Input Data
Dose
0
241. 3
1358
7385
15
14
15
15
Obs Mean
1491
1129
945
677
1. 744e + 004
161
Obs Std Dev
716
171
516
465
117
Dose
0
241. 3
1358
7385
1. 744e + 004
Estimated Values of Interest
Est Mean Est Std Scaled Residual
1201
1198
1153
694 . 8
154 . 3
583.1
581. 8
565.3
383.5
121. 2
1. 927
-0.4405
-1.427
-0.1801
0.1566
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-252 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Var{e(ij;
exp(lalpha + log(mean(i)
rho)
Model R: Yij
Var{e(ij)}
Mu + e(i)
Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -444.8329 6 901.6657
A2 -425.4028 10 870.8057
A3 -435.4894 7 884.9787
R -463.7537 2 931.5074
5 -440.8331 5 891.6662
Additive constant for all log-likelihoods = -61.57. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 76.7 8 < 0.0001
Test 2 38.86 4 < 0.0001
Test 3 20.17 3 0.0001563
Test 7a 10.69 2 0.004778
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 8 379.86
BMDL = 5143.92
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-253 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.42.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
2000
1500
1000
500
2000
4000
6000
8000 10000
dose
12000 14000 16000 18000
13:48 11/16 2009
E.2.42.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_PFC_per_cells.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_PFC_per_cells.plt
Mon Nov 16 13:48:01 2009
Anti Response to SRBCs, PFC per 10^6 cells, Table 4
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(
exp(lalpha + rho
In(mean(i)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 12.3562
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-254 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
intercept
0
1491
-1330
0. 478435
4033.79
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
-1
0.4
-0.047
-0.39
0.16
rho
-1
1
-0.41
0. 05
0.39
-0.16
intercept
0.4
-0.41
1
-0.15
-0.77
0.36
v
-0.047
0. 05
-0.15
1
0.37
-0. 95
n
-0.39
0.39
-0.77
0.37
1
-0. 65
k
0.16
-0.16
0.36
-0. 95
-0. 65
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
2 . 8564
1.38249
1309.76
-18748.9
0. 659862
1. 07 302e + 006
Std. Err.
2 . 65961
0.391339
139.641
37661.1
0.240194
. 27 003e + 006
Lower Conf. Limit
-2.35633
0.615484
1036.07
-92563.2
0.189091
-7 . 29608e + 006
Upper Conf. Limit
8 . 06913
2.1495
1583.45
55065.4
1.13063
9. 44213e + 006
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 15
241.3 14
1358 15
7385 15
1. 744e + 004
1.4 9e + 003 1.31e + 003 716
1.13e+003 1.24e+003 171
945 1.08e+003 516
677 633 465
8 161 149
596
572
522
360
1.18
-0.703
-1. 02
0.469
117
133
0.251
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-255 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Model Log(likelihood) # Param's AIC
A1 -444.832859 6 901.665718
A2 -425.402825 10 870.805651
A3 -435.489363 7 884.978727
fitted -439.536553 6 891.073107
R -463.753685 2 931.507371
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 76.7017 8 <.0001
Test 2 38.8601 4 <.0001
Test 3 20.1731 3 0.0001563
Test 4 8.09438 1 0.00444
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 604 9.7 4
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-256 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.42.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
2000
1500
1000
500
Power
BMDL
0 2000 4000 6000 8000 10000 12000 14000 16000 18000
dose
13:48 11/16 2009
E.2.42.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Power_Unrest_BMRl_PFC_per_cells.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Power_Unrest_BMRl_PFC_per_cells.plt
Mon Nov 16 13:48:05 2009
Anti Response to SRBCs, PFC per 10^6 cells, Table 4
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(
exp(lalpha + log(mean(i)
rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-257 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha = 12.3562
rho = 0
control = 1491
slope = -7 9.8343
power = 0.288026
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-1
0.39
-0.42
-0.4
rho
-1
1
-0.41
0.42
0.4
control
0.39
-0.41
1
-0. 81
-0.79
slope
-0.42
0.42
-0. 81
1
1
power
-0.4
0.4
-0.79
1
1
Parameter Estimates
95. 0%
Wald Confidence
Interval
Variable
Estimate
Std. Err.
Lower Conf. Limit Upper
Conf. Limit
lalpha
2.91272
2.64894
-2.2791
8.10454
rho
1.37364
0.389689
0.60986
2 .13741
control
1319.7
140.669
1043.99
1595.41
slope
-2.80443
6.05405
-
14.6701
9.06128
power
0.617853
0.211323
0
.203668
1.03204
Table of Data
and
Estimated Values
of Interest
Dose N Obs Mean Est Mean
Obs Std Dev
Est Std
Dev Scaled Res.
0 15 1.4 9e + 003
1.32e+003
716
597
1.11
241.3 14 1.13e+003
1.24e+003
171
571
-0.705
1358 15
945
1.08e+003
516
519
-0.992
7385 15
677
631
465
359
0. 494
1.744e+004 8
161 149
117
133 0.256
Model Descriptions
for
likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/N2
Model
Likelihoods of Interest
Log(likelihood) # Param's
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-258 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
A1 -444.832859 6 901.665718
A2 -425.402825 10 870.805651
A3 -435.489363 7 884.978727
fitted -439.417961 5 888.835922
R -463.753685 2 931.507371
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 76.7017 8 <.0001
Test 2 38.8601 4 <.0001
Test 3 20.1731 3 0.0001563
Test 4 7.85719 2 0.01967
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 5856.4
BMDL = 1632.55
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-259 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.43. Smialowicz et al. (2008): PFC per Spleen
2 E.2.43.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
0.00
5.76
0.12
377.56
7.4E+03
4.7E+03
nonconstant variance,
power restricted >1
exponential (M3)
2
0.00
5.34
0.07
379.14
8.5E+03
4.9E+03
nonconstant variance,
power restricted >1
exponential (M4)
3
0.00
5.76
0.12
377.56
7.4E+03
4.7E+03
nonconstant variance,
power restricted >1
exponential (M5)
1
0.00
5.34
0.02
381.14
8.5E+03
4.9E+03
nonconstant variance,
power restricted >1
exponential (M5)d
1
0.00
5.34
0.02
381.14
8.5E+03
4.9E+03
nonconstant variance,
power unrestricted
Hill
2
0.00
4.31
0.12
378.11
8.6E+03
error
nonconstant variance,
n restricted > 1, bound
hit
Hilld
1
0.00
2.66
0.10
378.46
6.6E+03
error
nonconstant variance,
n unrestricted
linearc
3
0.00
5.72
0.13
377.52
1.1E+04
8.9E+03
nonconstant
variance
polynomial
2
0.00
4.49
0.11
378.29
8.9E+03
5.7E+03
nonconstant variance
power
3
0.00
5.72
0.13
377.52
1.1E+04
8.9E+03
nonconstant variance,
power restricted >1,
bound hit
power d
2
0.00
2.62
0.27
376.42
6.5E+03
2.1E+03
nonconstant variance,
power unrestricted
exponential (M2)
3
0.00
4.38
0.22
391.51
5.8E+03
3.2E+03
constant variance,
power restricted >1
exponential (M3)
3
0.00
4.38
0.22
391.51
5.8E+03
3.2E+03
constant variance,
power restricted >1
exponential (M4)
2
0.00
4.38
0.11
393.51
5.8E+03
8.0E+02
constant variance,
power restricted >1
exponential (M5)
2
0.00
4.38
0.11
393.51
5.8E+03
8.0E+02
constant variance,
power restricted >1
exponential (M5)
2
0.00
4.38
0.11
393.51
5.8E+03
8.0E+02
constant variance,
power unrestricted
Hill
2
0.00
3.87
0.14
393.00
2.7E+03
4.0E+02
constant variance, n
restricted >1, bound
hit
Hill
1
0.00
1.06
0.30
392.19
1.8E+03
1.8E+02
constant variance, n
unrestricted
linear
3
0.00
5.59
0.13
392.72
9.0E+03
6.7E+03
constant variance
polynomial
2
0.00
4.61
0.10
393.74
6.5E+03
3.8E+03
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-260 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
power
3
0.00
5.59
0.13
392.72
9.0E+03
6.7E+03
constant variance,
power restricted >1,
bound hit
power
2
0.00
1.01
0.60
390.14
1.8E+03
1.8E+02
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
E.2.43.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
Linear
35
30
25
20
15
10
5
0
BMDI
BMD
0
2000
4000
6000
8000
10000 12000 14000 16000 18000
dose
13:45 11/16 2009
E.2.43.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_PFC_per_spleen.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Linear_BMRl_PFC_per_spleen.plt
Mon Nov 16 13:45:55 2009
Anti Response to SRBCs - PFC x 10 to the 4 per spleen, Table 4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-261 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose/s2 + ...
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.76607
rho = 0
beta 0 = 22.5956
beta~l = -0.00117245
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho beta 0 beta 1
lalpha 1 -0.97 0.031 -0.021
rho -0.97 1 -0.034 0.026
beta_0 0.031 -0.034 1 -0.88
beta 1 -0.021 0.026 -0.88 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
0. 4 91077
1.47094
22 .151
-0.00110204
Std. Err.
0.742891
0.264097
1.72621
0. 000118826
Lower Conf. Limit
-0.964962
0.953314
18.7677
-0.00133493
Upper Conf. Limit
1.94712
1.98856
25.5343
-0.000869145
Table of Data and Estimated Values of Interest
5 N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 15 27.8 22.2 13.4 12.5 1.75
241.3 14 21 21.9 13.6 12.4 -0.268
1358 15 17.6 20.7 9.4 11.9 -0.998
7385 15 12.6 14 8.7 8.91 -0.614
1.744e+004 8 3 2.93 3.1 2.82 0.0665
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-262 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -190.565019 6 393.130038
A2 -181.476284 10 382.952569
A3 -181.900030 7 377.800059
fitted -184.760998 4 377.521996
R -204.636496 2 413.272993
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
46.3204
18.1775
0.84749
5.72194
<.0001
0.001139
0.8381
0.126
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 11322.2
BMDL = 8 94 8.34
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-263 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.43.4.
Figure for UnrestrictedModel: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
35
30
25
20
15
10
5
0
BMDL
BMD
0
2000
4000
6000
8000
10000 12000 14000 16000 18000
dose
13:45 11/16 2009
E.2.43.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_Unrest_BMRl_PFC_per_spleen.(d)
Gnuplot Plotting File:
Mon Nov 16 13:45:56 2009
Anti Response to SRBCs - PFC x 10 to the 4 per spleen, Table 4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-264 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha 0.786146
rho 1.36372
a 29.19
b 0.000129431
c 0.000102775
d 1
Parameter Estimates
Variable
lnalpha
rho
Model 5
0.52811
1.45744
23.1604
9 . 96651e-005
6. 92509e-030
1. 23518
Table of Stats From Input Data
Dose
0
241. 3
1358
7385
15
14
15
15
Obs Mean
27 . 8
21
17 . 6
12 . 6
1. 744e + 004
Obs Std Dev
13. 4
13. 6
9.4
8 . 7
3.1
Dose
0
241. 3
1358
7385
1. 744e + 004
Estimated Values of Interest
Est Mean Est Std Scaled Residual
23.16
22 . 93
21.28
11. 68
3.2
12 .86
12 .77
12 .09
7 . 807
3. 04
1.397
-0.5656
-1.18
0.4578
-0.1864
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-265 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Var{e(ij;
exp(lalpha + log(mean(i)
rho)
Model R: Yij
Var{e(ij)}
Mu + e(i)
Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -190.565 6 393.13
A2 -181.4763 10 382.9526
A3 -181.9 7 377.8001
R -204.6365 2 413.273
5 -184.5689 6 381.1378
Additive constant for all log-likelihoods = -61.57. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 46.32 8 < 0.0001
Test 2 18.18 4 0.001139
Test 3 0.8475 3 0.8381
Test 7a 5.338 1 0.02087
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated
Confidence Level = 0.950000
BMD = 8 4 60
BMDL = 4 901
standard deviations from control
94
02
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-266 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.43.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
35
30
25
20
15
10
BMD
2000 4000 6000 8000 10000 12000 14000 16000 18000
dose
13:45 11/16 2009
E.2.43.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_PFC_per_spleen.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_PFC_per_spleen.plt
Mon Nov 16 13:45:57 2009
Anti Response to SRBCs - PFC x 10 to the 4 per spleen, Table 4
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(
exp(lalpha + rho
In(mean(i)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.76607
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-267 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
intercept
0
27 . 8
-24 . 8
0. 476652
4009.51
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
-0. 98
0.24
0. 03
-0.21
0. 019
rho
-0. 98
1
-0.3
-0.026
0.21
-0.021
intercept
0.24
-0.3
1
0. 079
-0.73
0.1
v
0. 03
-0.026
0. 079
1
0. 019
-0. 96
n
-0.21
0.21
-0.73
0. 019
1
-0.28
k
0. 019
-0.021
0.1
-0. 96
-0.28
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
0.742099
1.37015
25.3072
-1195.09
0.543247
2.57198e+007
Std. Err.
1.02085
0.355955
2.92734
4993.33
0.174917
2 .10767e + 008
Lower Conf. Limit
-1.25872
0.67249
19.5697
-10981.8
0.200417
-3.87 37 5e + 008
Upper Conf. Limit
2 . 74292
2 . 06781
31.0447
8591.65
0. 886078
4.38815e+008
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 15
241.3 14
1358 15
7385 15
1. 744e + 004
27 . 8
21
17 . 6
12 . 6
25.3
23.1
19.7
11. 2
13. 4
13. 6
9.4
8 . 7
13.3
12 . 5
11. 2
7 . 6
0.728
-0.629
-0.716
0. 691
3. 03
3.1
3.1
-0. 03
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-268 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Model Log(likelihood) # Param's AIC
A1 -190.565019 6 393.130038
A2 -181.476284 10 382.952569
A3 -181.900030 7 377.800059
fitted -183.230840 6 378.461681
R -204.636496 2 413.272993
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 46.3204 8 <.0001
Test 2 18.1775 4 0.001139
Test 3 0.84749 3 0.8381
Test 4 2.66162 1 0.1028
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 6615 . 87
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-269 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.43.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
E.2.43.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
BMDL BMD
0 2000 4000 6000 8000 10000 12000 14000 16000 18000
dose
13:45 11/16 2009
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Power_Unrest_BMRl_PFC_per_spleen.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Power_Unrest_BMRl_PFC_per_spleen.plt
Mon Nov 16 13:45:57 2009
Anti Response to SRBCs - PFC x 10 to the 4 per spleen, Table 4
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-270 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha = 4.76607
rho = 0
control = 27.8
slope = -1.51177
power = 0.286447
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 98
0.25
-0.24
-0.22
rho
-0. 98
1
-0.3
0.25
0.22
control
0.25
-0.3
1
-0.78
-0.74
slope
-0.24
0.25
-0.78
1
1
power
-0.22
0.22
-0.74
1
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
0.746924
1. 36826
25.3818
-0.124774
0.531205
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
1.02058 -1.25337 2.74721
0.355827 0.670849 2.06566
2.96695 19.5666 31.1969
0.226126 -0.567972 0.318425
0.175723 0.186794 0.875617
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 15
241.3 14
1358 15
7385 15
1. 744e + 004
27 . 8
21
17 . 6
12 . 6
25. 4
23.1
19.6
11. 2
13. 4
13. 6
9.4
8 . 7
13.3
12 . 4
11.1
7 . 6
3. 03
3.1
3.1
0.705
-0.626
-0.704
0.702
-0.0313
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/N2
Likelihoods of Interest
Model
Log(likelihood)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-271 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
A1 -190.565019 6 393.130038
A2 -181.476284 10 382.952569
A3 -181.900030 7 377.800059
fitted -183.210067 5 376.420134
R -204.636496 2 413.272993
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 46.3204 8 <.0001
Test 2 18.1775 4 0.001139
Test 3 0.84749 3 0.8381
Test 4 2.62008 2 0.2698
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 6542.48
BMDL = 2072.46
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-272 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.44. Toth et al. (1978): Amyloidosis
2 E.2.44.1. Summary Table of BMDS Modeling Results
Amyloidosis (Toth et al. (1978))
Model
Degrees
of
Freedom
X1 Test
Statistic
2
X
p-Valuea
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
2
6.45
0.04
149.12
1.1E+04
7.0E+03
power restricted >1, bound
hit
logistic
2
7.91
0.02
151.34
2.0E+04
1.6E+04
log-logisticb
2
5.86
0.05
148.27
8.3E+03
4.8E+03
slope restricted >1,
bound hit
log-logisticc
2
0.20
0.90
140.24
2.7E+02
2.9E+00
slope unrestricted
log-probit
2
9.94
0.007
153.52
2.2E+04
1.5E+4
slope restricted >1, bound
hit
log-probit
2
0.28
0.87
140.32
2.7E+02
4.0E+00
slope unrestricted
multistage
2
6.45
0.04
149.12
1.1E+04
7.0E+03
betas restricted >0
probit
2
7.75
0.02
151.11
1.9E+04
1.5E+04
Weibull
2
6.45
0.04
149.12
1.1E+04
7.0E+03
power restricted >1, bound
hit
Weibull
3
0.00
1.00
140.03
2.0E+02
1.9E+00
power unrestricted
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-273 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.2.44.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1
Log-Logistic Model with 0.95 Confidence Level
0.6
0.5
0.4
0.3
0.1
0
0 10000 20000 30000
dose
13:39 11/16 2009
E.2.44.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LogLogistic_BMRl_Amyloidosis.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS2l\AD\Blood\LogLogistic_BMRl_Amyloidosis.pit
Mon Nov 16 13:39:45 2009
Table 2
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 25 0
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-274 DRAFT—DO NOT CITE OR QUOTE
Log-Logistic
< BMDL| BMP
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
User has chosen the log transformed model
Default Initial Parameter Values
background = 0
intercept = -10.8548
slope = 1
the user,
background
intercept
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by
and do not appear in the correlation matrix )
background intercept
1 -0.49
-0.49 1
Parameter Estimates
Interval
Variable
Limit
background
intercept
slope
Estimate
0.0699641
-11.2157
1
Std. Err.
95.0% Wald Confidence
Lower Conf. Limit Upper Conf.
* - Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
Analysis of Deviance Table
#
Log(likelihood)
-68.017
-72.1329
-82.0119
Param's
4
2
1
Deviance Test d.f.
8.23187
27.99
P-value
0.01631
<.0001
AIC:
148.266
Dose
Est. Prob.
Goodness of Fit
Expected Observed Size
Scaled
Residual
0.0000
315.4949
7814.0188
50105.0000
0.0700
0.0739
0.1585
0.4446
2. 659
3.251
6. 973
19.117
0. 000
5.000
10.000
17.000
38
44
44
43
-1.691
1.008
1.250
-0.650
Chi^2
5.86
d.f. = 2
P-value
0.0535
Benchmark Dose Computation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-275 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Specified effect =
Risk Type =
Confidence level =
BMD =
BMDL =
0.1
Extra risk
0.95
8254 .29
4805.18
E.2.44.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
0.6
0.5
0.4
0.3
0.2
0.1
B yiDL BMD
13:39 11/16 2009
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
10000
20000 30000
dose
40000
50000
E.2.44.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LogLogistic_Unrest_BMRl_Amyloidosis.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\LogLogistic_Unrest_BMRl_Amyloidosis.pit
Mon Nov 16 13:39:45 2009
Table 2
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-276 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0
intercept = -3.91243
slope = 0.314588
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.98
slope -0.98 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0 * * *
intercept -4.01968 * * *
slope 0.326277 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -68.017 4
Fitted model -68.1202 2 0.206421 2 0.9019
Reduced model -82.0119 1 27.99 3 <.0001
AIC: 140.24
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0000 0.000 0.000 38 0.000
315.4949 0.1051 4.623 5.000 44 0.186
7814.0188 0.2507 11.029 10.000 44 -0.358
50105.0000 0.3802 16.348 17.000 43 0.205
Chi/S2 = 0.20 d.f. = 2 P-value = 0.9028
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-277 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
0.1
Extra risk
0. 95
266.567
2.92895
E.2.44.6. Figure for Unrestricted Model: Log-Probit, Slope Restricted >1
LogProbit Model with 0.95 Confidence Level
LogProbit
B flDUBMD
10000 20000 30000 40000
dose
50000
13:39 11/16 2009
E.2.44.7. Output File for Unrestricted Model: Log-Probit, Slope Restricted >1
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\LogProbit_BMRl_Amyloidosis.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\LogProbit_BMRl_Amyloidosis.plt
Mon Nov 16 13:39:45 2009
Table 2
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope*Log(Dose)),
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-278 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial (and Specified) Parameter Values
background = 0
intercept = -2.2812
slope = 0.180958
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.98
slope -0.98 1
Parameter Estimates
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-3.45134 -1.19365
0.0623389 0.308792
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Variable
background
intercept
slooe
Estimate
0
-2 . 3225
0.185565
Std. Err.
NA
0.57595
0.0628719
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -68.017 4
Fitted model -68.1574 2 0.280896 2 0.869
Reduced model -82.0119 1 27.99 3 <.0001
AIC: 140.315
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0000 0.000 0.000 38 0.000
315.4949 0.1048 4.611 5.000 44 0.192
7814.0188 0.2549 11.216 10.000 44 -0.421
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-279 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
50105.0000 0.3766 16.195 17.000 43 0.253
Chi/S2 = 0.28 d.f. = 2 P-value = 0.8704
Benchmark Dose
Specified effect :
Risk Type
Confidence level :
BMD
BMDL
Computation
0.1
Extra risk
0. 95
273.029
4.02007
E.2.45. Toth et al. (1978): Skin Lesions
E.2.45.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
X
p-Value a
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
gamma
2
6.89
0.03
156.34
5.7E+03
4.1E+03
power restricted >1,
bound hit
logistic
2
10.70
0.00
161.42
1.4E+04
1.1E+04
log-logisticb
2
5.09
0.08
153.96
3.5E+03
2.2E+03
slope restricted >1,
bound hit
log-logisticc
2
0.04
0.95
147.08
2.60E+02
3.18E+01
slope urestricted
log-probit
2
14.29
<0.001
164.79
1.24E+04
8.31E+03
slope restricted >1,
bound hit
log-probit
2
0.80
0.67
147.84
3.3E+02
4.5E+01
slope unrestricted >1
multistage
2
6.89
0.03
156.34
5.7E+03
4.1E+03
betas restricted >0
probit
2
10.39
0.01
160.99
1.3E+04
1.0E+04
Weibull
2
6.89
0.03
156.34
5.7E+03
4.1E+03
power restricted >1,
bound hit
Weibull
2
0.00
1.00
147.04
2.19E+02
2.08E+01
power unrestricted
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-280 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
E.2.45.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1
LogProbit Model with 0.95 Confidence Level
0.7
0.6
0.5
"O
1
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\A\Blood\LogLogistic_BMR2_Skin_lesion_lyr.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS2l\AD\Blood\LogLogistic_BMR2_Skin_lesion_lyr.pit
Mon Nov 16 13:30:07 2009
LogProbit
i/lDL BMP
Table 2
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 25 0
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-281 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
User has chosen the log transformed model
Default Initial Parameter Values
background = 0
intercept = -10.252
slope = 1
the user,
background
intercept
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by
and do not appear in the correlation matrix )
background intercept
1 -0.43
-0.43 1
Parameter Estimates
Interval
Variable
Limit
background
intercept
slope
Estimate
0.0564295
-10.3645
1
Std. Err.
95.0% Wald Confidence
Lower Conf. Limit Upper Conf.
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-71.5177
-74.9791
-95.8498
Param's
4
2
1
Deviance Test d.f.
6.92292
48.6642
P-value
0.03138
<.0001
AIC:
153.958
Dose
Est. Prob.
Goodness of Fit
Expected
Observed
Size
Scaled
Residual
0.0000
315.4949
7814.0188
50105.0000
0.0564
0.0657
0.2430
0.6343
2.144
2.892
10.690
27.273
0. 000
5.000
13.000
25.000
38
44
44
43
-1.508
1.283
0.812
-0.720
Chi^2
5.09
d.f. = 2
P-value
0.0783
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-282 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Benchmark Dose Computation
Specified effect =
Risk Type =
Confidence level =
BMD =
BMDL =
0.1
Extra risk
0.95
3523.85
2211.53
E.2.45.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
07:51 11/27 2009
20000 30000
dose
E.2.45.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\Usepa\Bmds2\Data\LogTcdSet.(d)
Gnuplot Plotting File: C:\Usepa\Bmds2\Data\LogTcdSet.plt
Fri Nov 27 07:51:12 2009
BMDS Model Run
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-283 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
Dependent variable = r_skin
Independent variable = DOSE
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 25 0
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0
intercept = -4.78342
slope = 0.469549
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -background
have been estimated at a boundary point, or have been specified by
the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.98
slope -0.98 1
Parameter Estimates
Interval
Variable
Limit
background
intercept
slope
* - Indicates that this value is not calculated.
95.0% Wald Confidence
Estimate Std. Err. Lower Conf. Limit Upper Conf.
q * * *
-4.84059 * * *
0.475472 * * *
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-71.5177
-71.5376
-95.8498
# Param's
4
2
1
Deviance Test d.f.
0. 0398444
48 . 6642
P-value
0.9803
<.0001
AIC:
147.075
Goodness of Fit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-284 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
Dose
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
0.0000
0. 000
0. 000
38
0.000
316.0000
0.1087
4.784
5.000
44
0.105
4714.0000
0.3060
13.464
13.000
44
-0.152
50105.0000
0.5756
24.753
25.000
43
0.076
Chi^2 = 0.04 d.f. = 2 P-value = 0.9803
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 259.682
BMDL = 31.788
E.2.46. Van Birgelen et al. (1995a): Hepatic Retinol
E.2.46.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
4
<0.0001
41.09
<0.0001
159.73
4.3E+03
2.3E+03
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
40.44
<0.0001
159.09
3.4E+04
2.4E+03
nonconstant variance,
power restricted >1
exponential (M4)
C
3
<0.0001
20.80
0.00
141.45
1.4E+04
1.9E+03
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
20.80
0.00
141.45
1.4E+04
1.9E+03
nonconstant variance,
power restricted >1
exponential (M5)d
3
<0.0001
20.80
0.00
141.45
1.4E+04
1.9E+03
nonconstant variance,
power unrestricted
Hill
3
<.0001
4.22
0.24
124.86
2.9E+03
error
nonconstant variance, n
restricted >1, bound hit
Hilld
2
<.0001
2.85
0.24
125.50
2.0E+03
error
nonconstant variance, n
unrestricted
linear
4
<.0001
58.18
<.0001
176.83
1.0E+05
7.9E+04
nonconstant variance
polynomial
4
<.0001
58.18
<.0001
176.83
1.0E+05
7.9E+04
nonconstant variance
power
4
<.0001
58.18
<.0001
176.83
1.0E+05
7.9E+04
nonconstant variance,
power restricted >1,
bound hit
power d
3
<.0001
11.12
0.01
131.77
2.1E+02
7.7E+00
nonconstant variance,
power unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-285 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
4
<0.0001
3.87
0.42
184.19
3.0E+03
1.9E+03
constant variance, power
restricted >1
exponential (M3)
4
<0.0001
3.87
0.42
184.19
3.0E+03
1.9E+03
constant variance, power
restricted >1
exponential (M4)
3
<0.0001
1.84
0.61
184.15
2.7E+03
1.7E+03
constant variance, power
restricted >1
exponential (M5)
3
<0.0001
1.84
0.61
184.15
2.7E+03
1.7E+03
constant variance, power
restricted >1
exponential (M5)
3
<0.0001
1.84
0.61
184.15
2.7E+03
1.7E+03
constant variance, power
unrestricted
Hill
3
<.0001
1.04
0.79
183.36
2.1E+03
1.1E+03
constant variance, n
restricted >1, bound hit
Hill
2
<.0001
0.98
0.61
185.29
1.7E+03
4.0E+01
constant variance, n
unrestricted
linear
4
<.0001
25.63
<.0001
205.94
6.8E+04
5.0E+04
constant variance
polynomial
4
<.0001
25.63
<.0001
205.94
6.8E+04
5.0E+04
constant variance
power
4
<.0001
25.63
<.0001
205.94
6.8E+04
5.0E+04
constant variance, power
restricted >1, bound hit
power
3
<.0001
2.28
0.52
184.60
2.1E+02
6.2E+00
constant variance, power
unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meetBMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
1
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-286 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.46.2.
Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
20
15
10
5
0
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:28 11/20 2009
E.2.46.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_BMRl_hepatic_retinol.(d)
Gnuplot Plotting File:
Fri Nov 20 12:28:01 2009
Tbl3, hepatic retinol
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-287 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -1.16065
rho 1.53688
a 15.645
b 4.61687e-005
c 0.0365247
d 1
Parameter Estimates
Variable
lnalpha
rho
Model 4
-0.926841
I.77261
II.5052
5 . 20223e-005
0.0653036
1
Table of Stats From Input Data
Dose N
0
3969
647 9
9968
4.761e+004
1.37 8e + 005
Obs Mean
14
9
8 . 4
8 . 2
5.1
2 . 2
0.6
Obs Std Dev
8.768
3.394
2 .263
0. 8485
0. 8485
0.5657
Dose
0
3969
647 9
9968
4.761e+004
1.37 8e + 005
Estimated Values of Interest
Est Mean Est Std Scaled Residual
11. 51
9.499
8 .428
7 .154
1. 655
0.7596
5. 483
4 . 627
4 .161
3.599
0.9832
0.4931
1.751
-0.6719
-0.1551
-1.614
1. 568
-0.9156
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-288 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -87.1567 7 188.3134
A2 -47.28742 12 118.5748
A3 -55.32422 8 126.6484
R -109.967 2 223.934
4 -65.72639 5 141.4528
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2^log(Likelihood Ratio)
125. 4
79.74
16. 07
20.8
10
5
4
3
p-value
< 0.0001
< 0.0001
0.002922
0.0001156
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 6a is less than .1. Model 4 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 13706.9
BMDL = 1852.89
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-289 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.46.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
20
15
10
5
0
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:28 11/20 2009
E.2.46.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_Unrest_BMRl_hepatic_retinol.(d)
Gnuplot Plotting File:
Fri Nov 20 12:28:10 2009
Tbl3, hepatic retinol
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-290 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -1.16065
rho 1.53688
a 15.645
b 4.61687e-005
c 0.0365247
d 1
Parameter Estimates
Variable
lnalpha
rho
Model 5
-0.926841
I.77261
II.5052
5 . 20223e-005
0.0653036
1
Table of Stats From Input Data
Dose N
0
3969
647 9
9968
4.761e+004
1.37 8e + 005
Obs Mean
14
9
8 . 4
8 . 2
5.1
2 . 2
0.6
Obs Std Dev
8.768
3.394
2 .263
0. 8485
0. 8485
0.5657
Dose
0
3969
647 9
9968
4.761e+004
1.37 8e + 005
Estimated Values of Interest
Est Mean Est Std Scaled Residual
11. 51
9.499
8 .428
7 .154
1. 655
0.7596
5. 483
4 . 627
4 .161
3.599
0.9832
0.4931
1.751
-0.6719
-0.1551
-1.614
1. 568
-0.9156
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-291 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -87.1567 7 188.3134
A2 -47.28742 12 118.5748
A3 -55.32422 8 126.6484
R -109.967 2 223.934
5 -65.72639 5 141.4528
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2^log(Likelihood Ratio)
125. 4
79.74
16. 07
20.8
10
5
4
3
p-value
< 0.0001
< 0.0001
0.002922
0.0001156
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is less than .1. Model 5 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 13706.9
BMDL = 1852.89
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-292 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.46.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
Hill
20
15
10
N
\
5
0
BMD
0
20000
40000
60000
80000 100000 120000 140000
dose
12:28 11/20 2009
E.2.46.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Hill_Unrest_BMRl_hepatic_retinol.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_hepatic_retinol.pit
Fri Nov 20 12:28:12 2009
Tbl3, hepatic retinol
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
Default Initial Parameter Values
1/15/10
E-293 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
lalpha = 2.76506
rho = 0
intercept = 14.9
v = -14.3
n = 3.62162
k = 6985.28
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -0.78 -0.04 0.012 0.036 0.033
rho -0.78 1 -0.099 0.12 -0.046 -0.052
intercept -0.04 -0.099 1 -0.94 -0.25 -0.81
v 0.012 0.12 -0.94 1 0.54 0.75
n 0.036 -0.046 -0.25 0.54 1 0.31
k 0.033 -0.052 -0.81 0.75 0.31 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
-1.16164
1.69911
16.6709
-17.0495
0.763329
4251.89
Std. Err.
0.374355
0.18585
2.08161
2.32002
0.19632
1440.45
Lower Conf. Limit
-1.89536
1. 33485
12 .591
-21.5967
0.378549
1428.66
Upper Conf. Limit
-0.427913
2.06337
20.7508
-12.5023
1.14811
7075.13
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8
3969 8
647 9 8
9968 8
4.761e+004
1.37 8e + 005
14
9
8 . 4
8 . 2
5.1
2 . 2
0.6
16.7
8 . 37
6.79
5.47
1. 95
0.741
8 .77
3.39
2 .26
0.849
0.849
0.566
6.11
3.4
2 . 85
2 . 37
0. 987
0. 434
-0. 82
0.0248
1. 4
-0.439
0.716
-0.919
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-294 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -87.156698 7 188.313395
A2 -47.287416 12 118.574833
A3 -55.324218 8 126.648436
fitted -56.747514 6 125.495027
R -109.967018 2 223.934036
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.'
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 125.359 10 <.0001
Test 2 79.7386 5 <.0001
Test 3 16.0736 4 0.002922
Test 4 2.84659 2 0.2409
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 198 0.8 8
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-295 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.46.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
S/IDLBMD
20000
40000
60000
80000
100000 120000 140000
E.2.46.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_hepatic_retinol.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_hepatic_retinol.plt
Fri Nov 20 12:28:14 2009
Tbl3, hepatic retinol
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-296 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha = 2.76506
rho = 0
control = 14.9
slope = -0.92667
power = 0.231239
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0.8
-0.042
0. 048
0. 063
rho
-0.8
1
-0.089
-0.038
-0.1
control
-0.042
-0.089
1
-0. 91
-0. 81
slope
0. 048
-0.038
-0. 91
1
0. 98
power
0. 063
-0.1
-0. 81
0. 98
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
-0.986245
1.67858
16.9266
-3.10665
0.139874
Std. Err.
0.394723
0.202896
2.23237
1.35883
0.0269583
Lower Conf. Limit
-1.75989
1.28091
12.5513
-5.76991
0.0870372
Upper Conf. Limit
-0.212602
2 . 07625
21.302
-0.443384
0.192712
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8
3969 8
647 9 8
9968 8
4.761e+004
1.37 8e + 005
14
9
8 . 4
8 . 2
5.1
2 . 2
0.6
16. 9
7 . 03
6.32
5. 67
2 . 91
0.666
8 .77
3.39
2 .26
0.849
0.849
0.566
6.56
3.14
2 . 87
2 . 62
1. 5
0. 434
-0.874
1.24
1. 85
-0.611
-1. 34
-0.427
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma^2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-297 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Model Log(likelihood) # Param's AIC
A1 -87.156698 7 188.313395
A2 -47.287416 12 118.574833
A3 -55.324218 8 126.648436
fitted -60.885852 5 131.771704
R -109.967018 2 223.934036
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2^log(Likelihood Ratio) Test df p-value
Test 1 125.359 10 <.0001
Test 2 79.7386 5 <.0001
Test 3 16.0736 4 0.002922
Test 4 11.1233 3 0.01108
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 209.4 98
BMDL = 7.67456
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-298 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.2.47. Van Birgelen et al. (1995a): Hepatic Retinol Palmitate
2 E.2.47.1. Summary Table of BMDS Modeling Results
Model
Degrees of
Freedom
Variance
/7-Value a
J2 Test
Statistic
X2P-
Valueb
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
exponential (M2)
4
<0.0001
57.51
<0.0001
460.28
error
error
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
57.51
<0.0001
460.28
error
error
nonconstant variance,
power restricted >1
exponential (M4)c
3
<0.0001
42.23
<0.0001
446.99
7.8E+04
2.0E+04
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
42.23
<0.0001
446.99
7.8E+04
2.0E+04
nonconstant variance,
power restricted >1
exponential (M5)d
3
<0.0001
42.23
<0.0001
446.99
7.8E+04
2.0E+04
nonconstant variance,
power unrestricted
Hill
3
<.0001
11.47
0.01
416.23
2.0E+03
error
nonconstant variance, n
restricted >1, bound hit
Hilld
3
<.0001
120.59
<.0001
525.36
5.0E-11
5.0E-11
nonconstant variance, n
unrestricted
linear
4
<.0001
83.61
<.0001
486.37
1.9E+05
1.3E+05
nonconstant variance
polynomial
4
<.0001
128.71
<.0001
531.47
6.2E+04
5.0E+04
nonconstant variance
power
4
<.0001
83.61
<.0001
486.37
1.9E+05
1.3E+05
nonconstant variance,
power restricted >1, bound
hit
power d
3
<.0001
4.22
0.24
408.98
2.9E+01
3.2E-02
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
142.00
<0.0001
649.06
error
error
constant variance, power
restricted >1
exponential (M3)
4
<0.0001
142.00
<0.0001
649.06
error
error
constant variance, power
restricted >1
exponential (M4)
3
<0.0001
2.84
0.42
511.95
7.9E+02
2.9E+00
constant variance, power
restricted >1
exponential (M5)
3
<0.0001
2.84
0.42
511.95
7.9E+02
2.0E+00
constant variance, power
restricted >1
exponential (M5)
3
<0.0001
2.84
0.42
511.95
7.9E+02
2.0E+00
constant variance, power
unrestricted
Hill
3
<.0001
0.93
0.82
510.04
3.9E+02
9.5E+01
constant variance, n
restricted >1, bound hit
Hill
2
<.0001
0.31
0.86
511.42
2.8E-01
2.8E-01
constant variance, n
unrestricted
linear
4
<.0001
43.71
<.0001
550.82
1.1E+05
7.0E+04
constant variance
polynomial
4
<.0001
43.71
<.0001
550.82
1.1E+05
7.0E+04
constant variance
power
4
<.0001
43.71
<.0001
550.82
1.1E+05
7.0E+04
constant variance, power
restricted >1, bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-299 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Model
Degrees of
Freedom
Variance
/7-Value a
J2 Test
Statistic
X2P-
Value"
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
power
3
<.0001
0.33
0.95
509.44
2.0E-04
2.0E-04
constant variance, power
unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
E.2.47.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
700
600
500
400
300
200
100
BMDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:29 11/20 2009
E.2.47.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_BMRl_hepatic_retinol_palmitate.(d)
Gnuplot Plotting File:
Fri Nov 20 12:29:00 2009
Tbl3, hepatic retinol palmitate
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-300 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'""d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)'""d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha 0.284674
rho 1.77158
a 4 95.6
b 6.13207e-005
c 0.00576502
d 1
Parameter Estimates
Variable Model 4
NC = No Convergence
lnalpha -0.241584
rho 2.03456
a 223.851
b 5.45885e-005
c 0.012925
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 472 271.5
3969 8 94 67.88
6479 8 107 76.37
9968 8 74 39.6
4.7 61e + 004 8 22 22.63
1.37 8e + 005 8 3 2.828
Estimated Values of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-301 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dose
Est Mean
Est Std
Scaled Residual
0
3969
647 9
9968
4.761e+004
1.37 8e + 005
223. 9
180.8
158
131.1
19.33
3. 013
217 . 8
175.3
152 . 9
126. 4
18 . 03
2 .721
3.222
-1.401
-0.9443
-1.278
0.4197
-0. 01317
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -250.5548 7 515.1096
A2 -196.7557 12 417.5115
A3 -197.3832 8 410.7663
R -276.7896 2 557.5793
4 -218.4969 5 446.9938
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
160.1
107 . 6
1. 255
42 . 23
10
5
4
3
p-value
< 0.0001
< 0.0001
0.869
< 0.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-302 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
variance appears to be appropriate here.
The p-value for Test 6a is less than .1. Model 4 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 77948.7
BMDL = 20092.3
E.2.47.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
700
600
500
400
300
200
100
Exponential
BMDL
20000 40000 60000 80000 100000 120000 140000
dose
12:29 11/20 2009
E.2.47.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File:
C:\USEPA\BMDS21\Nov2 0\Blood\Exp_Unrest_BMRl_hepatic_retinol_palmitate.(d)
Gnuplot Plotting File:
Fri Nov 20 12:29:18 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-303 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Tbl3, hepatic retinol palmitate
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)'""d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha 0.284674
rho 1.77158
a 4 95.6
b 6.13207e-005
c 0.00576502
d 1
Parameter Estimates
Variable Model 5
No Convergence
lnalpha -0.241584
rho 2.03456
a 223.851
b 5.45885e-005
c 0.012925
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 472 271.5
3969 8 94 67.88
6479 8 107 76.37
9968 8 74 39.6
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-304 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
4.761e+004
1.37 8e + 005
22
3
22 . 63
2 . 828
Estimated Values of Interest
Dose
0
3969
647 9
9968
4.761e+004
1.37 8e + 005
Est Mean
223. 9
180.8
158
131.1
19.33
3. 013
Est Std
217 . 8
175.3
152 . 9
126. 4
18 . 03
2 .721
Scaled Residual
3.222
-1.401
-0.9443
-1.278
0.4197
-0. 01317
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -250.5548 7 515.1096
A2 -196.7557 12 417.5115
A3 -197.3832 8 410.7663
R -276.7896 2 557.5793
5 -218.4969 5 446.9938
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
160.1
107 . 6
1. 255
42 . 23
10
5
4
3
p-value
< 0.0001
< 0.0001
0.869
< 0.0001
The p-value for Test 1 is less than .05. There appears to be
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-305 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is less than .1. Model 5 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 77 952.3
BMDL = 20092.3
E.2.47.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
700
600
500
400
300
200
100
B tfDLBMD
20000 40000 60000 80000 100000 120000 140000
dose
12:29 11/20 2009
E.2.47.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File:
C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_hepatic_retinol_palmitate.(d)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-306 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Gnuplot Plotting File:
C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_hepatic_retinol_palmitate.pit
Fri Nov 20 12:29:21 2009
Tbl3, hepatic retinol palmitate
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 9.57332
rho = 0
intercept = 472
v = -469
n = 1.6454
k = 2462.18
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -k
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha rho intercept v n
lalpha 1 -0.9 -0.0084 -0.05 0.00043
rho -0.9 1 0.33 -0.25 6.2e-005
intercept
v
n
-0.0084
-0. 05
0.00043
0.33
-0.25
6 . 2e-005
1
-1
0.00089
-1
1
-0.00081
0. 00089
-0.00081
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha 9.05753 0.813787 7.46254 10.6525
rho 0.296518 0.132793 0.0362478 0.556789
intercept 733.34 146.204 446.785 1019.89
v -707.607 132.71 -967.713 -447.5
n 0.620012 31.4549 -61.0305 62.2705
k 1.3782e-010 NA
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-307 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 e
3969 £
647 9 8
9968 8
4.761e+004
1.37 8e + 005
472
94
107
74
22
3
733
25.7
25.7
25.7
272
25.7
25.7
67 .
76.
39.
22 . 6
2 . 83
246
150
150
150
150
150
-3
1.29
1. 53
0. 91
-0.0704
-0.429
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -250.554817 7 515.109634
A2 -196.755746 12 417.511491
A3 -197.383174 8 410.766347
fitted -257.680271 5 525.360542
R -276.789644 2 557.579287
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
160.068
107.598
1. 25486
120.594
10
5
4
3
<.0001
<.0001
0.869
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-308 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 5.0137 6e-011
BMDL = 5.0137 6e-011
E.2.47.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
700
600
500
400
300
200
100
Power
B VIDLBMD
Power Model with 0.95 Confidence Level
~T
20000 40000 60000 80000 100000 120000 140000
dose
12:29 11/20 2009
E.2.47.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File:
C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_hepatic_retinol_palmitate.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_hepatic_retinol_palmitate.pit
Fri Nov 20 12:29:22 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-309 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Tbl3, hepatic retinol palmitate
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 9.57332
rho = 0
control = 472
slope = -204.597
power = 0.0711193
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho control slope power
lalpha 1 -0.95 0.3 -0.32 -0.3
rho -0.95 1 -0.41 0.37 0.29
control 0.3 -0.41 1 -0.96 -0.82
slope -0.32 0.37 -0.96 1 0.95
power -0.3 0.29 -0.82 0.95 1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
0.064014
1.81132
464.289
-216.594
0.0639105
Std. Err.
0.859473
0.197468
87.5706
73.4027
0.0139781
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-1.62052
1.42429
292.654
-360.461
0. 0365139
1.74855
2 .19835
635.925
-72.7275
0.0913071
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8
3969 8
647 9 8
9968 8
4.761e+004
1.37 8e + 005
472
94
107
74
464
96.5
84 . 8
74.2
272
67 .
76.
39.
269
64 . 7
57 . 6
51
22
3
33.2
2 .86
22 . 6
2 . 83
24 .
2 . t
0.0812
-0.108
1.09
-0.00938
-1.28
0.145
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-310 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
Likelihoods of Interest
Model Log(likelihood) # Param'
A1 -250.554817 7
A2 -196.755746 12
A3 -197.383174 8
fitted -199.490894 5
R -276.789644 2
AIC
515.109634
417.511491
410.766347
408.981788
557.579287
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
160.068
107.598
1. 25486
4 . 21544
10
5
4
3
<.0001
<.0001
0.869
0.2391
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-311 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 BMD = 28.9991
3
4
5 BMDL = 0.0324608
6
7
8
9 E.2.48. Van Birgelen et al. (1995a): Plasma FT4
10 E.2.48.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)c
4
0.01
2.70
0.61
214.98
4.4E+04
2.8E+04
nonconstant variance,
power restricted >1
exponential (M3)
4
0.01
2.70
0.61
214.98
4.4E+04
2.8E+04
nonconstant variance,
power restricted >1
exponential (M4)
3
0.01
1.96
0.58
216.24
3.0E+04
1.2E+04
nonconstant variance,
power restricted >1
exponential (M5)
3
0.01
1.96
0.58
216.24
3.0E+04
1.2E+04
nonconstant variance,
power restricted >1
exponential (M5)d
3
0.01
1.96
0.58
216.24
3.0E+04
1.2E+04
nonconstant variance,
power unrestricted
Hill
3
0.01
1.90
0.59
216.19
2.8E+04
8.8E+03
nonconstant variance, n
restricted >1, bound hit
Hilld
2
0.01
1.90
0.39
218.19
2.8E+04
7.9E+03
nonconstant variance, n
unrestricted
linear
4
0.01
3.98
0.41
216.27
6.0E+04
4.4E+04
nonconstant variance
polynomial
4
0.01
3.98
0.41
216.27
6.0E+04
4.4E+04
nonconstant variance
power
4
0.01
3.98
0.41
216.27
6.0E+04
4.4E+04
nonconstant variance,
power restricted >1,
bound hit
power d
3
0.01
2.30
0.51
216.59
3.0E+04
7.2E+03
nonconstant variance,
power unrestricted
exponential (M2)
4
0.01
3.21
0.52
213.50
4.1E+04
2.7E+04
constant variance,
power restricted >1
exponential (M3)
4
0.01
3.21
0.52
213.50
4.1E+04
2.7E+04
constant variance,
power restricted >1
exponential (M4)
3
0.01
2.47
0.48
214.76
2.7E+04
1.1E+04
constant variance,
power restricted >1
exponential (M5)
3
0.01
2.47
0.48
214.76
2.7E+04
1.1E+04
constant variance,
power restricted >1
exponential (M5)
3
0.01
2.47
0.48
214.76
2.7E+04
1.1E+04
constant variance,
power unrestricted
Hill
3
0.01
2.35
0.50
214.64
2.4E+04
8.1E+03
constant variance, n
restricted >1, bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-312 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
Hill
2
0.01
2.33
0.31
216.62
2.4E+04
7.0E+03
constant variance, n
unrestricted
linear
4
0.01
4.50
0.34
214.79
5.7E+04
4.3E+04
constant variance
polynomial
4
0.01
4.50
0.34
214.79
5.7E+04
4.3E+04
constant variance
power
4
0.01
4.50
0.34
214.79
5.7E+04
4.3E+04
constant variance,
power restricted >1,
bound hit
power
3
0.01
2.66
0.45
214.95
2.6E+04
6.4E+03
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
1
2
3 E.2.48.2. Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
4 Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
30
25
20
15
10
5
BMDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
5 12:30 11/20 2009
6
7
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-313 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.48.3. Output File for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_BMRl_plasma_FT4.(d)
Gnuplot Plotting File:
Fri Nov 20 12:30:05 2009
Tbl3, plasma FT4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
lnalpha 4.29134
rho -0.423761
a 25.725
b 2.47112e-005
c 0.381323
d 1
Parameter Estimates
Variable Model 2
lnalpha 1.7323
rho 0.534787
a 23.5975
b 1.50877e-005
c 0.358997
d 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-314 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Stats From Input Data
Dose
N
Obs
Mean
Obs Std Dev
0
8
23
. 4
3.111
3969
8
24
. 5
5. 657
647 9
8
22
. 4
2 . 828
9968
8
19
. 3
9.334
4.761e+004
8
16.3
4 .243
1.37 8e + 005
8
10.3
4 .808
Estimated
Values of
Interest
Dose
Est Mean
Est Std
Scaled Residi
0
23. 03
5.531
0.1896
3969
22 .47
5. 496
1.046
647 9
22 .12
5.474
0.1445
9968
21. 65
5.444
-1.219
7 61e + 004
17 .13
5.13
-0.4583
37 8e + 005
9.779
4.447
0.3314
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -102.145 7 218.2901
A2 -94.04963 12 212.0993
A3 -102.143 8 220.286
R -117.8175 2 239.635
2 -103.491 4 214.9821
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 47.54 10 < 0.0001
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-315 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Test 2
Test 3
Test 4
16.19
16.19
2 .696
5
4
4
0.0 0 632
0.002778
(j . 6 (J 9 9
The p-value for Test 1 is less than .05. There appears to be a
difference between resp'Orise andAor variances among the dase
levels, it seems appropriate to' model the 'data.
The p-value tor Test 2 is less than .1. A riO'ri-hO'mogerieO'US
variance model appears t'O be appropriate.
The p-value tor Test 3 is less than .1. YU'U may want t'O
cunsider a 'different variance model.
The p-value tor Test 4 is greater than .1. Model 2 seems
t'O adeguately describe the data.
E'enchmark DO'Se C'Oirputati'Oris :
Specified Effect = 1.000000
Risk Typ'e = Estimated standard deviations frc'in control
Confidence Level = 0.950000
E'MD = 4 4193.5
E'MDL = 2 8156.1
E.2.48.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
30
25
20
15
10
BMD
5
BMDL
0
20000
40000
60000
80000
100000
120000
140000
dose
12:30 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-316 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.2.48.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_Unrest_BMRl_plasma_FT4.(d)
Gnuplot Plotting File:
Fri Nov 20 12:30:11 2009
Tbl3, plasma FT4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha 4.29134
rho -0.423761
a 25.725
b 2.47112e-005
c 0.381323
d 1
Parameter Estimates
Variable Model 5
lnalpha 1.7323
rho 0.534787
a 23.5975
b 1.50877e-005
c 0.358997
d 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-317 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Stats From Input Data
Dose
Obs Mean
Obs Std Dev
0
3969
647 9
9968
4.761e+004
1.37 8e + 005
23. 4
24 . 5
22 . 4
19.3
16.3
10.3
3.111
5. 657
2 . 828
9.334
4 .243
4 .808
Estimated Values of Interest
Dose
0
3969
647 9
9968
4.761e+004
1.37 8e + 005
Est Mean
23. 6
22 .72
22.19
21.49
15. 85
10.36
Est Std
5.537
5. 481
5.446
5.4
4 . 978
4.443
Scaled Residual
-0.1009
0.9194
0.1096
-1.145
0.2575
-0.03965
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -102.145 7 218.2901
A2 -94.04963 12 212.0993
A3 -102.143 8 220.286
R -117.8175 2 239.635
5 -103.1224 5 216.2449
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
-2*log(Likelihood Ratio) D. F. p-value
Test
1:
Test
2 :
Test
3:
Test
7a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-318 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Test 1 47.54 10 < 0.0001
Test 2 16.19 5 0.00632
Test 3 16.19 4 0.002778
Test 7a 1.959 3 0.581
The p-value for Test 1 is less than .05. There appears to be a
difference between resp'Orise andAor variances among the duse
levels, it seems appropriate to' model the 'data.
The p-value tor Test 2 is less than .1. A riO'ri-hO'mogerieO'US
variance model appears to' be appropriate.
The p-value tor Test 3 is less than .1. YO'U may want to'
cunsider a 'different variance model.
The p-value tor Test 7a is greater than .1. Model 5 seems
t'O adeguately describe the data.
E'enchmark DO'Se C'Oirputati'Oris :
Specified Effect = 1.000000
Risk Typ'e = Estimated standard deviations frc'in centre,1
Confidence Level = 0.950000
EMD = 3 0208.5
EMDL = 1227 3.2
E.2.48.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
Hill
30
25
20
15
10
5
BMDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:30 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-319 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.2.48.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Hill_Unrest_BMRl_plasma_FT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_plasma_FT4.plt
Fri Nov 20 12730:12 2009
Tbl3, plasma FT4
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 3.38957
rho = 0
intercept = 23.4
v = -13.1
n = 0.996796
k = 40705.6
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -1 0.2 -0.16 -0.19 0.12
rho -1 1 -0.2 0.16 0.19 -0.12
intercept 0.2 -0.2 1 -0.39 -0.59 0.22
v -0.16 0.16 -0.39 1 0.9 -0.98
n -0.19 0.19 -0.59 0.9 1 -0.85
k 0.12 -0.12 0.22 -0.98 -0.85 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha 1.81761 2.18889 -2.47254 6.10776
rho 0.505217 0.744572 -0.954117 1.96455
intercept 23.7748 1.72532 20.3933 27.1564
v -21.6283 20.8713 -62.5352 19.2786
n 0.975863 0.6937 -0.383764 2.33549
k 83458.4 171511 -252696 419613
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-320 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 e
3969 £
647 9 e
9968 8
4.761e+004
1.37 8e + 005
23. 4
24 . 5
22 . 4
19.3
16.3
10.3
23. 8
22 . 7
22 .1
21. 4
15. 9
10.4
3.11
5. 66
2 . 83
9.33
24
81
5.52
5.46
5.43
5.38
99
48
-0.192
0. 921
0.143
-1. 08
0.255
-0.0414
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model
Log(likelihood)
# Param's
AIC
A1
-102.145036
7
218.290071
A2
-94.049629
12
212 .099258
A3
-102.143023
8
220.286046
fitted
-103.092664
6
218 .185329
R
-117.817514
2
239.635028
Explanation of Tests
Test 1:
Test 2
Test 3
Test 4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
47.5358
16.1908
16.1868
1.89928
10
5
4
2
<.0001
0.00632
0.002778
0.3869
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than
model appears to be appropriate
.1. A non-homogeneous variance
The p-value for Test 3 is less than .1. You may want to consider a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-321 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
different variance model
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 27884.5
BMDL = 7 907.26
E.2.48.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
30
25
20
15
10
5
BMDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:30 11/20 2009
E.2.48.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_plasma_FT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Pwr_Unrest_BMRl_plasma_FT4.plt
Fri Nov 20 12:30:13 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-322 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Tbl3, plasma FT4
The form of the response function is:
Y[dose] = control + slope 'k doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 3.38957
rho = 0
control = 2 4.5
slope = -0.0256219
power = 0.537235
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho control slope power
lalpha 1 -1 0.099 -0.069 -0.06
rho -1 1 -0.1 0.069 0.06
control 0.099 -0.1 1 -0.78 -0.75
slope -0.069 0.069 -0.78 1 1
power -0.06 0.06 -0.75 1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
1.99957
0.44594
24.0444
-0.0113184
0.601415
Std. Err.
2.14696
0.730207
1.65932
0.0287697
0.209424
Lower Conf. Limit
-2.20839
-0.98524
20.7922
-0.0677059
0.190952
Upper Conf. Limit
6.20753
1.87712
27.2966
0.0450692
1.01188
Table of Data and Estimated Values of Interest
e N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 23.4 24 3.11 5.52 -0.33
3969 8 24.5 22.4 5.66 5.44 1.1
6479 8 22.4 21.8 2.83 5.4 0.3
9968 8 19.3 21.2 9.33 5.37 -0.985
4.7 61e + 004 8 16.3 16.7 4.24 5.09 -0.212
1.37 8e + 005 8 10.3 10.1 4.81 4.55 0.129
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-323 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -102.145036 7 218.290071
A2 -94.049629 12 212.099258
A3 -102.143023 8 220.286046
fitted -103.295375 5 216.590750
R -117.817514 2 239.635028
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
47.5358
16.1908
16.1868
2.3047
10
5
4
3
<.0001
0.00632
0.002778
0.5116
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 29513.9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-324 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3 BMDL = 7158.08
4
5
6
7 E.2.49. Van Birgelen et al. (1995a): Plasma TT4
8 E.2.49.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
4
0.94
9.91
0.04
241.35
5.6E+04
3.6E+04
nonconstant variance,
power restricted >1
exponential (M2)
4
0.94
9.91
0.04
241.35
5.6E+04
3.6E+04
nonconstant variance,
power unrestricted
exponential (M3)
4
0.94
9.91
0.04
241.35
5.6E+04
3.6E+04
nonconstant variance,
power restricted >1
exponential (M3)
4
0.94
9.91
0.04
241.35
5.6E+04
3.6E+04
nonconstant variance,
power unrestricted
exponential (M4)
3
0.94
9.33
0.03
242.77
3.6E+04
6.8E+03
nonconstant variance,
power restricted >1
exponential (M4)
3
0.94
9.33
0.03
242.77
3.6E+04
6.8E+03
nonconstant variance,
power unrestricted
exponential (M5)
3
0.94
9.33
0.03
242.77
3.6E+04
5.5E+03
nonconstant variance,
power restricted >1
exponential (M5)
3
0.94
9.33
0.03
242.77
3.6E+04
5.5E+03
nonconstant variance,
power unrestricted
Hill
3
0.94
5.45
0.14
238.89
9.4E+03
error
nonconstant variance, n
restricted >1, bound hit
Hill
3
0.94
5.45
0.14
238.89
9.4E+03
error
nonconstant variance, n
unrestricted
linear
4
0.94
10.33
0.04
241.77
6.6E+04
4.5E+04
nonconstant variance
polynomial
4
0.94
10.33
0.04
241.77
6.6E+04
4.5E+04
nonconstant variance
power
4
0.94
10.33
0.04
241.77
6.6E+04
4.5E+04
nonconstant variance,
power restricted >1,
bound hit
power
3
0.94
8.78
0.03
242.22
2.9E+04
5.4E+03
nonconstant variance,
power unrestricted
exponential (M2)
C
4
0.94
9.33
0.05
239.35
5.7E+04
3.9E+04
constant variance,
power restricted >1
exponential (M3)
4
0.94
9.33
0.05
239.35
5.7E+04
3.9E+04
constant variance,
power restricted >1
exponential (M4)
3
0.94
8.75
0.03
240.78
3.7E+04
9.1E+03
constant variance,
power restricted >1
exponential (M5)
3
0.94
8.75
0.03
240.78
3.7E+04
9.1E+03
constant variance,
power restricted >1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-325 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M5)d
3
0.94
8.75
0.03
240.78
3.7E+04
9.1E+03
constant variance,
power unrestricted
Hill
3
0.94
6.31
0.10
238.33
9.5E+03
error
constant variance, n
restricted >1, bound hit
Hilld
3
0.94
6.31
0.10
238.33
9.5E+03
error
constant variance, n
unrestricted
linear
4
0.94
9.75
0.04
239.77
6.6E+04
4.8E+04
constant variance
polynomial
4
0.94
9.75
0.04
239.77
6.6E+04
4.8E+04
constant variance
power
4
0.94
9.75
0.04
239.77
6.6E+04
4.8E+04
constant variance,
power restricted >1,
bound hit
power d
3
0.94
8.21
0.04
240.24
3.0E+04
5.8E+03
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meetBMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
1
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-326 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.49.2. Figure for Selected Model: Exponential (M2), Constant Variance, Power Restricted
>1
Exponential_beta Model 2 with 0.95 Confidence Level
45
40
35
30
25
20
Exponential
BMDL
BMD
0 20000 40000 60000 80000 100000 120000 140000
dose
12:31 11/20 2009
E.2.49.3. Output File for Selected Model: Exponential (M2'), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_CV_BMRl_plasma_TT4.(d)
Gnuplot Plotting File:
Fri Nov 20 12:31:00 2009
Tbl3, plasma TT4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-327 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 2
lnalpha
rho(S)
3.66719
0
43.47
1.9827 7e-005
0.558678
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 2
3. 84955
0
40.4479
1.3887 6e-005
0.575097
1
Table of Stats From Input Data
Dose
N
Obs Mean
Obs Std Dev
0
8
40. 9
6.788
3969
8
41. 4
5.374
647 9
8
41. 4
6.505
9968
8
32 . 3
7 . 354
4.761e+004
8
33.
6
6.223
1.37 8e + 005
8
25.
5
7 . 637
Estimated Values
of
Interest
Dose
Est Mean
Est
Std
Scaled Residi
0
39.73
6 .
895
0.4797
3969
39.2
6 .
895
0. 901
647 9
U)
CO
CO
-J
6 .
895
1. 036
9968
C\]
CO
CO
6 .
895
-2.511
7 61e + 004
33. 85
6 .
895
-0.1024
37 8e + 005
24 . 99
6 .
895
0.2106
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-328 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) ^ rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma"2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -112.0125 7 238.025
A2 -111.4015 12 246.8029
A3 -112.0125 7 238.025
R -127.4455 2 258.891
2 -116.6748 3 239.3495
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Test
4 :
Does response and/or variances differ among Dose levels:
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
(A2 vs. R)
Test
Test 1
Test 2
Test 3
Test 4
Tests of Interest
-2^log(Likelihood Ratio)
32 .09
1. 222
1. 222
9.325
p-value
10
5
5
4
0.0003871
0.9427
0.9427
0.05348
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is less than .1. Model 2 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 56637.1
3MDL = 38643.8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-329 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.49.4. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
45
40
35
30
25
20
BMDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:31 11/20 2009
E.2.49.5. Output File for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_CV_Unrest_BMRl_plasma_TT4.(d)
Gnuplot Plotting File:
Fri Nov 20 12:31:08 2009
Tbl3, plasma TT4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-330 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 5
lnalpha
rho(S)
3.66719
0
43.47
1.9827 7e-005
0.558678
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 5
3. 84955
0
40.4479
1.3887 6e-005
0.575097
1
Table of Stats From Input Data
Dose
N
Obs Mean
Obs Std Dev
0
8
40. 9
6.788
3969
8
41. 4
5.374
647 9
8
41. 4
6.505
9968
8
32 . 3
7 . 354
4.761e+004
8
33.
6
6.223
1.37 8e + 005
8
25.
5
7 . 637
Estimated Values
of
Interest
Dose
Est Mean
Est
Std
Scaled Residi
0
40.45
6 .
854
0.1866
3969
39.53
6 .
854
0 .7733
647 9
38 . 97
6 .
854
1. 003
9968
38 . 23
6 .
854
-2.446
7 61e + 004
32 .13
6 .
854
0. 6049
37 8e + 005
25. 8
6 .
854
-0.1223
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-331 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) ^ rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma"2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -112.0125 7 238.025
A2 -111.4015 12 246.8029
A3 -112.0125 7 238.025
R -127.4455 2 258.891
5 -116.3891 4 240.7782
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
32 .09
1. 222
1. 222
8 .753
p-value
10
5
5
3
0.0003871
0.9427
0.9427
0.03276
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is less than .1. Model 5 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 36636.4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-332 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
BMDL =
9124.3
E.2.49.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model
45
40
35
30
25
20
0 20000
12:31 11/20 2009
40000 60000 80000 100000 120000 140000
dose
E.2.49.7. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Hill_CV_Unrest_BMRl_plasma_TT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Hill_CV_Unrest_BMRl_plasma_TT4.plt
Fri Nov 20 1273l7l0 2009
Tbl3, plasma TT4
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-333 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha
rho
intercept
44 . 7333
0
40. 9
-15. 4
3.33301
9622.52
Specified
Asymptotic Correlation Matrix of Parameter Estimates
alpha
intercept
The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
1
-6.6e-008
7 . 9e-008
-1. 5e-007
intercept
-6.6e-008
1
-0. 63
-0.12
v
7.9e-008
-0. 63
1
-0.29
k
-1. 5e-007
-0.12
-0.29
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
intercept
k
Estimate
44.6379
41. 2386
-11. 689
18
9336.14
Std. Err.
9.11167
1. 36525
2.1552
NA
666.631
Lower Conf. Limit
26.7793
38.5627
-15.9131
8029.56
Upper Conf. Limit
62.4964
43.9144
-7 .46484
10642.7
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
-0.143
0. 0683
0.0752
0.00058
1.71
-1.71
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
0 8 40.9 41.2 6.79 6.68
3969 8 41.4 41.2 5.37 6.68
6479 8 41.4 41.2 6.51 6.68
9968 8 32.3 32.3 7.35 6.68
4.7 61e + 004 8 33.6 29.5 6.22 6.68
1.37 8e + 005 8 25.5 29.5 7.64 6.68
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-334 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -112.012501 7 238.025002
A2 -111.401462 12 246.802924
A3 -112.012501 7 238.025002
fitted -115.165987 4 238.331973
R -127.445484 2 258.890968
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
32.088
1.22208
1.22208
6.30697
10
5
5
3
0.0003871
0.9427
0.9427
0.09759
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 94 8 6.87
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-335 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.49.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
45
40
35
30
25
20
BMD
BMDI
0
20000
40000
60000
80000
100000
120000
140000
dose
12:31 11/20 2009
E.2.49.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_CV_Unrest_BMRl_plasma_TT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Pwr_CV_Unrest_BMRl_plasma_TT4.plt
Fri Nov 20 12:31:11 2009
Tbl3, plasma TT4
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-336 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
alpha =
rho =
control =
slope =
power =
44 . 7333
0
41. 4
-1.4001
0.189211
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha control slope power
alpha 1 -5.9e-009 -5e-011 -3.8e-010
control -5.9e-009 1 -0.78 -0.75
slope -5e-011 -0.78 1 1
power -3.8e-010 -0.75 1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
46.4461
41. 4607
-0.0241896
0.547925
Std. Err.
9. 48077
2 .18095
0.0653588
0.223428
Lower Conf. Limit
27.8641
37.1861
-0.15229
0.110013
Upper Conf. Limit
65.028
45.7352
0.103911
0.985836
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8
3969 8
647 9 8
9968 8
4.761e+004
1.37 8e + 005
40. 9
41.
41.
32 . 3
33. 6
25.5
41. 5
39.2
38 . 5
37 . 7
32 .
25.
6.79
5.37
6.51
7 . 35
6.22
7 . 64
. 82
. 82
. 82
. 82
. 82
. 82
-0.233
0. 916
1. 21
-2 .24
0. 408
-0.0527
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-337 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
Model Log(likelihood) # Param's AIC
A1 -112.012501 7 238.025002
A2 -111.401462 12 246.802924
A3 -112.012501 7 238.025002
fitted -116.119011 4 240.238023
R -127.445484 2 258.890968
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2^log(Likelihood Ratio) Test df p-value
Test 1 32.088 10 0.0003871
Test 2 1.22208 5 0.9427
Test 3 1.22208 5 0.9427
Test 4 8.21302 3 0.04181
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 29589.5
BMDL = 582 6.38
E.2.50. White et al. (1986): CH50
E.2.50.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
5
0.09
19.19
0.00
389.66
1.1E+04
6.9E+03
nonconstant variance,
power restricted >1
exponential (M3)
5
0.09
19.19
0.00
389.66
1.1E+04
6.9E+03
nonconstant variance,
power restricted >1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-338 DRAFT—DO NOT CITE OR QUOTE
-------
exponential (M4)
4
0.09
18.15
0.00
390.63
7.8E+03
2.9E+03
nonconstant variance,
power restricted >1
exponential (M5)
4
0.09
18.15
0.00
390.63
7.8E+03
2.9E+03
nonconstant variance,
power restricted >1
Hillc
4
0.09
17.12
0.00
389.60
4.8E+03
8.3E+02
nonconstant variance,
n restricted >1, bound
hit
Hilld
3
0.09
7.05
0.07
381.53
8.2E+01
7.6E+01
nonconstant variance, n
unrestricted
linear
5
0.09
23.97
0.00
394.45
1.9E+04
1.4E+04
nonconstant variance
polynomial
5
0.09
23.97
0.00
394.45
1.9E+04
1.4E+04
nonconstant variance
power
5
0.09
23.97
0.00
394.45
1.9E+04
1.4E+04
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
5
0.09
19.89
0.00
388.58
9.6E+03
6.5E+03
constant variance,
power restricted >1
exponential (M3)
5
0.09
19.89
0.00
388.58
9.6E+03
6.5E+03
constant variance,
power restricted >1
exponential (M4)
4
0.09
18.80
0.00
389.48
6.5E+03
2.2E+03
constant variance,
power restricted >1
exponential (M5)
4
0.09
18.80
0.00
389.48
6.5E+03
2.2E+03
constant variance,
power restricted >1
Hill
4
0.09
17.39
0.00
388.07
3.3E+03
8.4E+02
constant variance, n
restricted >1, bound hit
Hill
3
0.09
7.07
0.07
379.75
1.5E+02
6.3E+00
constant variance, n
unrestricted
linear
5
0.09
24.48
0.00
393.16
1.8E+04
1.4E+04
constant variance
polynomial
5
0.09
24.48
0.00
393.16
1.8E+04
1.4E+04
constant variance
power
5
0.09
24.48
0.00
393.16
1.8E+04
1.4E+04
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meetBMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
1
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-339 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.2.50.2. Figure for Selected Model: Hill, Nonconstant Variance, n Restricted >1, Bound Hit
Hill Model with 0.95 Confidence Level
Hill
100
BMD
3MDI
0
10000
20000
30000
40000
50000
dose
13:28 11/16 2009
E.2.50.3. Output File for Selected Model: Hill, Nonconstant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_BMRl_CH50.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Hill_BMRl_CH50.plt
Mon Nov 16 13:28:23 2009
[insert study notes]
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter restricted to be greater than 1
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 7
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-340 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha
rho
intercept
5.60999
0
91
-74
0.118036
602.74
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha
rho
intercept
lalpha
1
-0. 99
0.27
0.23
-0.32
rho
-0. 99
1
-0.28
-0.24
0.33
intercept
0.27
-0.28
1
0.39
-0.78
v
0.23
-0.24
0.39
1
-0. 85
k
-0.32
0.33
-0.78
-0. 85
1
Parameter Estimates
Variable
lalpha
rho
intercept
k
Estimate
4 . 581
0.312931
74 . 6365
-66.2095
1
11475.6
Std. Err.
1.66271
0. 431612
6.33658
14.7868
NA
11747.8
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
1.32215
-0.533012
62 . 217
-95.1911
-11549.6
7.83986
1.15887
87.056
-37.2278
34500.8
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 91 74.6 14.1 19.4 2.39
602.7 8 54 71.3 8.49 19.3 -2.54
2250 8 63 63.8 11.3 18.9 -0.117
3934 8 56 57.7 25.5 18.6 -0.263
1.477 e+ 0 0 4 8 41 37.4 17 17.4 0.589
2.684e+004 8 32 28.3 17 16.7 0.636
4.99e+004 8 17 20.8 17 15.9 -0.678
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-341 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -181.340979 8 378.681959
A2 -175.820265 14 379.640529
A3 -181.238690 9 380.477380
fitted -189.800260 5 389.600520
R -212.367055 2 428.734109
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
73.0936
11.0414
10.8369
17.1231
12
<.0001
0.0871
0.05471
0.001829
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 4 7 56.06
BMDL = 825.553
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-342 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.2.50.4. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
WIDLBMD
10000
20000
30000
40000
50000
E.2.50.5. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_CH50.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_CH50.plt
Mon Nov 16 13:28:23 2009
[insert study notes]
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 7
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 5.60999
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-343 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho = 0
intercept = 91
v = -74
n = 0.118036
k = 602.74
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -1 0.16 0.19 -0.4 -0.013
rho -1 1 -0.16 -0.19 0.4 0.011
intercept 0.16 -0.16 1 0.15 -0.58 0.015
v 0.19 -0.19 0.15 1 -0.011 -0.93
n -0.4 0.4 -0.58 -0.011 1 -0.36
k -0.013 0.011 0.015 -0.93 -0.36 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
6.54258
-0.246247
89.6313
-615.173
0.246754
2.44083e+008
Std. Err.
2.08981
0.541898
5.59369
706.037
0.0587686
1.35075e+009
Lower Conf. Limit
2.44663
-1.30835
78.6679
-1998.98
0.13157
-2.40334e+009
Upper Conf. Limit
10.6385
0.815854
100.595
768.633
0.361938
2.89151e+009
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 91 89.6 14.1 15.1 0.256
602.7 8 54 65.2 8.49 15.8 -2.01
2250 8 63 56.3 11.3 16 1.17
3934 8 56 51.7 25.5 16.2 0.747
1.477e+004 8 41 38.3 17 16.8 0.453
2.684e+004 8 32 30.9 17 17.3 0.175
4.99e+004 8 17 22.3 17 18 -0.833
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) ) = Sigma'"2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-344 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -181.340979 8 378.681959
A2 -175.820265 14 379.640529
A3 -181.238690 9 380.477380
fitted -184.762700 6 381.525401
R -212.367055 2 428.734109
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 73.0936 12 <.0001
Test 2 11.0414 6 0.0871
Test 3 10.8369 5 0.05471
Test 4 7.04802 3 0.07038
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 81.6596
BMDL = 7 5.5039
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-345 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3. ADMINISTERED DOSE BMDS RESULTS
2 E.3.1. Amin et al. (2000): Saccharin Consumed, Female (0.25%)
3 E.3.1.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linearc
1
0.00
0.84
0.36
179.70
8.8E+01
5.9E+01
nonconstant variance
polynomial
1
0.00
0.84
0.36
179.70
8.8E+01
5.9E+01
nonconstant variance
power
1
0.00
0.84
0.36
179.70
8.8E+01
5.9E+01
nonconstant variance,
power restricted >1, bound
hit
linear
1
0.00
0.12
0.73
191.80
6.6E+01
4.3E+01
constant variance
polynomial
1
0.00
0.12
0.73
191.80
6.6E+01
4.3E+01
constant variance
power
1
0.00
0.12
0.73
191.80
6.6E+01
4.3E+01
constant variance, power
restricted >1, bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-346 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.1.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
Linear
BMDL
13:41 11/11 2009
E.3.1.3. Output file for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Linear_BMRl_25_s_c.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Linear_BMRl_25_s_c.plt
Wed Nov 11 13:41:21 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = beta 0 + beta l*dose + beta 2*dose^2 +
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-347 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
lalpha
rho
beta_0
beta 1
5.29482
0
30.8266
-0.204134
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
beta_0
beta 1
lalpha
1
-0. 99
-0.016
0. 03
rho
-0. 99
1
0. 013
-0.026
beta_0
-0.016
0. 013
1
-0. 94
beta_l
0. 03
-0.026
-0. 94
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
-2.55843
2 .42056
30.3968
-0.196699
Std. Err.
1.66185
0.545617
4 . 03582
0.0443352
Lower Conf. Limit
-5.8156
1.35117
22.4868
-0.283594
Upper Conf. Limit
0.698746
3. 48995
38 . 3069
-0.109803
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
25 10
100 10
31. 7
24 . 6
10.7
30. 4
25.5
10.7
20.6
12
5.33
17 . 3
14
4 . 92
0.233
-0.2
-0.0204
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma^2
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -92.841935 4 193.683870
A2 -85.255316 6 182.510632
A3 -85.429148 5 180.858295
fitted -85.851107 4 179.702213
R -98.136607 2 200.273213
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-348 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
25.7626
15.1732
0.347663
0. 843918
<.0001
0.0005072
0.5554
0.3583
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 88 . 1623
BMDL = 58.9029
E.3.2. Amin et al. (2000): Saccharin Consumed, Female (0.50%)
E.3.2.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linearc
1
<0001
4.68
0.03
159.74
9.9E+01
6.4E+01
nonconstant
variance
polynomial
1
<.0001
4.68
0.03
159.74
9.9E+01
6.4E+01
nonconstant variance
power
1
<.0001
4.68
0.03
159.74
9.9E+01
6.4E+01
nonconstant variance,
power restricted >1,
bound hit
linear
1
<.0001
2.57
0.11
175.96
6.5E+01
4.3E+01
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-349 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
polynomial
1
<.0001
2.57
0.11
175.96
6.5E+01
4.3E+01
constant variance
power
1
<.0001
2.57
0.11
175.96
6.5E+01
4.3E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
E.3.2.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
35
30
25
20
15
10
Linear
BMDL
BMO
20
40 60
dose
80
100
13:41 11/11 2009
E.3.2.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Linear_BMRl_50_s_c.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Linear_BMRl_50_s_c.plt
Wed Nov 11 13:41:42 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = beta 0 + beta l*dose + beta 2*dose^2 +
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-350 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.68512
rho = 0
beta_G = 19.3484
beta 1 = -0.158141
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho beta 0 beta 1
lalpha 1 -0.97 0.018 -0.0021
rho -0.97 1 -0.027 0.014
beta_0 0.018 -0.027 1 -0.95
beta 1 -0.0021 0.014 -0.95 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
-0.997428
2 .13634
18 .1144
-0.135736
Std. Err.
0.992786
0. 404 989
3.10302
0.0331501
Lower Conf. Limit
-2.94325
1.34257
12.0326
-0.200709
Upper Conf. Limit
0. 948397
2.9301
24.1962
-0.0707631
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10 22.4 18.1 16 13.4 1
25 10 11.4 14.7 7.66 10.7 -0.983
100 10 4.54 4.54 3.33 3.06 -0.00393
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-351 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Model R: Yi = Mu + e(i)
Var{e(i) ) = Sigma'"2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -83.696404 4 175.392808
A2 -73.511830 6 159.023660
A3 -73.530233 5 157.060467
fitted -75.868688 4 159.737377
R -90.294746 2 184.589492
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2^log(Likelihood Ratio) Test df p-value
Test 1 33.5658 4 <.0001
Test 2 20.3691 2 <.0001
Test 3 0.0368066 1 0.8479
Test 4 4.67691 1 0.03057
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 98.7409
BMDL = 64.169
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-352 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.3. Amin et al. (2000): Saccharin Preference Ratio, Female (0.25%)
2 E.3.3.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linearc
1
0.01
9.80
0.00
228.09
1.3E+02
6.1E+01
nonconstant variance
polynomial
1
0.01
9.80
0.00
228.09
1.3E+02
6.1E+01
nonconstant variance
power
1
0.01
9.80
0.00
228.09
1.3E+02
6.1E+01
nonconstant variance,
power restricted >1,
bound hit
linear
1
0.01
3.36
0.07
226.51
1.1E+02
6.1E+01
constant variance
polynomial
1
0.01
3.36
0.07
226.51
1.1E+02
6.1E+01
constant variance
power
1
0.01
3.36
0.07
226.51
1.1E+02
6.1E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
5 E.3.3.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
Linear
90
80
70
60
50
40
BMC I
BMDL
30
0
20
40
60
80
100
120
dose
6 13:42 11/11 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-353 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.3.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Linear_BMRl_2 5_s_p_f.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Linear_BMRl_25_s_p_f.plt
Wed Nov 11 13:42:05 2009
Rel Male Thymus wt Tbl 2
The form of the response function is:
Y[dose] = beta 0 + beta l^dose + beta 2*dose/s2 +
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 6.34368
rho = 0
beta_Q = 74.2008
beta 1 = -0.219781
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho beta 0 beta 1
lalpha 1 -1 0.2 -0.28
rho -1 1 -0.19 0.28
beta_0 0.2 -0.19 1 -0.76
beta 1 -0.28 0.28 -0.76 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha 0.338774 9.23768 -17.7667 18.4443
rho 1.43998 2.21674 -2.90476 5.78472
beta_0 73.6633 6.6623 60.6054 86.7211
beta 1 -0.207175 0.101074 -0.405276 -0.00907442
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10 82.1 73.7 13.3 26.2 1.02
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-354 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
25
100
10
10
58 .1
54 . 9
68 . 5
52 . 9
33. 9
19.5
24 . S
20. (
-1. 32
0.295
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma'"2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) ) = Sigma'"2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -108.574798 4 225.149597
A2 -104.269377 6 220.538754
A3 -105.147952 5 220.295903
fitted -110.046917 4 228.093834
R -112.382522 2 228.765045
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
16.2263
8.61084
1.75715
9.7 97 93
0.00273
0.0135
0.185
0.001747
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-355 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Confidence level =
BMD =
0. 95
126.365
BMDL = 61.2812
E.3.4. Amin et al. (2000): Saccharin Preference Ratio, Female (0.50%)
E.3.4.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linearc
1
0.56
2.60
0.11
236.57
9.2E+01
5.2E+01
nonconstant variance
polynomial
1
0.56
2.60
0.11
236.57
9.2E+01
5.2E+01
nonconstant variance
power
1
0.56
2.60
0.11
236.57
9.2E+01
5.2E+01
nonconstant variance,
power restricted >1,
bound hit
linear
1
0.56
2.92
0.09
234.94
8.3E+01
5.1E+01
constant variance
polynomial
1
0.56
2.92
0.09
234.94
8.3E+01
5.1E+01
constant variance
power
1
0.56
2.92
0.09
234.94
8.3E+01
5.1E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-356 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.4.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
Linear
90
80
70
60
50
40
30
20
BMDL
BMD
10
0
20
40
60
80
100
dose
13:42 11/11 2009
E.3.4.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\Linear_BMRl_50_s_p_f.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Linear_BMRl_50_s_p_f.plt
Wed Nov 11 13:42:27 2009
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = beta 0 + beta l*dose + beta 2*dose/s2 +
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-357 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
lalpha
rho
beta_0
beta 1
6. 63936
0
64.1858
-0.332668
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
beta_0
beta 1
lalpha
1
-1
0.11
-0.18
rho
-1
1
-0.11
0.18
beta_0
0.11
-0.11
1
-0.75
beta_l
-0.18
0.18
-0.75
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
4 .43902
0.562378
63.7204
-0.320869
Std. Err.
3.53662
0. 909867
7 .4 9597
0.114882
Lower Conf. Limit
-2.49263
-1. 22093
49.0286
-0.546034
Upper Conf. Limit
11.3707
2 . 34569
78.4122
-0.0957048
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
25 10
100 10
12.1
44.5
33. 8
63.7
55.7
31. 6
24 . 6
32 . 9
24 . 6
29.6
28 . 5
24 . 3
0. 962
-1.24
0 . 277
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma^2
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -113.009921 4 234.019841
A2 -112.428886 6 236.857773
A3 -112.984528 5 235.969055
fitted -114.283840 4 236.567679
R -117.976057 2 239.952114
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-358 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
Explanation of Tests
Test 1:
Test 2
Test 3
Test 4
Do responses and/or variances differ among Dose levels:
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
Test 1
Test 2
Test 3
Test 4
-2*log(Likelihood Ratio) Test df
11.0943
1.16207
1.11128
2 .59862
p-value
0.02552
0.5593
0.2918
0.107
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1.
homogeneous model
The p-value for Test 3 is greater than .1.
to be appropriate here
The p-value for Test 4 is greater than .1.
to adequately describe the data
Consider running a
The modeled variance appears
The model chosen seems
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 92.2435
BMDL = 51.5208
E.3.5. Bell et al. (2007): Balano-Preputial Separation in Male Pups (10% Extra Risk)
E.3.5.1. Summary Table of BMDS modeling results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
1 P"
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
2
1.99
0.37
113.51
7.3E+00
4.7E+00
power restricted >1, bound
hit
logistic
2
2.88
0.24
114.85
1.5E+01
1.1E+01
log-logistic b
2
1.57
0.46
112.95
5.2E+00
2.9E+00
slope restricted >1,
bound hit
log-logisticc
1
0.49
0.48
113.91
2.1E+00
1.4E-01
slope unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-359 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
log-probit
1
0.60
0.44
114.02
2.2E+00
1.7E-01
slope restricted >1
multistage, 1-
degree
2
1.99
0.37
113.51
7.3E+00
4.7E+00
betas restricted >0, bound
hit
probit
2
2.79
0.25
114.72
1.4E+01
1.1E+01
Weibull
2
1.99
0.37
113.51
7.3E+00
4.7E+00
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
E.3.5.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
10 20 30 40
dose
11:43 11/29 2009
—i—i i—i—i—i—i
Log-Logistic
BMDL BMD
i I I ¦ ¦ I- ¦ iL-1 I I I I—I I I I I I I I I I I I I I I I I I I I I I I I I I I I I L
E.3.5.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\LogLogistic_BMR2_BPS_d49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\LogLogistic_BMR2_BPS_d4 9.plt
Sun Nov 29 11:43:52 2009
0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-360 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0.0333333
intercept = -3.75371
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.58
intercept -0.58 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0635251 * * *
intercept -3.84765 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -54.476 2 1.53661 2 0.4638
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 112.952
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0635 1.906 1.000 30 -0.678
2.4000 0.1091 3.274 5.000 30 1.011
8.0000 0.2000 6.001 6.000 30 -0.000
46.0000 0.5273 15.819 15.000 30 -0.300
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-361 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Chi' 2 = 1.57 d.f. = 2
E'-value =
0.4 55 9
E'.enchmark Dose
Specified effect :
Risk Type
Confidence level :
BMD
BMDL
CC'irp'UtatiC'ri
0 .1
E::tra risk
0.95
5.20918
2.86991
E.3.5.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
E.3.5.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\LogLogistic_Unrest_BMR2_BPS_d49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\LogLogistic_Unrest_BMR2_BPS_d4 9.plt
Sun Nov 29 11:43:54 2009
0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-362 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0.0333333
intercept = -2.54947
slope = 0.615936
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.49 0.35
intercept -0.49 1 -0.93
slope 0.35 -0.93 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0354714 * * *
intercept -2.70296 * * *
slope 0.670238 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -53.9541 3 0.492844 1 0.4827
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 113.908
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0355 1.064 1.000 30 -0.063
2.4000 0.1392 4.176 5.000 30 0.435
8.0000 0.2405 7.216 6.000 30 -0.520
46.0000 0.4848 14.544 15.000 30 0.167
Chi'A2 = 0.49 d.f. = 1 P-value = 0.4836
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-363 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Benchmark Dose
Specified effect :
Risk Type
Confidence level :
BMD
BMDL
Computation
0.1
Extra risk
0. 95
2.12667
0.13633
E.3.6. Bell et al. (2007): Balano-Preputial Separation in Male Pups (5% Extra Risk)
E.3.6.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
1 P"
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
2
1.99
0.37
113.51
3.6E+00
2.3E+00
power restricted >1, bound
hit
logistic
2
2.88
0.24
114.85
8.4E+00
6.2E+00
log-logistic b
2
1.57
0.46
112.95
2.5E+00
1.4E+00
slope restricted >1,
bound hit
log-logisticc
1
0.49
0.48
113.91
7.0E-01
1.1E-02
slope unrestricted
log-probit
1
0.60
0.44
114.02
8.6E-01
2.1E-02
slope restricted >1
multistage, 1-
degree
2
1.99
0.37
113.51
3.6E+00
2.3E+00
betas restricted >0, bound
hit
probit
2
2.79
0.25
114.72
7.7E+00
5.7E+00
Weibull
2
1.99
0.37
113.51
3.6E+00
2.3E+00
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-364 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.3.6.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
0.7
0.6
0.5
0.4
0.3
0.2
0.1
BMDL BMD
b ¦ I -I ¦ I ¦ ¦ i_
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
10
20
30
40
dose
11:43 11/29 2009
E.3.6.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\LogLogistic_BMRl_BPS_d49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\LogLogistic_BMRl_BPS_d4 9.plt
Sun Nov 29 11:43:49 2009
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-365 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Default Initial Parameter Values
background = 0.0333333
intercept = -3.75371
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.58
intercept -0.58 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0635251 * * *
intercept -3.84765 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -54.476 2 1.53661 2 0.4638
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 112.952
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0635
1. 906
1. 000
30
-0.678
2.4000
0.1091
3.274
5. 000
30
1. 011
8.0000
0.2000
6. 001
6. 000
30
-0.000
46.0000
0.5273
15.819
15.000
30
-0.300
Chi'A2 = 1.57 d.f. = 2 P-value = 0.4559
Benchmark Dose Computation
Specified effect = 0.05
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2.46751
BMDL = 1.35943
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-366 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.3.6.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
0.7
0.6
0.5
0.4
0.3
0.2
0
10
20
30
40
dose
11:43 11/29 2009
E.3.6.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\LogLogistic_Unrest_BMRl_BPS_d49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\LogLogistic_Unrest_BMRl_BPS_d4 9.plt
Sun Nov 29 11:43:53 2009
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-367 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
Default Initial Parameter Values
background = 0.0333333
intercept = -2.54947
slope = 0.615936
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.49 0.35
intercept -0.49 1 -0.93
slope 0.35 -0.93 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0354714 * * *
intercept -2.70296 * * *
slope 0.670238 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -53.9541 3 0.492844 1 0.4827
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 113.908
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0355
1. 064
1. 000
30
-0.063
2.4000
0.1392
4.176
5. 000
30
0. 435
8.0000
0.2405
7 . 216
6. 000
30
-0.520
46.0000
0.4848
14.544
15.000
30
0.167
Chi/N2 = 0.49 d.f. = 1 P-value = 0.4836
Benchmark Dose Computation
Specified effect = 0.05
Risk Type = Extra risk
Confidence level = 0.95
BMD = 0.697474
BMDL = 0.0111259
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-368 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.7. Cantoni et al. (1981): Urinary Copro-Porhyrins
2 E.3.7.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
logistic
2
20.02
0.00
131.96
4.6E+03
3.8E+01
log-logistic
2
20.02
0.00
131.96
8.0E+07
3.8E+01
slope restricted >1, bound
hit
log-probit
2
20.02
0.00
131.96
4.6E+03
error
slope restricted >1
multistage, 2-
degree
3
20.02
0.00
129.96
error
error
betas restricted >0, bound
hit
probitb
2
20.02
0.00
131.96
4.6E+03
3.8E+01
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
3
4
5 E.3.7.2. Figure for Selected Model: Probit
Probit Model with 0.95 Confidence Level
Probit
13:34 11/11 2009
1000
2000
3000
dose
4000
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-369 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.7.3. Output File for Selected Model: Probit
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Probit_BMR2_BPS_pnd49.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Probit_BMR2_BPS_pnd4 9.plt
Wed Nov 11 13:34:24 2009
The form of the probability function is:
P[response] = CumNorm(Intercept+Slope* Dose),
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial (and Specified) Parameter Values
background = 0 Specified
intercept = 1.29116
slope = -0.0292594
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.64
slope -0.64 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
intercept 0.755415 0.164576 0.432851 1.07798
slope 0 0.00651162 -0.0127625 0.0127625
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -53.7077 4
Fitted model -63.9797 2 20.544 2 3.4588333e-005
Reduced model -63.9797 1 20.544 3 0.0001309
AIC: 131.959
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-370 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Dose
Est. Prob.
Goodness of Fit
Expected Observed Size
Scaled
Residual
0.0000 0.7750 23.250 29.000 30 2.514
2.4000 0.7750 23.250 25.000 30 0.765
8.0000 0.7750 23.250 24.000 30 0.328
46.0000 0.7750 23.250 15.000 30 -3.607
Chi'"'2 = 20.02 d.f. = 2 P-value = 0.0000
Slope parameter essentially zero. BMD set to 100 * max(Dose).
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 4 600
BMDL = 38.23 94
E.3.8. Cantoni et al. (1981): Urinary Porphyrins
E.3.8.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
<0.0001
19.41
<0.0001
58.75
1.2E+01
9.0E+00
nonconstant variance,
power restricted >1
exponential (M3)
2
<0.0001
19.41
<0.0001
58.75
1.2E+01
9.0E+00
nonconstant variance,
power restricted >1
exponential (M4)
1
<0.0001
21.80
<0.0001
63.14
2.2E-01
1.1E-01
nonconstant variance,
power restricted >1
exponential (M5)
1
<0.0001
21.80
<0.0001
63.14
2.2E-01
1.1E-01
nonconstant variance,
power restricted >1
Hill
0
<.0001
19.02
NA
62.36
9.4E+00
4.7E+00
nonconstant variance,
n restricted >1
linearc
2
<0001
23.15
<.0001
62.49
7.7E-01
2.8E-01
nonconstant
variance
polynomial
1
<.0001
18.19
<.0001
59.53
6.3E+00
2.0E+00
nonconstant variance
power
2
<.0001
23.15
<.0001
62.49
7.7E-01
2.8E-01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
<0.0001
0.05
0.98
108.89
7.0E+01
5.0E+01
constant variance,
power restricted >1
exponential (M3)
2
<0.0001
0.05
0.98
108.89
7.0E+01
5.0E+01
constant variance,
power restricted >1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-371 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M4)
1
<0.0001
0.86
0.35
111.70
1.8E+01
1.0E+01
constant variance,
power restricted >1
exponential (M5)
0
<0.0001
0.04
N/A
112.88
4.2E+01
1.4E+01
constant variance,
power restricted >1
Hill
0
<.0001
0.04
NA
112.88
4.2E+01
1.0E+01
constant variance, n
restricted >1
linear
2
<.0001
0.85
0.66
109.69
1.8E+01
1.3E+01
constant variance
polynomial
1
<.0001
0.03
0.86
110.88
4.4E+01
1.1E+01
constant variance
power
1
<.0001
0.04
0.85
110.88
4.2E+01
1.4E+01
constant variance,
power restricted >1
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
1
2
3 E.3.8.2. Figure for Selected Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
Linear
V1DLBMD
14:06 10/06 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-372 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.8.3. Output File for Selected Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\Linear_BMRl_Urinary_porphyrins.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\Linear_BMRl_Urinary_porphyrins.plt
Tue Oct 06 14:06:04 2009
Table 1, dose converted to ng per kg per day
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose/s2 + ...
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 6.68244
rho = 0
beta_Q = -1.7736
beta 1 = 1.38238
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho beta 0 beta 1
lalpha 1 -0.89 -0.27 0.2
rho -0.89 1 0.18 -0.082
beta_0 -0.27 0.18 1 -0.31
beta 1 0.2 -0.082 -0.31 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha -2.51366 0.863504 -4.2061 -0.821228
rho 2.27539 0.314031 1.6599 2.89088
beta_0 2.57041 0.376521 1.83244 3.30838
beta 1 1.07729 0.234062 0.618541 1.53605
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 4 2.27 2.57 0.49 0.833 -0.721
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-373 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
1.43 4 5.55
14.3 3 7 . 62
143 3 197
4.11 0.85
18 1.79
157 63.1
1.42 2.03
7.61 -2.36
89.4 0.78
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma(i)~2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -51.421748 5 112.843496
A2 -15.312111 8 46.624223
A3 -15.669627 6 43.339255
fitted -27.243469 4 62.486938
R -68.750584 2 141.501167
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
106.877
12.2193
0.715032
23.1477
<.0001
<.0001
0.6 9 9 4
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-374 DRAFT—DO NOT CITE OR QUOTE
-------
1 Risk Type
2 Confidence level
3 BMD
4 BMDL
5
6
7
8 E.3.9. Crofton et al. (2005): Serum T4
9 E.3.9.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
8
0.76
46.09
<0.0001
518.24
2.1E+03
1.2E+03
constant variance,
power restricted >1
exponential (M3)
8
0.76
46.09
<0.0001
518.24
2.1E+03
1.2E+03
constant variance,
power restricted >1
exponential (M4)
C
7
0.76
2.05
0.96
476.20
5.6E+01
3.0E+01
constant variance,
power restricted >1
exponential (M5)
7
0.76
2.05
0.96
476.20
5.6E+01
3.0E+01
constant variance,
power restricted >1
Hill
6
0.76
1.28
0.97
477.43
5.6E+01
2.6E+01
constant variance, n
restricted >1
linear
8
0.76
51.36
<.0001
523.52
4.2E+03
3.1E+03
constant variance
polynomial
8
0.76
51.36
<.0001
523.52
4.2E+03
3.1E+03
constant variance
power
8
0.76
51.36
<.0001
523.52
4.2E+03
3.1E+03
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
10
11
Estimated standard deviations from the control mean
0. 95
0.773193
0.281589
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-375 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.9.2. Figure for Selected Model: Exponential (M4), Constant Variance, Power Restricted
>1
Exponential_beta Model 4 with 0.95 Confidence Level
; ;
Exponential
- J
I
N
"
" i
VIDL
BMD
-
0 2000 4000 6000 8000 10000
dose
12:00 11/11 2009
E.3.9.3. Output File for selected Model: Exponential (M4), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\ExpConstVar_BMRl_SerumT4.(d)
Gnuplot Plotting File:
Wed Nov 11 12:00:47 2009
0
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-376 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 10
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 4
lnalpha
rho(S)
5.47437
0
104.999
0.000371694
0.445764
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 4
5.50283
0
99.776
0.00728387
0.533516
1
Table of Stats From Input Data
Obs Mean
Obs Std Dev
0
14
100
15. 44
0.1
6
96.27
14 . 98
3
12
98 . 57
18 .11
10
6
99.76
19. 04
30
6
93.32
12 .11
100
6
70. 94
12.74
300
6
62 . 52
14.75
1000
6
52 . 68
22 .73
3000
6
54 . 66
19.71
le+004
4
49.15
11.15
Estimated Values of Interest
Est Mean
Est Std
Scaled Residual
0
99.78
15. 66
0.05325
0.1
99.74
15. 66
-0.5434
3
98 .77
15. 66
-0. 04357
10
96.51
15. 66
0.5085
30
90. 64
15. 66
0.4195
100
75 . 7
15. 66
-0.744
300
58 .47
15. 66
0.6334
1000
53.26
15. 66
-0.09133
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-377 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
3000
le+004
53.23
53.23
15. 66
15. 66
0.2237
-0.5218
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma"2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + log(mean(i)) ^ rho)
Model R: Yij = Mu + e(1)
Var{e(ij)} = Sigma"2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -233.0774 11 488.1549
A2 -230.2028 20 500.4056
A3 -233.0774 11 488.1549
R -268.4038 2 540.8076
4 -234.1019 4 476.2038
Additive constant for all log-likelihoods = -66.16. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
76.4
5.749
5.749
2.049
D. F.
18
9
9
7
p-value
< 0.0001
0.7647
0.7647
0.9571
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-378 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Specified Effect = 1.000000
3 Risk Type = Estimated standard deviations from control
4 Confidence Level = 0.950000
5 BMD = 56.3321
6 BMDL = 30.0635
7
8 E.3.10. DeCaprio et al. (1986): Absolute Kidney Weight, Males
9 E.3.10.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.67
10.85
0.00
-9.87
2.6E+00
1.7E+00
nonconstant variance,
power restricted >1
exponential (M3)
2
0.67
10.85
0.00
-9.87
2.6E+00
1.7E+00
nonconstant variance,
power restricted >1
exponential (M4)
1
0.67
0.54
0.46
-18.18
2.7E-01
1.2E-01
nonconstant variance,
power restricted >1
exponential (M5)
1
0.67
0.54
0.46
-18.18
2.7E-01
1.2E-01
nonconstant variance,
power restricted >1
Hill
1
0.67
0.08
0.78
-18.64
1.9E-01
7.3E-02
nonconstant variance, n
restricted >1, bound hit
linear
2
0.67
11.14
0.00
-9.58
2.8E+00
1.9E+00
nonconstant variance
polynomial
1
0.67
1.10
0.29
-17.62
3.6E-01
2.3E-01
nonconstant variance
power
2
0.67
11.14
0.00
-9.58
2.8E+00
1.9E+00
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.67
10.90
0.00
-11.71
2.5E+00
1.7E+00
constant variance,
power restricted >1
exponential (M3)
2
0.67
10.90
0.00
-11.71
2.5E+00
1.7E+00
constant variance,
power restricted >1
exponential (M4)
1
0.67
0.52
0.47
-20.10
2.6E-01
1.2E-01
constant variance,
power restricted >1
exponential (M5)
1
0.67
0.52
0.47
-20.10
2.6E-01
1.2E-01
constant variance,
power restricted >1
Hillc
1
0.67
0.07
0.79
-20.54
1.9E-01
7.1E-02
constant variance, n
restricted >1, bound
hit
linear
2
0.67
11.21
0.00
-11.40
2.6E+00
1.9E+00
constant variance
polynomial
1
0.67
1.08
0.30
-19.53
3.5E-01
2.4E-01
constant variance
power
2
0.67
11.21
0.00
-11.40
2.6E+00
1.9E+00
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-379 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.10.2. Figure for Selected Model: Hill, Constant Variance, n Restricted >1, Bound Hit
Hill Model with 0.95 Confidence Level
ISMDL BMD
14:07 10/06 2009
E.3.10.3. Output File for Selected Model: Hill, Constant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_abs_male_kidney_wt.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_abs_male_kidney_wt.pit
Tue Oct 06 14:07:06 2009
Abs Male Kidney wt, Tbl 2
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-380 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha
rho
intercept
0.202865
0
5.49
-1.19
1.12255
0.399186
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
-1. 7e-009
4 . 6e-009
-2e-009
intercept
-1.7e-009
1
-0.49
-0.54
v
4 . 6e-009
-0.49
1
-0.27
k
-2e-009
-0.54
-0.27
1
Parameter Estimates
Variable
alpha
intercept
k
Estimate
0.183391
5. 47882
-1. 25656
1
0.361009
Std. Err.
0.0405044
0.130251
0.197258
NA
0.220645
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
0.104004
5.22353
-1.64318
-0.0714485
0.262778
5.7341
-0.869946
0.793466
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
0.12 10
0.61 11
4.9 10
5.49
5.14
4 .71
4 . 3
5.48
5.17
4 .69
4 . 31
0.538
0.379
0.398
0.474
0.428
0.428
0.428
0.428
0.0826
-0.187
0.159
-0.0626
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-381 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Model R: Yi = Mu + e(i
Var{e(i) ) = Sigma'"2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 14.306322 5 -18.612644
A2 15.093716 8 -14.187432
A3 14.306322 5 -18.612644
fitted 14.270730 4 -20.541459
R -0.636909 2 5.273819
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
31.4613
1.57479
1.57479
0.0711848
0001
6651
6651
7896
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.186641
BMDL = 0.07147 93
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-382 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.11. DeCaprio et al. (1986): Absolute Thymus Weight, Males
2 E.3.11.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
<0.0001
0.74
0.69
-93.74
1.8E+00
4.8E-01
nonconstant variance,
power restricted >1
exponential (M3)
1
<0.0001
0.70
0.40
-91.78
2.4E+00
4.8E-01
nonconstant variance,
power restricted >1
exponential (M4)
2
<0.0001
0.74
0.69
-93.74
1.8E+00
4.4E-01
nonconstant variance,
power restricted >1
exponential (M5)
0
<0.0001
0.70
N/A
-89.78
2.4E+00
4.8E-01
nonconstant variance,
power restricted >1
Hill
0
<.0001
0.70
NA
-89.78
8.4E-01
error
nonconstant variance, n
restricted >1
linear
2
<.0001
0.69
0.71
-93.79
2.4E+00
error
nonconstant variance
polynomial
1
<.0001
0.69
0.41
-91.79
2.4E+00
error
nonconstant variance
powerc
2
<0001
0.69
0.71
-93.79
2.4E+00
1.0E+00
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
<0.0001
1.01
0.60
-60.51
1.1E+01
3.7E+00
constant variance,
power restricted >1
exponential (M3)
2
<0.0001
1.01
0.60
-60.51
1.1E+01
3.7E+00
constant variance,
power restricted >1
exponential (M4)
1
<0.0001
0.60
0.44
-58.92
error
error
constant variance,
power restricted >1
exponential (M5)
1
<0.0001
0.60
0.44
-58.92
error
error
constant variance,
power restricted >1
Hill
1
<.0001
0.40
0.53
-59.12
error
error
constant variance, n
restricted >1, bound hit
linear
2
<.0001
1.04
0.59
-60.48
9.0E+00
4.2E+00
constant variance
polynomial
1
<.0001
0.67
0.41
-58.85
error
5.8E+00
constant variance
power
2
<.0001
1.04
0.59
-60.48
9.0E+00
4.2E+00
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-383 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.11.2. Figure for Selected Model: Power, Nonconstant Variance, Power Restricted >1,
Bound Hit
Power Model with 0.95 Confidence Level
0.8
Power
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
BMDI
!MD
-0.1
0
1
2
3
4
5
dose
14:07 10/06 2009
E.3.11.3. Output File for Selected Model: Power, Nonconstant Variance, Power Restricted >1,
Bound Hit
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\Pwr_BMRl_abs_thymus_wt.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\Pwr_BMRl_abs_thymus_wt.plt
Tue Oct 06 14:07:51 2009
Abs Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is restricted to be greater than or egual to 1
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-384 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha
rho
control
slope
power
-2.54423
0
0.562
-0.216619
-0.019526
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -power
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha rho control slope
lalpha 1 0.99 0.39 -0.97
rho 0.99 1 0.33 -0.98
control 0.39 0.33 1 -0.4
slope -0.97 -0.98 -0.4 1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
-7 .29904
-4 . 37824
0.500643
-0.0492183
1
Std. Err.
2 . 85129
3. 97762
0.0242054
0.0335618
NA
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-12 .8875
-12 .1742
0.453202
-0.114998
-1.71062
3. 41775
0.548085
0.0165617
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10 0.562 0.501 0.158 0.118 1.64
0.12 10 0.448 0.495 0.0696 0.121 -1.22
0.61 11 0.445 0.471 0.113 0.135 -0.628
4.9 10 0.352 0.259 0.528 0.498 0.587
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-385 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 33.761192 5 -57.522384
A2 57.094821 8 -98.189643
A3 51.239109 6 -90.478218
fitted 50.896393 4 -93.792785
R 32.253943 2 -60.507885
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 49.6818 6 <.0001
Test 2 46.6673 3 <.0001
Test 3 11.7114 2 0.002863
Test 4 0.685433 2 0.7098
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 2.40256
BMDL = 1.007 4 6
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-386 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.12. DeCaprio et al. (1986): Body Weight, Females
2 E.3.12.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
0.10
1.49
0.68
380.09
5.5E+00
3.6E+00
nonconstant variance,
power restricted >1
exponential (M3)
3
0.10
1.49
0.68
380.09
5.5E+00
3.6E+00
nonconstant variance,
power restricted >1
exponential (M4)
2
0.10
1.27
0.53
381.87
4.5E+00
2.4E+00
nonconstant variance,
power restricted >1
exponential (M5)
2
0.10
1.27
0.53
381.87
4.5E+00
2.4E+00
nonconstant variance,
power restricted >1
Hill
2
0.10
1.26
0.53
381.86
4.5E+00
error
nonconstant variance, n
restricted >1, bound hit
linear
3
0.10
2.06
0.56
380.66
6.9E+00
4.7E+00
nonconstant variance
polynomial
2
0.10
1.28
0.53
381.88
4.6E+00
2.6E+00
nonconstant variance
power
3
0.10
2.06
0.56
380.66
6.9E+00
4.7E+00
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
C
3
0.10
1.11
0.78
379.55
6.1E+00
4.5E+00
constant variance,
power restricted >1
exponential (M3)
3
0.10
1.11
0.78
379.55
6.1E+00
4.5E+00
constant variance,
power restricted >1
exponential (M4)
2
0.10
0.75
0.69
381.19
4.7E+00
2.6E+00
constant variance,
power restricted >1
exponential (M5)
2
0.10
0.75
0.69
381.19
4.7E+00
2.6E+00
constant variance,
power restricted >1
Hill
2
0.10
0.74
0.69
381.18
4.7E+00
2.4E+00
constant variance, n
restricted >1, bound hit
linear
3
0.10
1.73
0.63
380.17
7.6E+00
5.9E+00
constant variance
polynomial
2
0.10
0.76
0.68
381.20
4.7E+00
2.7E+00
constant variance
power
3
0.10
1.73
0.63
380.17
7.6E+00
5.9E+00
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-387 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.12.2. Figure for Selected Model: Exponential (M2), Constant Variance, Power Restricted
>1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
600
500
400
300
200
BMDL
BMD
0
5
10
15
20
25
30
dose
14:08 10/06 2009
E.3.12.3. Output File for Selected Model: Exponential (M2), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\ExpConstVar_BMRl_fem_BW.(d)
Gnuplot Plotting File:
Tue Oct 06 14:08:53 2009
Female BW Tbl 1
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-388 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
Specified
lnalpha 8.08396
rho(S) 0
a 632.1
b 0.0928666
c 0.528849
d 1
Parameter Estimates
Variable Model 2
lnalpha 8.10223
rho 0
a 588.488
b 0.0403005
c 0.434663
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 602 33.94
0.12 10 583 69.57
0.68 9 570 66
4.86 10 531 44.27
31 4 351 98
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 584.9 57.71 0.839
0.12 583.7 57.71 -0.0379
0.68 578.2 57.71 -0.4248
4.86 538.6 57.71 -0.4161
31 345.7 57.71 0.1845
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-389 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -186.2212 6 384.4424
A2 -182.3775 10 384.755
A3 -186.2212 6 384.4424
R -204.9225 2 413.8449
2 -186.7751 3 379.5501
Additive constant for all log-likelihoods = -37.68. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Test
4 :
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Test
Test 1
Test 2
Test 3
Test 4
Tests of Interest
-2*log(Likelihood Ratio)
45.09
7 . 687
7 . 687
1.108
p-value
< 0.0001
0.1037
0.1037
0.7752
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is greater than .1. Model 2 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 6.12391
BMDL = 4.52632
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-390 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.13. DeCaprio et al. (1986): Body Weight, Males
2 E.3.13.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
0.79
6.75
0.08
419.55
4.5E+00
2.9E+00
nonconstant variance,
power restricted >1
exponential (M3)
3
0.79
6.75
0.08
419.55
4.5E+00
2.9E+00
nonconstant variance,
power restricted >1
exponential (M4)
2
0.79
4.68
0.10
419.49
2.8E+00
1.5E+00
nonconstant variance,
power restricted >1
exponential (M5)
2
0.79
4.68
0.10
419.49
2.8E+00
1.5E+00
nonconstant variance,
power restricted >1
Hill
2
0.79
4.56
0.10
419.36
2.6E+00
error
nonconstant variance,
n restricted >1, bound
hit
linear
3
0.79
8.23
0.04
421.04
5.8E+00
3.9E+00
nonconstant variance
polynomial
2
0.79
4.82
0.09
419.63
2.9E+00
1.8E+00
nonconstant variance
power
3
0.79
8.23
0.04
421.04
5.8E+00
3.9E+00
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
3
0.79
6.14
0.11
417.87
4.8E+00
3.7E+00
constant variance,
power restricted >1
exponential (M3)
3
0.79
6.14
0.11
417.87
4.8E+00
3.7E+00
constant variance,
power restricted >1
exponential (M4)
2
0.79
4.08
0.13
417.81
2.9E+00
1.8E+00
constant variance,
power restricted >1
exponential (M5)
2
0.79
4.08
0.13
417.81
2.9E+00
1.8E+00
constant variance,
power restricted >1
Hillc
2
0.79
3.98
0.14
417.70
2.8E+00
1.6E+00
constant variance, n
restricted >1, bound
hit
linear
3
0.79
7.56
0.06
419.29
6.1E+00
4.8E+00
constant variance
polynomial
2
0.79
4.20
0.12
417.93
3.1E+00
2.0E+00
constant variance
power
3
0.79
7.56
0.06
419.29
6.1E+00
4.8E+00
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-391 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.13.2. Figure for Selected Model: Hill, Constant Variance, n Restricted >1, Bound Hit
Hill Model with 0.95 Confidence Level
Hill
700
600
500
400
300
BMD
BMDL
0
5
10
15
20
25
dose
14:09 10/06 2009
E.3.13.3. Output File for Selected Model: Hill, Constant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_male_BW.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_male_BW.plt
Tue Oct 06 14:09:41 2009
Male BW Tbl 1
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-392 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha
rho
intercept
3408.2
0 Specified
713
-280
0.5774
3.62353
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
1. le-008
5.5e-008
-5.8e-008
intercept
1. le-008
1
0.27
-0.46
v
5.5e-008
0.27
1
-0. 94
k
-5.8e-008
-0.46
-0. 94
1
Parameter Estimates
Variable
alpha
intercept
k
Estimate
3309.29
688.613
-430.869
1
18 .1326
Std. Err.
6 97.6 6
11.4849
146.654
NA
12.6279
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
1941.9
666.103
-718.305
-6.61762
4676.68
711.123
-143.432
42.8828
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
0.12
0. 61
4 . 9
26
10
10
11
10
4
713
682
651
603
433
689
686
675
597
435
47 . 4
50. 6
63
63.2
76
57 . 5
57 . 5
57 . 5
57 . 5
57 . 5
1. 34
-0.208
-1. 36
0.333
-0.0617
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-393 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Model R: Yi = Mu + e(i
Var{e(i) ) = Sigma'"2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -202.863522 6 417.727044
A2 -202.022697 10 424.045393
A3 -202.863522 6 417.727044
fitted -204.851020 4 417.702041
R -226.717147 2 457.434293
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2^log(Likelihood Ratio)
Test df
p-value
Test 1
Test 2
Test 3
Test 4
49.3889
1.68165
1.68165
3. 975
<.0001
0.794
0.794
0.137
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 2.7 9396
BMDL = 1.55261
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-394 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.14. DeCaprio et al. (1986): Relative Brain Weight, Males
2 E.3.14.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.99
5.25
0.07
-194.78
2.9E+00
2.0E+00
nonconstant variance,
power restricted >1
exponential (M3)
2
0.99
5.25
0.07
-194.78
2.9E+00
2.0E+00
nonconstant variance,
power restricted >1
exponential (M4)
1
0.99
0.02
0.89
-198.01
4.4E-01
1.8E-01
nonconstant variance,
power restricted >1
exponential (M5)
1
0.99
0.02
0.89
-198.01
4.4E-01
1.8E-01
nonconstant variance,
power restricted >1
Hill
0
0.99
0.00
NA
-196.03
4.1E-01
1.3E-01
nonconstant variance,
n restricted >1
linear
2
0.99
5.09
0.08
-194.93
2.8E+00
1.9E+00
nonconstant variance
polynomial
1
0.99
0.08
0.78
-197.95
4.9E-01
2.7E-01
nonconstant variance
power
2
0.99
5.09
0.08
-194.93
2.8E+00
1.9E+00
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.99
5.27
0.07
-196.72
2.8E+00
2.1E+00
constant variance,
power restricted >1
exponential (M3)
2
0.99
5.27
0.07
-196.72
2.8E+00
2.1E+00
constant variance,
power restricted >1
exponential (M4)
C
1
0.99
0.02
0.89
-199.97
4.6E-01
2.1E-01
constant variance,
power restricted >1
exponential (M5)
1
0.99
0.02
0.89
-199.97
4.6E-01
2.1E-01
constant variance,
power restricted >1
Hill
0
0.99
0.00
NA
-197.99
4.3E-01
1.5E-01
constant variance, n
restricted >1
linear
2
0.99
5.10
0.08
-196.88
2.7E+00
1.9E+00
constant variance
polynomial
1
0.99
0.08
0.78
-199.91
5.0E-01
3.0E-01
constant variance
power
2
0.99
5.10
0.08
-196.88
2.7E+00
1.9E+00
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-395 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.14.2. Figure for Selected Model: Exponential (M4), Constant Variance, Power Restricted
>1
Exponential_beta Model 4 with 0.95 Confidence Level
0.7
Exponential
0.65
0.6
0.55
0.5
BMD
BMDL
0
1
2
3
4
5
dose
14:11 10/06 2009
E.3.14.3. Output File for Selected Model: Exponential (M4), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\ExpConstVar_BMRl_rel_male_brain_wt.(d)
Gnuplot Plotting File:
Tue Oct 06 14:11:05 2009
Rel Male Brain wt, Tbl 2
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-396 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
Specified
lnalpha -6.07283
rho(S) 0
a 0.5111
b 0.351325
c 1.33125
d 1
Parameter Estimates
Variable Model 4
lnalpha -6.07239
rho 0
a 0.53911
b 1.25838
c 1.20224
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 10 0.538 0.04743
0.12 10 0.556 0.0506
0.61 11 0.597 0.05307
4.9 10 0.648 0.0506
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 0.5391 0.04802 -0.0731
0.12 0.5544 0.04802 0.106
0.61 0.5975 0.04802 -0.03703
4.9 0.6479 0.04802 0.005972
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-397 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 103.9931 5 -197.9862
A2 104.0646 8 -192.1293
A3 103.9931 5 -197.9862
R 92.4089 2 -180.8178
4 103.9841 4 -199.9682
Additive constant for all log-likelihoods = -37.68. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2^log(Likelihood Ratio)
23.31
0.1431
0.1431
0.01798
p-value
0.0006986
0.9862
0.9862
0.8933
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 0.461347
BMDL = 0.209454
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-398 DRAFT—DO NOT CITE OR QUOTE
-------
2 E.3.15. DeCaprio et al. (1986): Relative Liver Weight, Females
3 E.3.15.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
C
2
0.02
5.20
0.07
38.53
3.4E+00
2.1E+00
nonconstant variance,
power restricted >1
exponential (M3)
1
0.02
4.15
0.04
39.48
4.7E+00
2.3E+00
nonconstant variance,
power restricted >1
exponential (M4)
1
0.02
5.33
0.02
40.66
3.3E+00
1.9E+00
nonconstant variance,
power restricted >1
exponential (M5)
0
0.02
4.15
N/A
41.48
4.6E+00
7.3E-01
nonconstant variance,
power restricted >1
Hill
0
0.02
4.15
NA
41.48
4.6E+00
7.5E-01
nonconstant variance, n
restricted >1
linear
2
0.02
5.33
0.07
38.66
3.3E+00
1.9E+00
nonconstant variance
polynomial
1
0.02
2.86
0.09
38.20
4.9E+00
2.7E+00
nonconstant variance
power
1
0.02
4.15
0.04
39.48
4.7E+00
2.2E+00
nonconstant variance,
power restricted >1
exponential (M2)
2
0.02
0.63
0.73
39.73
3.9E+00
2.7E+00
constant variance, power
restricted >1
exponential (M3)
1
0.02
0.30
0.58
41.40
4.7E+00
2.7E+00
constant variance, power
restricted >1
exponential (M4)
1
0.02
0.69
0.41
41.78
3.8E+00
1.2E+00
constant variance, power
restricted >1
exponential (M5)
0
0.02
0.30
N/A
43.40
4.7E+00
7.2E-01
constant variance, power
restricted >1
Hill
0
0.02
0.30
NA
43.40
4.7E+00
7.3E-01
constant variance, n
restricted >1
linear
2
0.02
0.68
0.71
39.78
3.8E+00
2.5E+00
constant variance
polynomial
1
0.02
0.24
0.62
41.34
4.6E+00
1.2E+00
constant variance
power
1
0.02
0.30
0.58
41.40
4.7E+00
2.5E+00
constant variance, power
restricted >1
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
4
5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-399 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.15.2. Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
6.5
5.5
4.5
3.5
Exponential
dose
14:10 10/06 2009
E.3.15.3. Output File for Selected Model: Exponential (M2'), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\Exp_BMRl_rel_fem_liver_wt.(d)
Gnuplot Plotting File:
Tue Oct 06 14:10:17 2009
Relative Female Liver wt, Tbl 2
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign
exp{sign
[c—(c—1)
[c—(c—1)
b * dose}
(b * dose)^d}
exp{-b * dose}]
exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable
Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-400 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
lnalpha
rho
Model 2
-9.34924
5. 89997
4.0565
0.378048
1.43399
1
Parameter Estimates
Variable Model 2
lnalpha
-6.47175
rho
4 . 07772
a
4 . 36037
b
0.196303
c
1.57543
d
9.64518
Table
of Stats From Input
Data
Dose
N
Obs Mean
Obs Std
0
8
4 . 3
0.7354
0.12
10
4.49
1 .107
CD
<£>
O
9
4 . 27
0.48
CO
10
5.54
1. 36
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 4.312 0.7964 -0.04371
0.12 4.338 0.8054 0.5953
0.68 4.462 0.8483 -0.6795
4.86 5.505 1.25 0.08976
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-401 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Model R: Yij = Mu + e(i
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -16.54794 5 43.09588
A2 -11.40563 8 38.81126
A3 -12.66678 6 37.33356
R -21.58737 2 47.17474
2 -15.2671 4 38.53419
Additive constant for all log-likelihoods = -34. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 20.36 6 0.002385
Test 2 10.28 3 0.0163
Test 3 2.522 2 0.2833
Test 4 5.201 2 0.07425
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is less than .1. Model 2 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 3.37444
BMDL = 2.10781
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-402 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.16. DeCaprio et al. (1986): Relative Liver Weight, Males
2 E.3.16.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
<0.0001
25.86
<0.0001
65.17
6.2E+00
3.7E+00
nonconstant variance,
power restricted >1
exponential (M3)
1
<0.0001
25.30
<0.0001
66.60
5.1E+00
3.9E+00
nonconstant variance,
power restricted >1
exponential (M4)
C
1
<0.0001
16.82
<0.0001
58.12
3.8E-01
1.4E-01
nonconstant variance,
power restricted >1
exponential (M5)
0
<0.0001
7.80
N/A
51.11
3.5E-01
1.3E-01
nonconstant variance,
power restricted >1
Hill
1
<.0001
7.80
0.01
49.11
2.8E-01
error
nonconstant variance, n
restricted >1, bound hit
linear
2
<.0001
25.90
<.0001
65.20
6.3E+00
3.4E+00
nonconstant variance
polynomial
2
<.0001
25.39
<.0001
64.69
5.8E+00
4.4E+00
nonconstant variance
power
1
<.0001
25.30
<.0001
66.60
5.1E+00
3.7E+00
nonconstant variance,
power restricted >1
exponential (M2)
2
<0.0001
5.09
0.08
64.15
5.6E+00
3.4E+00
constant variance,
power restricted >1
exponential (M3)
2
<0.0001
5.09
0.08
64.15
5.6E+00
3.4E+00
constant variance,
power restricted >1
exponential (M4)
1
<0.0001
2.15
0.14
63.21
1.1E+00
2.7E-01
constant variance,
power restricted >1
exponential (M5)
0
<0.0001
0.72
N/A
63.78
6.3E-01
1.3E-01
constant variance,
power restricted >1
Hill
0
<.0001
0.72
NA
63.78
6.5E-01
error
constant variance, n
restricted >1
linear
2
<.0001
5.00
0.08
64.06
5.5E+00
3.2E+00
constant variance
polynomial
2
<.0001
5.00
0.08
64.06
5.5E+00
3.2E+00
constant variance
power
2
<.0001
5.00
0.08
64.06
5.5E+00
3.2E+00
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-403 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.16.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
"
"|
: '
I
/
-
BMDL
BMD
0 1 2 3 4 5
dose
14:11 10/06 2009
E.3.16.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\Exp_BMRl_rel_male_liver_wt.(d)
Gnuplot Plotting File:
Tue Oct 06 14:11:44 2009
Rel Male Liver wt, Tbl 2
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-404 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
lnalpha
rho
Model 4
-10.8833
6.71347
3. 895
0. 428412
1.51772
1
Parameter Estimates
Variable Model 4
lnalpha
-12.146
rho
7 . 6297
a
4 . 32
b
2.79927
c
1.2705
d
18
Table
of Stats From Input
Data
Dose
N
Obs Mean
Obs Std
0
10
4 . 54
0 .7273
0.12
10
4 .1
0.4427
0. 61
11
5.36
2 . 023
4 . 9
10
5. 63
0.9171
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 4.333 0.6741 0.9699
0.12 4.633 0.8928 -1.889
0.61 5.172 1.417 0.4393
4.9 5.338 1.617 0.5712
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-405 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
Model R: Yij = Mu + e(i
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -26.53142 5 63.06284
A2 -13.9487 8 43.89739
A3 -15.65277 6 43.30554
R -31.57211 2 67.14421
4 -24.06213 5 58.12426
Additive constant for all log-likelihoods = -37.68. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 4 fit the data? (A3 vs 4)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 35.25 6 < 0.0001
Test 2 25.17 3 < 0.0001
Test 3 3.408 2 0.1819
Test 6a 16.82 1 < 0.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is less than .1. Model 4 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 0.376095
BMDL = 0.137425
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-406 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.17. DeCaprio et al. (1986): Relative Thymus Weight, Males
2 E.3.17.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
J2 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.07
7.12
0.03
-308.01
6.0E+00
3.4E+00
nonconstant variance,
power restricted >1
exponential (M3)
2
0.07
7.12
0.03
-308.01
6.0E+00
3.4E+00
nonconstant variance,
power restricted >1
exponential (M4)
1
0.07
5.77
0.02
-307.36
error
error
nonconstant variance,
power restricted >1
exponential (M5)
1
0.07
5.77
0.02
-307.36
error
error
nonconstant variance,
power restricted >1
Hill
1
0.07
4.71
0.03
-308.42
error
error
nonconstant variance,
n restricted >1, bound
hit
linear
2
0.07
7.19
0.03
-307.94
5.9E+00
3.6E+00
nonconstant variance
polynomial
1
0.07
5.98
0.01
-307.16
2.0E+00
6.8E-01
nonconstant variance
power
2
0.07
7351250
<0.0001
7350937
error
error
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
C
2
0.07
4.25
0.12
-306.73
5.1E+00
2.8E+00
constant variance,
power restricted >1
exponential (M3)
2
0.07
4.25
0.12
-306.73
5.1E+00
2.8E+00
constant variance,
power restricted >1
exponential (M4)
1
0.07
3.29
0.07
-305.69
1.8E+00
7.6E-03
constant variance,
power restricted >1
exponential (M5)
1
0.07
3.29
0.07
-305.69
1.8E+00
7.1E-03
constant variance,
power restricted >1
Hill
1
0.07
2.10
0.15
-306.88
3.2E-01
5.1E-07
constant variance, n
restricted >1, bound
hit
linear
2
0.07
4.30
0.12
-306.68
5.2E+00
3.1E+00
constant variance
polynomial
1
0.07
3.48
0.06
-305.50
1.4E+00
5.1E-01
constant variance
power
2
0.07
4.30
0.12
-306.68
5.2E+00
3.1E+00
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-407 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.17.2.
Figure for selected Model: Exponential (M2), Constant Variance, Power Restricted
>1
Exponential_beta Model 2 with 0.95 Confidence Level
0.095
Exponential
0.09
0.085
0.08
0.075
0.07
0.065
0.06
0.055
BMDL
0.05
BMP
0
1
2
3
4
5
dose
14:12 10/06 2009
E.3.17.3. Output File for Selected Model: Exponential (M2), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\ExpConstVar_BMRl_rel_male_thymus_wt.(d)
Gnuplot Plotting File:
Tue Oct 06 14:12:33 2009
Rel Male Thymus wt, Tbl 2
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-408 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
Specified
lnalpha -8.73123
rho(S) 0
a 0.0819
b 0.468839
c 0.686086
d 1
Parameter Estimates
Variable Model 2
lnalpha -8.65107
rho 0
a 0.0739582
b 1.27978
c 0.80201
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 10 0.078 0.01897
0.12 10 0.066 0.009487
0.61 11 0.068 0.01327
4.9 10 0.059 0.009487
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 0.07145 0.01338 1.548
0.12 0.0711 0.01338 -1.205
0.61 0.0697 0.01338 -0.4219
4.9 0.05857 0.01338 0.1014
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-409 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A3:
Yij
Var{e(ij;
Mu(i) + e(ij)
exp(lalpha + log(mean(
rho)
Model R: Yij
Var{e(ij)}
Mu + e(i
Sigma""" 2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 158.4903 5 -306.9805
A2 161.9563 8 -307.9126
A3 158.4903 5 -306.9805
R 153.4442 2 -302.8885
2 156.3648 3 -306.7296
Additive constant for all log-likelihoods = -37.68. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs.
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 17.02 6 0.009195
Test 2 6.932 3 0.07409
Test 3 6.932 3 0.07409
Test 4 4.251 2 0.1194
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. Consider running
a non-homogeneous variance model.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 4 is greater than .1. Model 2 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 5.11373
BMDL = 2.82487
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-410 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.18. Hojo et al. (2002): DRL Reinforce per Min
2 E.3.18.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
J2 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.43
9.01
0.01
10.48
6.9E+02
1.7E+02
nonconstant variance,
power restricted >1
exponential (M3)
1
0.43
8.95
0.00
12.43
3.9E+02
1.8E+02
nonconstant variance,
power restricted >1
exponential (M4)
1
0.43
2.92
0.09
6.39
1.0E+01
1.5E-01
nonconstant variance,
power restricted >1
exponential (M5)
0
0.43
2.31
N/A
7.78
2.0E+01
error
nonconstant variance,
power restricted >1
Hill
1
0.43
1JQNAN
<.0001
7.60
error
error
nonconstant variance,
n restricted >1
linear
2
0.43
9.21
0.01
11.02
error
error
nonconstant variance
polynomial
2
0.43
8.79
0.01
10.60
error
4.4E+02
nonconstant variance
power
2
0.43
8.28
0.02
9.98
error
error
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.43
10.12
0.01
9.89
3.0E+02
1.5E+02
constant variance,
power restricted >1
exponential (M3)
2
0.43
10.12
0.01
9.89
3.0E+02
1.5E+02
constant variance,
power restricted >1
exponential (M4)
1
0.43
3.47
0.06
5.24
1.7E+01
3.8E-02
constant variance,
power restricted >1
exponential (M5)
0
0.43
2.70
N/A
6.46
2.1E+01
1.2E-05
constant variance,
power restricted >1
Hill
0
0.43
2.70
NA
6.46
2.1E+01
1.7E-05
constant variance, n
restricted >1
linear0
2
0.43
9.78
0.01
9.55
2.7E+02
1.1E+02
constant variance
polynomial
2
0.43
9.78
0.01
9.55
2.7E+02
1.1E+02
constant variance
power
2
0.43
9.78
0.01
9.55
2.7E+02
1.1E+02
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-411 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.18.2. Figure for Selected Model: Linear, Constant Variance
Linear Model with 0.95 Confidence Level
Linear
1
0.5
0)
w
0
-0.5
-1
-1.5
BMDL
BMC I
0
50
100
150
200
250
dose
14:33 10/06 2009
E.3.18.3. Output File for Selected Model: Linear, Constant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\LinearConstVar_BMRl_DRL_reinforce_per_min.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AniDose\LinearConstVar_BMRl_DRL_reinforce_per_min.pit
Tue Oct 06 14:33:14 2009
Table 5
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose^2 + ...
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Signs of the polynomial coefficients are not restricted
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-412 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha = 0.337763
rho = 0 Specified
beta_0 = -0.4 04
beta~l = 0.00249615
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
beta_0
beta 1
alpha
1
-1. 4e-008
2 . 2e-008
beta_0
-1.4e-008
1
-0.69
beta_l
2 . 2e-008
-0.69
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
beta_0
beta 1
Estimate
0. 435671
-0.372098
0.00246548
Std. Err.
0.134451
0.198702
0. 00211361
Lower Conf. Limit
0.172152
-0.761547
-0.00167711
Upper Conf. Limit
0 . 6 9 919
0. 017352
0. 00660807
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
20
60
180
-0.814
-0.364
0.374
-0.163
-0.372
-0.323
-0.224
0.0717
0.448
0. 821
0.54
0.443
0. 66
0. 66
0. 66
0. 66
-1. 5
-0.14
2 . 22
-0.795
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma'*-2
Likelihoods of Interest
Model Log(likelihood)
A1 3.115550
A2 4.489557
A3 3.115550
Param'
5
AIC
3.768900
7 . 020886
3.768900
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-413 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
fitted -1.775882 3 9.551763
R -2.435087 2 8.870174
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 13.8493 6 0.03137
Test 2 2.74801 3 0.4321
Test 3 2.74801 3 0.4321
Test 4 9.78286 2 0.007511
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 267.718
BMDL = 110.032
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-414 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.19. Hojo et al. (2002): DRL Response per Min
2 E.3.19.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
J2 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.30
-0.14
N/A
122.33
4.3E+00
error
nonconstant variance,
power restricted >1
exponential (M3)
2
0.30
25.76
<0.0001
148.23
error
error
nonconstant variance,
power restricted >1
exponential (M4)
2
0.30
-0.12
N/A
122.35
4.7E+00
3.2E-01
nonconstant variance,
power restricted >1
exponential (M5)
1
0.30
0.16
0.69
124.63
6.4E+00
4.1E-01
nonconstant variance,
power restricted >1
Hill
0
0.30
l.#QNAN
NA
127.53
1.3E+01
1.8E-13
nonconstant variance,
n restricted >1
linear
2
0.30
11.09
0.00
133.30
2.1E+02
9.9E+01
nonconstant variance
polynomial
2
0.30
11.09
0.00
133.30
2.1E+02
error
nonconstant variance
power
2
0.30
12.33
0.00
133.30
2.2E+02
9.6E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.30
1.13
0.57
122.98
6.2E+00
error
constant variance,
power restricted >1
exponential (M3)
2
0.30
1.13
0.57
122.98
6.2E+00
error
constant variance,
power restricted >1
exponential (M4)
C
1
0.30
0.50
0.48
124.36
4.8E+00
2.7E-01
constant variance,
power restricted >1
exponential (M5)
0
0.30
0.50
N/A
126.35
1.1E+01
2.1E-01
constant variance,
power restricted >1
Hill
0
0.30
0.50
NA
126.35
1.6E+01
1.8E-13
constant variance, n
restricted >1
linear
2
0.30
10.97
0.00
132.83
2.1E+02
9.8E+01
constant variance
polynomial
2
0.30
10.97
0.00
132.83
2.1E+02
9.8E+01
constant variance
power
2
0.30
10.97
0.00
132.83
2.1E+02
9.8E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-415 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.19.2.
Figure for Selected Model: Exponential (M4), Constant Variance, Power Restricted
>1
Exponential_beta Model 4 with 0.95 Confidence Level
30
20
10
-10
Exponential
~i r~
BJV1DL- BMP
0 20
14:34 10/06 2009
40 60 80 100 120 140 160 180
dose
E.3.19.3. Output File for Selected Model: Exponential (M4'), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\ExpConstVar_BMRl_DRL_response_per_min.(d)
Gnuplot Plotting File:
Tue Oct 06 14:34:01 2009
Table 5, values adjusted by a constant to allow exponential model
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign
exp{sign
[c—(c—1)
[c—(c—1)
b * dose}
(b * dose)^d}
exp{-b * dose}]
exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable
Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-416 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 4
lnalpha
rho(S)
4.51689
0
24.6362
0.0212679
0.0184785
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 4
4.54064
0
23.463
0. 073228
0.100111
2 .44375
Table of Stats From Input Data
Obs Mean
Obs Std Dev
0
20
60
180
5
23. 46
4 . 013
0.478
4 .594
7 . 986
10. 96
7.194
15.23
0
20
60
180
Estimated Values of Interest
Est Mean Est Std Scaled Residual
23. 47
3. 973
2 . 37
2 . 361
9. 683
9. 683
9. 683
9. 683
-0.0004677
0.009182
-0.4787
0.5157
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-417 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -57.92733 5 125.8547
A2 -56.09669 8 128.1934
A3 -57.92733 5 125.8547
R -64.49611 2 132.9922
4 -58.17787 4 124.3557
Additive constant for all log-likelihoods = -19.3. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2^log(Likelihood Ratio)
16.8
3. 661
3. 661
0.5011
p-value
0.01005
0.3004
0.3004
0.479
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 4.77493
BMDL = 0.270447
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-418 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.20. Kattainen et al. (2001): 3rd Molar Mesio-Distal Length (Molar Development)
2 E.3.20.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
<0.0001
38.91
<0.0001
-122.95
4.0E+02
2.4E+02
nonconstant variance,
power restricted >1
exponential (M3)
3
<0.0001
38.91
<0.0001
-122.95
4.0E+02
2.4E+02
nonconstant variance,
power restricted >1
exponential (M4)
2
<0.0001
79.12
<0.0001
-80.75
error
error
nonconstant variance,
power restricted >1
exponential (M5)
1
<0.0001
79.12
<0.0001
-78.75
error
error
nonconstant variance,
power restricted >1
Hillc
2
<0001
8.72
0.01
-151.15
4.1E+00
2.1E+00
nonconstant
variance, n
restricted >1, bound
hit
linear
3
<.0001
39.54
<.0001
-122.33
4.7E+02
3.0E+02
nonconstant variance
polynomial
2
<.0001
36.57
<.0001
-123.30
1.9E+02
9.0E+01
nonconstant variance
power
3
<.0001
39.54
<.0001
-122.33
4.7E+02
3.0E+02
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
3
<0.0001
7.81
0.05
-99.70
8.5E+02
5.6E+02
constant variance,
power restricted >1
exponential (M3)
3
<0.0001
7.81
0.05
-99.70
8.5E+02
5.6E+02
constant variance,
power restricted >1
exponential (M4)
2
<0.0001
5.05
0.08
-100.47
2.3E+02
1.0E+00
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
5.05
0.08
-100.47
2.3E+02
8.1E-01
constant variance,
power restricted >1
Hill
2
<.0001
3.23
0.20
-102.29
8.1E+01
1.1E+01
constant variance, n
restricted >1, bound
hit
linear
3
<.0001
8.07
0.04
-99.45
8.8E+02
6.2E+02
constant variance
polynomial
2
<.0001
5.88
0.05
-99.64
3.7E+02
1.8E+02
constant variance
power
3
<.0001
8.07
0.04
-99.45
8.8E+02
6.2E+02
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-419 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.20.2. Figure for Selected Model: Hill, Nonconstant Variance, n Restricted >1, Bound Hit
1.9
1.8
1.7
1.6
1.5
1.4
1.3
1.2
1.1
Hill Model with 0.95 Confidence Level
Hill
i/IDLBMD
200
400 600
dose
800
1000
14:36 10/06 2009
E.3.20.3. Output File for Selected Model: Hill, Nonconstant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\Hill_BMRl_3rd_molar.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\Hill_BMRl_3rd_molar.plt
Tue Oct 06 14:36:22 2009
Figure 3 female only
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter restricted to be greater than 1
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-420 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha
rho
intercept
-2 . 37155
0
1. 85591
-0.507874
0. 826204
27 . 3305
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha
rho
intercept
lalpha
1
-0. 98
-0.16
0.84
-0.37
rho
-0. 98
1
0.2
-0.79
0.39
intercept
-0.16
0.2
1
-0.31
-0.11
v
0.84
-0.79
-0.31
1
-0.48
k
-0.37
0.39
-0.11
-0.48
1
Parameter Estimates
Variable
lalpha
rho
intercept
k
Estimate
3.34561
-14.3325
1. 8548
-0.441166
1
24.0343
Std. Err.
1.40443
2 . 62129
0.0159017
0.058852
NA
7.84495
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
0.592981
-19.4701
1.82364
-0.556513
8.65852
6.09824
-9.19484
1.88597
-0.325818
39.4101
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
30
100
300
1000
16
17
15
12
19
1.86
1. 58
1. 6
1. 5
1. 35
1. 85
1. 61
1. 5
1.45
1.42
0.0661
0.185
0.265
0.221
0.515
0.0637
0.176
0.293
0.378
0. 423
0.0692
-0.768
1.28
0.527
-0.783
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e( ij ) }
Model A2: Yij
Var{e(ij)}
Model A3: Yij
Var{e(ij)}
Mu(i) + e(ij)
Sigma/S2
Mu(i) + e(ij)
Sigma(i)^2
Mu ( i) + e(ij )
exp(lalpha + rho*ln(Mu(i)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-421 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 56.758717 6 -101.517434
A2 85.856450 10 -151.712901
A3 84.934314 7 -155.868628
fitted 80.575940 5 -151.151880
R 45.373551 2 -86.747101
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 80.9658 8 <.0001
Test 2 58.1955 4 <.0001
Test 3 1.84427 3 0.6053
Test 4 8.71675 2 0.0128
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 4 . 05231
BMDL = 2.14357
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-422 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.21. Kattainen et al. (2001): Females 3rd Molar Eruption
2 E.3.21.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
logistic
3
3.73
0.29
89.06
1.0E+02
7.3E+01
logistic
3
3.73
0.29
89.06
1.9E+02
1.4E+02
log-logistic b
3
0.48
0.92
85.53
2.3E+01
1.2E+01
slope restricted >1,
bound hit
log-logistic
3
0.48
0.92
85.53
4.8E+01
2.5E+01
slope restricted >1, bound
hit
log-probit
2
0.12
0.94
87.18
1.3E+01
5.2E-01
slope restricted >1
log-probit
2
0.12
0.94
87.18
2.8E+01
2.3E+00
slope restricted >1
multistage, 1-
degree
3
1.68
0.64
86.80
4.2E+01
2.7E+01
betas restricted >0
multistage, 1-
degree
3
1.68
0.64
86.80
8.7E+01
5.5E+01
betas restricted >0
probit
3
3.62
0.31
88.92
9.8E+01
7.1E+01
probit
3
3.62
0.31
88.92
1.9E+02
1.4E+02
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-423 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.21.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
.8
0.6
0.4
0.2
0
[iMDL BMD
0
200
400
600
800
1000
dose
11:05 10/15 2009
E.3.21.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File:
C:\USEPA\BMDS21\AniDose2\LogLogistic_BMRl_Female_3rd_molar_eruption.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AniDose2\LogLogistic_BMRl_Female_3rd_molar_eruption.pit
Thu Oct 15 11:05:20 2009
Figure 2
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope <= 1
Total number of observations = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-424 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
User has chosen the log transformed model
Default Initial Parameter Values
background = 0.0625
intercept = -6.063
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background
intercept
background
1
-0.56
intercept
-0.56
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0846785 * * *
intercept -6.06063 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -40.5286 5
Fitted model -40.7674 2 0.477533 3 0.9238
Reduced model -50.7341 1 20.411 4 0.0004142
AIC: 8 5.5347
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0847
1. 355
1.
000
16
-0.319
30.0000
0.1445
2 .457
3.
000
17
0.374
100.0000
0.2578
3. 867
4 .
000
15
0. 078
300.0000
0. 4615
5.538
6 .
000
12
0.267
1000.0000
0.7254
13.782
13.
000
19
-0.402
Chi ^2 = 0.48
d.f.
= 3 P
-value
= 0.9231
Benchmark Dose Computation
Specified effect = 0.05
Risk Type = Extra risk
Confidence level = 0.95
BMD = 22 . 5603
BMDL = 11.7531
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-425 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.22. Keller et al. (2006): Missing Mandibular Molars in CBA J Mice
2 E.3.22.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
1
2.63
0.10
52.49
5.2E+01
9.9E+00
power restricted >1
gamma
1
2.63
0.10
52.49
7.3E+01
2.0E+01
power restricted >1
logistic
2
2.28
0.32
50.10
4.6E+01
3.1E+01
logistic
2
2.28
0.32
50.10
7.2E+01
5.1E+01
log-logistic
1
2.62
0.11
52.52
8.5E+01
3.6E+01
slope restricted >1
log-logistic
1
2.62
0.11
52.52
9.3E+01
5.3E+01
slope restricted >1
log-probit
1
2.62
0.11
52.52
7.8E+01
3.9E+01
slope restricted >1
log-probit
1
2.62
0.11
52.52
8.9E+01
5.3E+01
slope restricted >1
multistage, 2-
degree
1
2.34
0.13
51.52
2.4E+01
1.1E+01
betas restricted >0
multistage, 1-
degree b
3
3.87
0.28
49.41
1.4E+01
9.2E+00
betas restricted >0
multistage, 2-
degree
1
2.34
0.13
51.52
4.6E+01
2.2E+01
betas restricted >0
multistage, 1-
degree
3
3.87
0.28
49.41
2.8E+01
1.9E+01
betas restricted >0
probit
2
2.25
0.32
50.03
4.3E+01
2.8E+01
probit
2
2.25
0.32
50.03
6.8E+01
4.8E+01
Weibull
1
2.58
0.11
52.22
3.7E+01
1.0E+01
power restricted >1
Weibull
1
2.58
0.11
52.22
6.1E+01
2.1E+01
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-426 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.22.2. Figure for Selected Model: Multistage, 1-Degree, Betas Restricted >0
Multistage Model with 0.95 Confidence Level
Multistage
1
.8
0.6
0.4
0.2
0
EMDLBMD
0
200
400
600
800
1000
dose
11:04 10/15 2009
E.3.22.3. Output File for Selected Model: Multistage, 1-Degree, Betas Restricted >0
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AniDose2\Multilst_BMRl_CBA_J_mandibular.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose2\Multilst_BMRl_CBA_J_mandibular.plt
Thu Oct 15 11:04:49 2009
Table 1 using mandibular molars only
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-427 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta (1) = 1.02909e + 017
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 0.00379264 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -21.5798 4
Fitted model -23.7044 1 4.24924 3 0.2358
Reduced model -71.326 1 99.4926 3 <.0001
AIC: 49.4088
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0000 0.000 0.000 29 0.000
10.0000 0.0372 0.856 2.000 23 1.260
100.0000 0.3156 9.153 6.000 29 -1.260
1000.0000 0.9775 29.324 30.000 30 0.832
Chi/S2 = 3.87 d.f. = 3 P-value = 0.2762
Benchmark Dose Computation
Specified effect = 0.05
Risk Type = Extra risk
Confidence level = 0.95
BMD = 13.5244
BMDL = 9.17 426
BMDU = 20.3135
Taken together, (9.17426, 20.3135) is a 90 % two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-428 DRAFT—DO NOT CITE OR QUOTE
-------
1 interval for the BMD
2
3
4
5 E.3.23. Kociba et al. (1978): Urinary Coproporphyrins, Females (Table 2)
6 E.3.23.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.03
19.68
<0.0001
84.01
7.1E+01
4.3E+01
nonconstant variance,
power restricted >1
exponential (M3)
2
0.03
19.68
<0.0001
84.01
7.1E+01
4.3E+01
nonconstant variance,
power restricted >1
exponential (M4)
C
1
0.03
4.23
0.04
70.56
1.6E+00
7.3E-01
nonconstant
variance, power
restricted >1
exponential (M5)
0
0.03
0.76
N/A
69.09
3.1E+00
1.0E+00
nonconstant variance,
power restricted >1
linear
2
0.03
19.38
<.0001
83.71
6.2E+01
3.1E+01
nonconstant variance
polynomial
2
0.03
19.38
<.0001
83.71
6.2E+01
3.1E+01
nonconstant variance
power
2
0.03
19.38
<.0001
83.71
6.2E+01
3.1E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.03
12.65
0.00
82.04
6.9E+01
4.7E+01
constant variance,
power restricted >1
exponential (M3)
2
0.03
12.65
0.00
82.04
6.9E+01
4.7E+01
constant variance,
power restricted >1
exponential (M4)
1
0.03
1.96
0.16
73.36
2.7E+00
1.1E+00
constant variance,
power restricted >1
exponential (M5)
0
0.03
0.41
N/A
73.80
8.3E+00
1.0E+00
constant variance,
power restricted >1
Hill
0
0.03
0.41
NA
73.80
7.6E+00
error
constant variance, n
restricted >1
linear
2
0.03
12.32
0.00
81.72
6.1E+01
3.8E+01
constant variance
polynomial
2
0.03
12.32
0.00
81.72
6.1E+01
3.8E+01
constant variance
power
2
0.03
12.32
0.00
81.72
6.1E+01
3.8E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
7
8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-429 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.23.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
24
22
20
18
% 16
C
O
Q.
CO
S. 14
c
05
0
5 12
10
8
6
13:21 11/11
E.3.23.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential
EJMDL BMD
2009
20
40
60
80
100
dose
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Exp_BMRl_urin_copropor_f.(d)
Gnuplot Plotting File:
Wed Nov 11 13:21:02 2009
Table2-UrinaryCoproporphyrin
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-430 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
lnalpha
rho
Model 4
-5.58269
2.98472
8 .17
0.0259469
2.23623
1
Parameter Estimates
Variable
Model 4
lnalpha
-5.49254
rho
2 . 91176
a
9.2
b
0.295128
c
1. 83696
d
18
Table
of Stats From Input
Data
Dose
N
Obs Mean
Obs Std
0
5
CO
CO
<—i
1
5
CO
2
10
5
16.4
4 . 7
100
5
17 . 4
4
0
1
10
100
Estimated Values of Interest
Est Mean Est Std Scaled Residual
8 . 93
10. 04
15. 42
17 . 64
1.733
2 . 038
3. 683
4 .436
1.122
-1.582
0.5967
-0.1211
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-431 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -31.69739 5 73.39478
A2 -27.21541 8 70.43081
A3 -28.16434 6 68.32868
R -41.73188 2 87.46376
4 -30.27804 5 70.55608
Additive constant for all log-likelihoods = -18.38. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 4 fit the data? (A3 vs 4)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 29.03 6 < 0.0001
Test 2 8.964 3 0.02977
Test 3 1.898 2 0.3872
Test 6a 4.227 1 0.03978
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is less than .1. Model 4 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1.62505
BMDL = 0.729987
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-432 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.24. Kociba et al. (1978): Uroporphyrin per Creatinine, Females
2 E.3.24.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.49
1.45
0.48
-93.10
3.8E+01
2.6E+01
nonconstant variance,
power restricted >1
exponential (M3)
2
0.49
1.45
0.48
-93.10
3.8E+01
2.6E+01
nonconstant variance,
power restricted >1
exponential (M4)
1
0.49
0.73
0.39
-91.82
1.4E+01
4.4E+00
nonconstant variance,
power restricted >1
exponential (M5)
0
0.49
0.51
N/A
-90.03
1.0E+01
4.5E+00
nonconstant variance,
power restricted >1
Hill
0
0.49
0.51
NA
-90.03
1.0E+01
7.7E+00
nonconstant variance, n
restricted >1
linear
2
0.49
1.20
0.55
-93.35
2.9E+01
1.8E+01
nonconstant variance
polynomial
1
0.49
0.72
0.40
-91.83
1.3E+01
4.8E+00
nonconstant variance
power
2
0.49
1.20
0.55
-93.35
2.9E+01
1.8E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.49
0.83
0.66
-93.56
4.4E+01
3.3E+01
constant variance,
power restricted >1
exponential (M3)
2
0.49
0.83
0.66
-93.56
4.4E+01
3.3E+01
constant variance,
power restricted >1
exponential (M4)
1
0.49
0.31
0.58
-92.08
1.7E+01
5.5E+00
constant variance,
power restricted >1
exponential (M5)
0
0.49
0.20
N/A
-90.19
1.1E+01
5.6E+00
constant variance,
power restricted >1
linearc
2
0.49
0.66
0.72
-93.73
3.5E+01
2.5E+01
constant variance
polynomial
1
0.49
0.31
0.58
-92.08
1.7E+01
6.1E+00
constant variance
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-433 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.24.2. Figure for Selected Model: Linear, Constant Variance
Linear Model with 0.95 Confidence Level
Linear
0.4
0.35
0.3
0.25
0.2
0.15
.1
BMDI
BMD
0
20
40
60
80
100
dose
15:00 10/06 2009
E.3.24.3. Output File for Selected Model: Linear, Constant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File:
C:\USEPA\BMDS21\AniDose\LinearConstVar_BMRl_Females_uroporphyrin_per_creatinine.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AniDose\LinearConstVar_BMRl_Females_uroporphyrin_per_creatinine.pit
Tue Oct 06 15:00:16 2009
Table 2
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose^2 + ...
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Signs of the polynomial coefficients are not restricted
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-434 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha = 0.0030385
rho = 0 Specified
beta_Q = 0.154759
beta 1 = 0.0014231
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
beta_0
beta 1
alpha
1
-2 . 2e-009
3.5e-009
beta_0
-2 . 2e-009
1
-0.55
beta_l
3.5e-009
-0.55
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
beta_0
beta 1
Estimate
0. 00251184
0.154759
0.0014231
Std. Err.
0.000794315
0.0134422
0.000267497
Lower Conf. Limit
0.000955015
0.128413
0. 000898818
Upper Conf. Limit
0. 00406867
0.181105
0. 00194739
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
1
10
100
0.157
0.143
0.181
0.296
0.155
0.156
0.169
0.297
0. 05
0. 037
0. 053
0. 074
0.0501
0.0501
0.0501
0.0501
0.1
-0.588
0.536
-0.0477
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 50.195349 5 -90.390697
A2 51.400051 8 -86.800103
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-435 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
A3 50.195349 5 -90.390697
fitted 49.867385 3 -93.734769
R 41.049755 2 -78.099510
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 20.7006 6 0.002076
Test 2 2.40941 3 0.4919
Test 3 2.40941 3 0.4919
Test 4 0.655928 2 0.7204
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 35.2176
BMDL = 25.0024
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-436 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.25. Latchoumycandane and Mathur (2002): Daily sperm Production
2 E.3.25.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.85
20.91
<0.0001
96.01
9.2E+01
4.5E+01
nonconstant variance,
power restricted >1
exponential (M3)
2
0.85
20.91
<0.0001
96.01
9.2E+01
4.5E+01
nonconstant variance,
power restricted >1
exponential (M4)
1
0.85
0.16
0.69
77.26
2.4E-01
8.8E-02
nonconstant variance,
power restricted >1
exponential (M5)
0
0.85
0.16
N/A
79.26
2.9E-01
8.8E-02
nonconstant variance,
power restricted >1
Hill
1
0.85
0.04
0.85
77.14
1.4E-01
1.3E-02
nonconstant variance, n
restricted >1, bound hit
linear
2
0.85
21.07
<.0001
96.18
9.5E+01
5.4E+01
nonconstant variance
polynomial
1
0.85
10.98
0.00
88.08
6.2E+00
3.7E+00
nonconstant variance
power
2
0.85
21.07
<.0001
96.18
9.5E+01
5.4E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.85
21.99
<0.0001
95.11
7.6E+01
4.0E+01
constant variance,
power restricted >1
exponential (M3)
2
0.85
21.99
<0.0001
95.11
7.6E+01
4.0E+01
constant variance,
power restricted >1
exponential (M4)
1
0.85
0.15
0.70
75.26
2.4E-01
1.0E-01
constant variance,
power restricted >1
exponential (M5)
0
0.85
0.15
N/A
77.26
3.7E-01
1.0E-01
constant variance,
power restricted >1
Hillc
1
0.85
0.03
0.86
75.14
1.4E-01
1.6E-02
constant variance, n
restricted >1, bound
hit
linear
2
0.85
22.20
<.0001
95.31
8.3E+01
4.9E+01
constant variance
polynomial
1
0.85
12.98
0.00
88.09
5.0E+00
3.2E+00
constant variance
power
2
0.85
22.20
<.0001
95.31
8.3E+01
4.9E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-437 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.25.2. Figure for Selected Model: Hill, Constant Variance, n Restricted >1, Bound Hit
Hill Model with 0.95 Confidence Level
i/lDLBMD
dose
15:09 10/06 2009
E.3.25.3. Output File for Selected Model: Hill, Constant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_sperm_prod.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_sperm_prod.plt
Tue Oct 06 15:09:27 2009
(xl0^6) Table 1 without Vitamin E
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-438 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha
rho
intercept
7 . 23328
0
22.19
-9.09
1.80484
0. 697086
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
). 3e-010
3e-008
5.3e-009
intercept
6.3e-010
1
-0.78
-0.23
v
3e-008
-0.78
1
-0.17
k
J.3e-009
-0.23
-0.17
1
Parameter Estimates
Variable
alpha
intercept
k
Estimate
6.03567
22.1885
-9.00869
1
0.386669
Std. Err.
1.74235
1.00316
1.26801
NA
0.265663
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
2.62073
20.2223
-11.4939
-0.134021
9. 45061
24 .1547
-6.52343
0. 907359
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
1
10
100
22 . 2
15.7
13.7
13.1
22 . 2
15.7
13.5
13.2
2 . 67
2 . 65
2.19
3.16
2.46
2.46
2.46
2.46
0.00151
-0.0218
0.134
-0.114
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-439 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
Model R:
Yi
Mu + e(i
Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -33.556444 5 77.112E
A2 -33.158811 8 82
A3 -33.556444 5 77
fitted -33.572245 4 75
R -47.392394 2 98
317623
112888
144490
784788
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio)
Test df
p-value
Test 1
Test 2
Test 3
Test 4
28.4672
0.795266
0.795266
0.031602
<.0001
0.8506
0.8506
0.8589
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.144988
BMDL = 0.0155926
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-440 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.26. Li et al. (2006): Hormone Levels (Estradiol)
2 E.3.26.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.47
4.98
0.08
272.78
3.0E+02
1.0E+02
nonconstant variance,
power restricted >1
exponential (M3)
2
0.47
4.98
0.08
272.78
3.0E+02
1.0E+02
nonconstant variance,
power restricted >1
exponential (M4)
1
0.47
0.32
0.57
270.12
error
error
nonconstant variance,
power restricted >1
exponential (M5)
0
0.47
0.32
N/A
272.12
error
error
nonconstant variance,
power restricted >1
Hill
1
0.47
0.32
0.57
270.12
error
error
nonconstant variance,
n restricted >1, bound
hit
linear
2
0.47
4.92
0.09
272.72
3.4E+02
9.7E+01
nonconstant variance
polynomial
2
0.47
4.92
0.09
272.72
3.4E+02
9.7E+01
nonconstant variance
power
2
0.47
4.92
0.09
272.72
3.4E+02
9.7E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
C
2
0.47
3.84
0.15
270.81
3.2E+02
1.1E+02
constant variance,
power restricted >1
exponential (M3)
2
0.47
3.84
0.15
270.81
3.2E+02
1.1E+02
constant variance,
power restricted >1
exponential (M4)
1
0.47
0.92
0.34
269.90
error
error
constant variance,
power restricted >1
exponential (M5)
0
0.47
0.92
N/A
271.90
error
error
constant variance,
power restricted >1
Hill
0
0.47
0.92
NA
271.90
error
error
constant variance, n
restricted >1
linear
2
0.47
3.78
0.15
270.75
3.6E+02
1.1E+02
constant variance
polynomial
2
0.47
3.78
0.15
270.75
3.6E+02
1.1E+02
constant variance
power
2
0.47
3.78
0.15
270.75
3.6E+02
1.1E+02
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-441 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.26.2.
Figure for Selected Model: Exponential (M2), Constant Variance, Power Restricted
>1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
35
30
25
20
15
10
5
0
BMDL
BMC I
0
50
100
150
200
250
300
dose
13:21 11/11 2009
E.3.26.3. Output File for Selected Model: Exponential (M2), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\ExpConst_BMRl_Li_Estradiol.(d)
Gnuplot Plotting File:
Wed Nov 11 13:21:55 2009
Figure 3
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-442 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
Specified
lnalpha 5.52431
rho(S) 0
a 9.5
b 0.0162139
c 2.73407
d 1
Parameter Estimates
Variable Model 2
lnalpha 5.54738
rho 0
a 10
b 0.842953
c 2.13158
d 1.46715
NC = No Convergence
Table of Stats From Input Data
N Obs Mean Obs Std Dev
0 10 10 12.48
2 10 20 19.97
50 10 24.74 14.98
100 10 17.89 18.31
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 16.66 16.61 -1.267
2 16.73 16.61 0.6227
50 18.57 16.61 1.173
100 20.71 16.61 -0.5362
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-443 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -130.4861 5 270.9723
A2 -129.2199 8 274.4398
A3 -130.4861 5 270.9723
R -132.6269 2 269.2537
2 -132.404 3 270.8079
Additive constant for all log-likelihoods = -36.76. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Test
4 :
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Test
Test 1
Test 2
Test 3
Test 4
Tests of Interest
-2*log(Likelihood Ratio)
6. 814
2 . 533
2 . 533
3. 836
p-value
0.3384
0.4694
0.4694
0.1469
The p-value for Test 1 is greater than .05. There may not be a
diffence between responses and/or variances among the dose levels
Modelling the data with a dose/response curve may not be appropriate.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is greater than .1. Model 2 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 317.4 97
BMDL = 111.954
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-444 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.27. Li et al. (2006): Hormone Levels (Progesterone)
2 E.3.27.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.00
16.99
0.00
330.20
4.9E+01
error
nonconstant variance,
power restricted >1
exponential (M3)
2
0.00
16.99
0.00
330.20
4.9E+01
error
nonconstant variance,
power restricted >1
exponential (M4)
C
1
0.00
0.82
0.37
316.03
1.6E-01
1.0E-01
nonconstant variance,
power restricted >1
exponential (M5)
0
0.00
0.82
N/A
318.03
4.9E-01
7.9E-02
nonconstant variance,
power restricted >1
Hill
1
0.00
0.81
0.37
316.02
2.2E-02
6.4E-05
nonconstant variance, n
restricted >1, bound hit
linear
2
0.00
17.93
0.00
331.13
7.5E+01
5.2E+01
nonconstant variance
polynomial
2
0.00
17.93
0.00
331.13
7.5E+01
5.2E+01
nonconstant variance
power
2
0.00
17.93
0.00
331.13
7.5E+01
4.5E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.00
3.97
0.14
329.50
6.8E+01
error
constant variance,
power restricted >1
exponential (M3)
2
0.00
3.97
0.14
329.50
6.8E+01
error
constant variance,
power restricted >1
exponential (M4)
1
0.00
0.14
0.71
327.66
2.2E+00
1.3E-01
constant variance,
power restricted >1
exponential (M5)
0
0.00
0.14
N/A
329.66
2.2E+00
2.8E-01
constant variance,
power restricted >1
Hill
1
0.00
0.12
0.73
327.64
2.4E+00
3.5E-05
constant variance, n
restricted >1, bound hit
linear
2
0.00
4.97
0.08
330.49
9.2E+01
5.7E+01
constant variance
polynomial
2
0.00
4.97
0.08
330.49
9.2E+01
5.7E+01
constant variance
power
2
0.00
4.97
0.08
330.49
9.2E+01
5.7E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-445 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.27.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
60
40
20
-20
BS/IDLBMD
13:22 11/11 2009
Exponential
20
40 60
dose
80
100
E.3.27.3. Output File for Selected Model: Exponential (M4'), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Exp_BMRl_Li_Progesterone.(d)
Gnuplot Plotting File:
Wed Nov 11 13:22:19 2009
Figure 4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-446 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
lnalpha
rho
Model 4
11.2757
-1.43319
65.0395
0.0460242
0.162232
1
Parameter Estimates
Variable
lnalpha
rho
Model 4
14 . 0852
-2 .26856
61.9568
1.02041
0.315961
1.78188
Dose
Table of Stats From Input Data
Obs Mean
Obs Std Dev
0
2
50
100
10
10
10
10
61. 94
30.72
16. 62
11. 08
11.15
39. 81
33. 44
44.59
0
2
50
100
Estimated Values of Interest
Est Mean Est Std Scaled Residual
61. 96
20.78
19.58
19.58
10. 61
36. 65
39.21
39.21
-0.0043
0. 858
-0.2385
-0.6853
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-447 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
Model R: Yij = Mu + e(i
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -159.7613 5 329.5225
A2 -151.9206 8 319.8412
A3 -152.6038 6 317.2077
R -165.9023 2 335.8046
4 -153.0132 5 316.0265
Additive constant for all log-likelihoods = -36.76. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 4 fit the data? (A3 vs 4)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 27.96 6 < 0.0001
Test 2 15.68 3 0.001318
Test 3 1.366 2 0.505
Test 6a 0.8188 1 0.3655
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 0.161712
BMDL = 0.100383
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-448 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.28. Markowski et al. (2001): FRIO Run Opp
2 E.3.28.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.17
6.17
0.05
119.08
2.3E+02
5.3E+01
nonconstant variance,
power restricted >1
exponential (M3)
1
0.17
10.00
0.00
124.91
1.6E+05
1.8E+02
nonconstant variance,
power restricted >1
exponential (M4)
1
0.17
2.09
0.15
117.00
error
error
nonconstant variance,
power restricted >1
exponential (M5)
0
0.17
1.52
N/A
118.43
error
error
nonconstant variance,
power restricted >1
Hill
0
0.17
1.51
NA
118.43
error
error
nonconstant variance, n
restricted >1
linear
2
0.17
6.66
0.04
119.57
2.5E+02
1.1E+02
nonconstant variance
polynomial
1
0.17
0.00
1.00
114.91
6.2E+01
2.7E+01
nonconstant variance
power
2
0.17
6.66
0.04
119.57
2.5E+02
1.1E+02
nonconstant variance,
power restricted >1, bound
hit
exponential (M2)
C
2
0.17
2.79
0.25
117.56
1.7E+02
5.0E+01
constant variance, power
restricted >1
exponential (M3)
2
0.17
2.79
0.25
117.56
1.7E+02
5.0E+01
constant variance, power
restricted >1
exponential (M4)
1
0.17
0.67
0.41
117.44
4.7E+01
1.7E-01
constant variance, power
restricted >1
exponential (M5)
0
0.17
0.15
N/A
118.92
3.2E+01
4.0E-05
constant variance, power
restricted >1
Hill
0
0.17
0.15
NA
118.92
2.3E+01
6.7E-06
constant variance, n
restricted >1
linear
2
0.17
3.32
0.19
118.09
2.1E+02
1.1E+02
constant variance
polynomial
2
0.17
3.32
0.19
118.09
2.1E+02
1.1E+02
constant variance
power
2
0.17
3.32
0.19
118.09
2.1E+02
1.1E+02
constant variance, power
restricted >1, bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-449 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.28.2. Figure for Selected Model: Exponential (M2), Constant Variance, Power Restricted
>1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
20
15
10
5
BMDL
BMD
0
0
20
40
60
80
100
120
140
160
180
dose
15:18 10/06 2009
E.3.28.3. Output File for Selected Model: Exponential (M2), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\ExpConstVar_BMRl_FRl0_run_opp.(d)
Gnuplot Plotting File:
Tue Oct 06 15:18:28 2009
Table 3
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-450 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
Specified
lnalpha 3.5321
rho(S) 0
a 13.9545
b 0.0143568
c 0.392432
d 1
Parameter Estimates
Variable Model 2
lnalpha 3.53824
rho 0
a 13.29
b 0.0376253
c 0.483301
d 3.66691
Table of Stats From Input Data
use N Obs Mean Obs Std Dev
0 7 13.29 8.65
20 4 11.25 5.56
60 6 5.75 3.53
180 7 7 6.01
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 11.94 6.197 0.5745
20 10.93 6.197 0.1025
60 9.158 6.197 -1.347
180 5.385 6.197 0.6897
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-451 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -54.38526 5 118.7705
A2 -51.88568 8 119.7714
A3 -54.38526 5 118.7705
R -57.45429 2 118.9086
2 -55.77871 3 117.5574
Additive constant for all log-likelihoods = -22.05. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test
1:
Test
2 :
Test
3:
Test
4 :
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Test
Test 1
Test 2
Test 3
Test 4
Tests of Interest
-2*log(Likelihood Ratio)
11.14
4.999
4.999
2.787
p-value
0.08423
0.1719
0.1719
0.2482
The p-value for Test 1 is greater than .05. There may not be a
diffence between responses and/or variances among the dose levels
Modelling the data with a dose/response curve may not be appropriate.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is greater than .1. Model 2 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 165.284
BMDL = 5 0.24 88
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-452 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.29. Markowski et al. (2001): FR2 Revolutions
2 E.3.29.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.11
5.71
0.06
216.09
3.6E+02
1.1E+02
nonconstant variance,
power restricted >1
exponential (M3)
2
0.11
5.71
0.06
216.09
3.6E+02
1.1E+02
nonconstant variance,
power restricted >1
exponential (M4)
1
0.11
1.94
0.16
214.33
error
error
nonconstant variance,
power restricted >1
exponential (M5)
0
0.11
0.45
N/A
214.83
error
error
nonconstant variance,
power restricted >1
Hill
1
0.11
0.45
0.50
212.83
error
error
nonconstant variance, n
restricted >1, bound hit
linear
2
0.11
6.08
0.05
216.46
3.3E+02
1.4E+02
nonconstant variance
polynomial
2
0.11
6.08
0.05
216.46
3.3E+02
1.4E+02
nonconstant variance
power
2
0.11
6.08
0.05
216.46
3.3E+02
1.4E+02
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
C
2
0.11
3.31
0.19
217.64
1.6E+02
5.8E+01
constant variance,
power restricted >1
exponential (M3)
2
0.11
3.31
0.19
217.64
1.6E+02
5.8E+01
constant variance, power
restricted >1
exponential (M4)
1
0.11
1.08
0.30
217.41
4.7E+01
2.0E-01
constant variance, power
restricted >1
exponential (M5)
0
0.11
0.20
N/A
218.53
3.3E+01
1.2E+01
constant variance, power
restricted >1
Hill
0
0.11
0.20
NA
218.53
2.4E+01
7.3E+00
constant variance, n
restricted >1, bound hit
linear
2
0.11
3.80
0.15
218.13
2.0E+02
1.0E+02
constant variance
polynomial
2
0.11
3.80
0.15
218.13
2.0E+02
1.0E+02
constant variance
power
2
0.11
3.80
0.15
218.13
2.0E+02
1.0E+02
constant variance, power
restricted >1, bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-453 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.29.2. Figure for Selected Model: Exponential (M2), Constant Variance, Power Restricted
>1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
200
150
100
50
0
BMDL
BMD
0
20
40
60
80
100
120
140
160
180
dose
15:18 10/06 2009
E.3.29.3. Output File for Selected Model: Exponential (M2), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\ExpConstVar_BMRl_FR2_revolutions.(d)
Gnuplot Plotting File:
Tue Oct 06 15:18:54 2009
Table 3
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-454 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 2
lnalpha
rho(S)
7.68046
0
125.255
0.0134965
0.429602
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 2
7.68885
0
119.29
0.0345516
0.526177
4.19941
Table of Stats From Input Data
Obs Mean
Obs Std Dev
0
20
60
180
119.3
108 . 5
56.5
68 .14
6 9.9
61
31. 21
33.23
0
20
60
180
Estimated Values of Interest
Est Mean Est Std Scaled Residual
108 . 9
101
86. 93
55.35
49. 85
49. 85
49. 85
49. 85
0.5497
0.2994
-1.495
0. 6786
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-455 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A3:
Yij
Var{e(ij;
Mu(i) + e(ij)
exp(lalpha + log(mean(
rho)
Model R: Yij
Var{e(ij)}
Mu + e(i
Sigma""" 2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -104.1655 5 218.331
A2 -101.1402 8 218.2803
A3 -104.1655 5 218.331
R -107.5993 2 219.1985
2 -105.8179 3 217.6357
Additive constant for all log-likelihoods = -22.05. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 12.92 6 0.04435
Test 2 6.051 3 0.1092
Test 3 6.051 3 0.1092
Test 4 3.305 2 0.1916
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is greater than .1. Model 2 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 162.682
BMDL = 58.0677
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-456 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.30. Markowski et al. (2001): FR5 Run Opp
2 E.3.30.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
2
0.23
5.55
0.06
134.53
1.4E+02
5.3E+01
nonconstant variance,
power restricted >1
exponential (M3)
2
0.23
5.55
0.06
134.53
1.4E+02
5.3E+01
nonconstant variance,
power restricted >1
exponential (M4)
1
0.23
1.05
0.31
132.03
3.6E+01
1.2E+01
nonconstant variance,
power restricted >1
exponential (M5)
0
0.23
0.08
N/A
133.06
2.8E+01
1.4E+01
nonconstant variance,
power restricted >1
Hill
0
0.23
0.08
NA
133.06
2.2E+01
error
nonconstant variance,
n restricted >1
linear
2
0.23
6.35
0.04
135.33
1.8E+02
9.3E+01
nonconstant variance
polynomial
1
0.23
0.06
0.81
131.04
4.0E+01
2.2E+01
nonconstant variance
power
2
0.23
6.35
0.04
135.33
1.8E+02
9.3E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
2
0.23
3.80
0.15
133.83
9.5E+01
4.3E+01
constant variance,
power restricted >1
exponential (M3)
2
0.23
3.80
0.15
133.83
9.5E+01
4.3E+01
constant variance,
power restricted >1
exponential (M4)
1
0.23
1.06
0.30
133.09
3.0E+01
9.4E+00
constant variance,
power restricted >1
exponential (M5)
0
0.23
0.01
N/A
134.03
2.9E+01
1.2E+01
constant variance,
power restricted >1
Hillc
1
0.23
0.01
0.94
132.03
2.2E+01
1.1E+01
constant variance, n
restricted >1, bound
hit
linear
2
0.23
4.80
0.09
134.82
1.3E+02
8.1E+01
constant variance
polynomial
1
0.23
0.36
0.55
132.39
3.1E+01
1.8E+01
constant variance
power
2
0.23
4.80
0.09
134.82
1.3E+02
8.1E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-457 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.30.2. Figure for Selected Model: Hill, Constant Variance, n Restricted >1, Bound Hit
Hill Model with 0.95 Confidence Level
¦ i ¦ ¦' ¦ ¦ ¦ ¦' M ¦ ¦'' ¦ ¦ ¦' M ¦ ¦ ¦' ¦ ¦ ¦ ¦ M ¦
Hill
i
BMDL BMD
0 20 40
15:22 10/06 2009
60 80 100 120 140 160 180
dose
E.3.30.3. Output File for Selected Model: Hill, Constant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_FR5_run_opp.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_FR5_run_opp.plt
Tue Oct 06 15:22:42 2009
Table 3
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-458 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha
rho
intercept
77.4849
0
26.14
-13.34
2 . 36002
35.0654
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
-3.6e-009
9 . 8e-009
3. 6e-008
intercept
-3.6e-009
1
-0. 81
-0.51
v
9 . 8e-009
-0. 81
1
0.36
k
3. 6e-008
-0.51
0.36
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
intercept
k
Estimate
64 . 5863
26.14
-13.1569
18
21.5963
Std. Err.
18 . 6445
3. 03753
3.7676
NA
2 . 68136
Lower Conf. Limit
28.0438
20.1865
-20.5413
16.3409
Upper Conf. Limit
101.129
32 .0935
-5.77257
26.8517
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
20
60
180
26.1
23.5
12 . 8
13.1
26.1
23.5
13
13
12 . 3
7 . 04
6.17
7 .14
5. 04
J. 04
J. 04
]. 04
1.02e-008
-1.39e-007
-0.0558
0.0517
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-459 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
Model R: Yi = Mu + e(i
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -62.013133 5 134.026266
A2 -59.839035 8 135.678070
A3 -62.013133 5 134.026266
fitted -62.016024 4 132.032049
R -67.530040 2 139.060081
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio)
Test df
p-value
Test 1
Test 2
Test 3
Test 4
15.382
4.3482
4.3482
0.0057833
0.01748
0.2262
0.2262
0.9394
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 22.14 4
BMDL = 11.165
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-460 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.31. Mietinnin et al. (2006): Caries
2 E.3.31.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
3
3.32
0.34
162.70
3.7E+01
2.0E+01
power restricted >1, bound
hit
gamma
3
3.32
0.34
162.70
7.5E+01
4.1E+01
power restricted >1, bound
hit
logistic
3
3.54
0.32
162.91
4.4E+01
2.6E+01
logistic
3
3.54
0.32
162.91
9.0E+01
5.2E+01
log-logistic b
3
2.33
0.51
161.77
1.5E+01
5.0E+00
slope restricted >1,
bound hit
log-logistic
3
2.33
0.51
161.77
3.1E+01
1.1E+01
slope restricted >1, bound
hit
log-probit
2
0.64
0.73
161.99
1.2E-01
error
slope restricted >1
log-probit
2
0.64
0.73
161.99
5.1E-01
error
slope restricted >1
multistage, 2-
degree
3
3.32
0.34
162.70
3.7E+01
2.0E+01
betas restricted >0, bound
hit
multistage, 2-
degree
3
3.32
0.34
162.70
7.5E+01
4.1E+01
betas restricted >0, bound
hit
probit
3
3.67
0.30
163.03
4.9E+01
3.1E+01
probit
3
3.67
0.30
163.03
9.9E+01
6.2E+01
Weibull
3
3.32
0.34
162.70
3.7E+01
2.0E+01
power restricted >1, bound
hit
Weibull
3
3.32
0.34
162.70
7.5E+01
4.1E+01
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-461 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.31.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
0.9
0.7
0.6
0.5
0.4 BMDL
Log-Logistic
f
BMD
200
400 600
dose
800
1000
15:23 10/06 2009
E.3.31.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\LogLogistic_BMRl_Caries.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\LogLogistic_BMRl_Caries.plt
Tue Oct 06 15:23:23 2009
Table 2 converting the percentage into the number of animals, and control is Control II from the
study. Dose is in ng per kg and is from Table 1
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope <= 1
Total number of observations = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-462 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Default Initial Parameter Values
background = 0.595238
intercept = -5.52519
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.64
intercept -0.64 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.658158 * * *
intercept -5.64068 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -77.6769 5
Fitted model -78.8837 2 2.41374 3 0.4911
Reduced model -83.2067 1 11.0597 4 0.0259
AIC: 161.767
Dose
Est
. Prob.
Goodness of Fit
Expected Observed
Size
Scale
Residu
0.0000
0.
6582
27 . 643
25.000
42
-0.860
30.0000
0.
6911
20.041
23.000
29
1.189
100.0000
0.
7477
18.693
19.000
25
0.141
300.0000
0.
8345
20.027
20.000
24
-0.015
1000.0000
0.
924 9
29.596
29.000
32
-0.400
Chi/S2 = 2.33
d.f. =
3 P
-value = 0.5062
Benchmark Dose Computation
Specified effect = 0.05
Risk Type = Extra risk
Confidence level = 0.95
BMD = 14.824
BMDL = 4.9904 4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-463 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 E.3.32. National Toxicology Program (1982): Male Mice, Toxic Hepatitis
3 E.3.32.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
1
4.93
0.03
113.10
1.6E+01
5.2E+00
power restricted >1
logistic b
2
4.76
0.09
110.71
1.8E+01
1.4E+01
log-logistic
1
4.93
0.03
113.09
1.5E+01
6.6E+00
slope restricted >1
log-probit
1
4.89
0.03
113.11
1.4E+01
7.2E+00
slope restricted >1
multistage, 2-
degree
1
5.18
0.02
112.86
1.2E+01
4.6E+00
betas restricted >0
probit
2
4.86
0.09
110.70
1.6E+01
1.3E+01
Weibull
1
4.99
0.03
113.06
1.6E+01
4.9E+00
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
E.3.32.2. Figure for Selected Model: Logistic
Logistic Model with 0.95 Confidence Level
0.6
0.4
0.2
Logistic
BMDL
12:25 11/04 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-464 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.32.3. Output file for Selected Model: Logistic
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\Logistic_BMR2_Toxic_hepatitis.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\Logistic_BMR2_Toxic_hepatitis.plt
Wed Nov 04 12:25:18 2009
The form of the probability function is:
P[response] = 1/[1+EXP(-intercept-slope^dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
background = 0 Specified
intercept = -3.05581
slope = 0.0703319
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.66
slope -0.66 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
intercept -3.08708 0.358526 -3.78978 -2.38438
slope 0.07156 0.00813416 0.0556174 0.0875027
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -51.0633 4
Fitted model -53.3562 2 4.58581 2 0.101
Reduced model -121.743 1 141.358 3 <.0001
AIC: 110.712
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-465 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Goodness of Fit
Scaled
Dose
Est. Prob.
Expected
Observed
Size
Residu.
0.0000
0. 0436
3.186
1. 000
73
-1.252
1.4000
0.0480
2 . 353
5. 000
49
1.769
7.1000
0.0705
3. 455
3. 000
49
-0.254
71.0000
0.8801
44.007
44.000
50
-0.003
Chi/S2 = 4.76 d.f. = 2 P-value = 0.0925
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 17.68 85
BMDL = 13.827 2
E.3.33. National Toxicology Program (2006): Alveolar Metaplasia
E.3.33.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X Test
Statistic
X2P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
34.09
0.00
340.13
2.2E+00
1.8E+00
power restricted >1, bound
hit
logistic
4
45.56
0.00
358.35
5.0E+00
4.1E+00
log-logistic
4
3.98
0.41
312.97
6.6E-01
5.0E-01
slope restricted >1, bound
hit
log-probitb
3
1.31
0.73
312.54
3.3E-01
9.0E-02
slope restricted >1
multistage, 2-
degree
4
34.09
0.00
340.13
2.2E+00
1.8E+00
betas restricted >0, bound
hit
probit
4
46.73
0.00
362.18
5.7E+00
4.8E+00
Weibull
4
34.09
0.00
340.13
2.2E+00
1.8E+00
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-466 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.33.2. Figure for Selected Model: Log-Probit, Slope Restricted >1
LogProbit Model with 0.95 Confidence Level
LogProbit
E.3.33.3. Output File for Selected Model: Log-Probit, Slope Restricted >1
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\LogProbit_BMR2_Alveolar_metaplasia.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\LogProbit_BMR2_Alveolar_metaplasia.plt
Wed Nov 04 12:26:52 2009
0
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope*Log(Dose)),
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-467 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
User has chosen the log transformed model
Default Initial (and Specified) Parameter Values
background = 0.0377358
intercept = -0.759264
slope = 0.469642
Asymptotic Correlation Matrix of Parameter Estimates
background
intercept
slope
background
1
-0.24
0.12
intercept
-0.24
1
-0. 9
slope
0.12
-0. 9
1
Parameter Estimates
Variable
background
intercept
slope
Estimate
0.0374101
-0.761678
0.471021
Std. Err.
0.0259232
0.210613
0.0755121
95.0% Wald Confidence Interval
Lower Conf. Limit
-0.0133985
-1.17447
0.32302
Upper Conf. Limit
0.0882186
-0.348885
0. 619022
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Param's Deviance Test d.f.
Log(likelihood)
-152.615
-153.271
-216.802
312.543
1.31226
128.374
P-value
0.7262
<.0001
Goodness of Fit
Est. Prob.
Expected
Observed
Scaled
Residual
0.0000
0
0374
1. 983
2 . 000
53
0. 012
2.1400
0
3679
19.868
19.000
54
-0.245
7.1400
0
5815
30.819
33.000
53
0. 607
15.7000
0
7149
37 .174
35.000
52
-0.668
32.9000
0
8187
43.389
45.000
53
0.574
71.4000
0
8981
46.701
46.000
52
-0.321
Chi'" 2
1. 31
d.f.
P-value
0 .7272
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
0.1
Extra risk
0. 95
0.331636
0.0896842
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-468 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.34. National Toxicology Program (2006): Gingival Hyperplasia Squamous, 2 Years
2 E.3.34.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
1 P"
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
12.82
0.01
318.87
2.3E+01
1.4E+01
power restricted >1, bound
hit
logistic
4
13.78
0.01
320.91
3.6E+01
2.6E+01
log-logistic b
4
12.38
0.01
317.97
1.8E+01
1.0E+01
slope restricted >1,
bound hit
log-logisticc
3
1.53
0.68
307.42
3.7E-01
1.5E-07
slope unrestricted
log-probit
3
1.47
0.69
307.35
4.7E-01
8.9E-07
slope restricted >1
multistage, 1-
degree
4
12.82
0.01
318.87
2.3E+01
1.4E+01
betas restricted >0, bound
hit
probit
4
13.67
0.01
320.69
3.4E+01
2.4E+01
Weibull
4
12.82
0.01
318.87
2.3E+01
1.4E+01
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
3
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-469 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.3.34.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
0.4
0.3
0.2
0.1
0
BMDL
BMD
0
10
20
30
40
50
60
70
dose
11:44 11/29 2009
E.3.34.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\LogLogistic_BMR2_Ging_Hyp_2yr.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\LogLogistic_BMR2_Ging_Hyp_2yr.plt
Sun Nov 29 11:44:28 2009
[insert study notes]
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope >= 1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-470 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Default Initial Parameter Values
background = 0.0188679
intercept = -4.5509
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.71
intercept -0.71 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.117717 * * *
intercept -5.10866 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-149.95
-156.985
-162.631
Param's Deviance Test d.f.
14.0696
25.3627
P-value
0. 007076
0.0001186
AIC:
317 . 969
Est. Prob.
Goodness of Fit
Expected Observed
Scaled
Residual
0.0000
0.1177
6.239
1
000
53
-2 . 233
2.1400
0.1290
6. 965
7
000
54
0. 014
7.1400
0.1542
8.174
14
000
53
2 . 216
15.7000
0.1942
10.292
13
000
53
0. 940
32.9000
0.2641
13.995
15
000
53
0.313
71.4000
0.3837
20.335
16
000
53
-1. 225
Chi'" 2 = 12.38 d.f. = 4 P-value = 0.0147
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 18.3832
BMDL = 10.4359
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-471 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
E.3.34.4. Figure for Unrestricted Model: Log-Logistic, Slope Unrestricted
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
0.4
0.3
0.2
10
20
30
40
50
60
70
dose
11:44 11/29 2009
E.3.34.5. Output File for Unrestricted Model: Log-Logistic, Slope Unrestricted
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\LogLogistic_Unrest_BMR2_Ging_Hyp_2yr.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\LogLogistic_Unrest_BMR2_Ging_Hyp_2yr.plt
Sun Nov 29 11:44:31 2009
[insert study notes]
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-472 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Default Initial Parameter Values
background = 0.0188679
intercept = -2.04571
slope = 0.299277
Asymptotic Correlation Matrix of Parameter Estimates
background
intercept
slope
background
1
-0.3
0.12
intercept
-0.3
1
-0. 91
slope
0.12
-0. 91
1
Variable
background
intercept
slope
Parameter Estimates
Estimate
0.0185126
-1.93464
0.264795
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-149.95
-150.708
-162.631
Param's Deviance Test d.f.
1.5163
25.3627
P-value
0.6785
0.0001186
307.416
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0185
0. 981
1
000
53
0
019
2.1400
0.1659
8 . 959
7
000
54
-0
717
7.1400
0.2105
11.155
14
000
53
0
959
15.7000
0.2447
12 . 972
13
000
53
0
009
32.9000
0.2806
14 . 873
15
000
53
0
039
71.4000
0.3219
17 . 059
16
000
53
-0
311
Chi'A2 = 1.53 d.f. = 3 P-value = 0.6750
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 0.370958
BMDL = 1.504 94e-007
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-473 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.35. National Toxicology Program (2006): Heart, Cardiomyopathy
2 E.3.35.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
6.23
0.18
398.01
6.5E+00
4.8E+00
power restricted >1, bound
hit
logistic
4
10.70
0.03
402.78
1.1E+01
9.2E+00
log-logistic b
4
2.42
0.66
394.22
3.7E+00
2.5E+00
slope restricted >1,
bound hit
log-probit
3
0.93
0.82
394.80
2.1E+00
5.1E-01
slope restricted >1
multistage, 2-
degree
4
6.23
0.18
398.01
6.5E+00
4.8E+00
betas restricted >0, bound
hit
probit
4
10.72
0.03
402.80
1.1E+01
9.3E+00
Weibull
4
6.23
0.18
398.01
6.5E+00
4.8E+00
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
3
4
5 E.3.35.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 10 20 30 40 50 60 70
dose
6 11:58 11/11 2009
7
Log-Logistic
BMDL BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-474 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.35.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\LogLogistic_BMR2_Cardiomyopathy.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\LogLogistic_BMR2_Cardiomyopathy.plt
Wed Nov 11 11:58:41 2009
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose)
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope <= 1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial Parameter Values
background = 0.188679
intercept = -3.47661
slope = 1
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
background intercept
background 1 -0.65
intercept -0.65 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.20346 * * *
intercept -3.50681 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -193.93 6
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-475 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
Fitted model -195.111 2 2.36161 4 0.6696
Reduced model -216.802 1 45.7449 5 <.0001
AIC: 394.221
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.2035
10.783
10.000
53
-0
267
2.1400
0.2515
13.581
12.000
54
-0
496
7.1400
0.3440
18.229
22.000
53
1
090
15.7000
0.4585
23.840
25.000
52
0
323
32.9000
0.5991
31.751
32.000
53
0
070
71.4000
0.7464
38.815
36.000
52
-0
897
Chi^2 = 2.42 d.f. = 4 P-value = 0.6589
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 3.70462
BMDL = 2.50223
E.3.36. National Toxicology Program (2006): Hepatocyte Hypertrophy, 2 Years
E.3.36.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X Test
Statistic
X2P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
26.48
0.00
290.37
1.6E+00
1.3E+00
power restricted >1, bound
hit
logistic
4
35.54
0.00
310.49
4.3E+00
3.6E+00
log-logistic b
5
15.18
0.01
278.08
7.0E-01
5.5E-01
slope restricted >1,
bound hit
log-probit
4
14.46
0.01
279.20
7.2E-01
3.3E-01
slope restricted >1
multistage, 2-
degree
4
26.48
0.00
290.37
1.6E+00
1.3E+00
betas restricted >0, bound
hit
probit
4
41.23
0.00
313.84
4.6E+00
3.9E+00
Weibull
4
26.48
0.00
290.37
1.6E+00
1.3E+00
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-476 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.36.2. Figure for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Log-Logistic Model with 0.95 Confidence Level
1
0.8
2 0.6
o
5
c
o
"G
E 0.4
0.2
0
E
0 10 20 30 40 50 60 70
dose
18:49 10/06 2009
E.3.36.3. Output File for Selected Model: Log-Logistic, Slope Restricted >1, Bound Hit
Logistic Model. (Version: 2.12; Date: 05/16/2008)
Input Data File:
C:\USEPA\BMDS21\AniDose\LogLogistic_BMR2_Hepatocyte_hypertrophy_2years.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AniDose\LogLogistic_BMR2_Hepatocyte_hypertrophy_2years.pit
Tue Oct 06 18:49:36 2009
[insert study notes]
The form of the probability function is:
P[response] = background+(1-background)/[1+EXP(-intercept-slope*Log(dose))]
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is restricted as slope <= 1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Log-Logistic
MDLBMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-477 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
User has chosen the log transformed model
Default Initial Parameter Values
background = 0
intercept = -2.2119
slope = 1.23746
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -background -slope
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
intercept
intercept 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0 * * *
intercept -1.83737 * * *
slope 1 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -129.986 6
Fitted model -138.041 1 16.1104 5 0.00653£
Reduced model -219.97 1 179.968 5 <.0001
AIC: 278.082
Goodness of Fit
Scaled
Est. Prob. Expected Observed Size Residual
0.0000
0
0000
0. 000
0. 000
53
0. 000
2.1400
0
2542
13.724
19.000
54
1.649
7.1400
0
5320
28.198
19.000
53
-2.532
15.7000
0
7143
37.857
42.000
53
1.260
32.9000
0
8397
44.505
41.000
53
-1.312
71.4000
0
9192
48.715
52.000
53
1. 655
Chi'" 2 = 15.18 d.f. = 5 P-value = 0.0096
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 0.697776
BMDL = 0.545416
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-478 DRAFT—DO NOT CITE OR QUOTE
-------
E.3.37. National Toxicology Program (2006): Liver, Eosinophilic Focus, Multiple
E.3.37.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
4.30
0.37
330.46
5.7E+00
4.5E+00
power restricted >1, bound
hit
logistic
4
6.46
0.17
333.34
1.3E+01
1.1E+01
log-logistic
3
5.90
0.12
334.15
4.7E+00
2.9E+00
slope restricted >1
log-probit
3
6.58
0.09
334.85
4.8E+00
1.8E+00
slope restricted >1
multistage, 2-
degree
3
4.18
0.24
332.36
6.2E+00
4.5E+00
betas restricted >0
probit
4
6.16
0.19
332.96
1.2E+01
1.0E+01
Weibull"
4
4.30
0.37
330.46
5.7E+00
4.5E+00
power restricted >1,
bound hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
E.3.37.2. Figure for Selected Model: WeibuII, Power Restricted >1, Bound Hit
Weibull Model with 0.95 Confidence Level
0.6
"O
0
0
£
<
1 0,
0.2
Weibull
BMDL BMD
10 20 30 40 50 60 70
dose
11:44 11/11 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-479 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.37.3. Output File for Selected Model: Weibull, Power Restricted >1, Bound Hit
Weibull Model using Weibull Model (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Weibull_BMR2_liver_eosin_focus.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Weibull_BMR2_liver_eosin_focus.plt
Wed Nov 11 11:44:27 2009
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(-slope*doseApower)
Dependent variable = DichEff
Independent variable = Dose
Power parameter is restricted as power >=1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial (and Specified) Parameter Values
Background = 0.0648148
Slope = 0.00246576
Power = 1.4 9873
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Power
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Background Slope
Background 1 -0.49
Slope -0.49 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.0893152 0.0297295 0.0310464 0.147584
Slope 0.0185641 0.00270697 0.0132586 0.0238697
Power 1 NA
NA - Indicates that this parameter has hit a bound
implied by some ineguality constraint and thus
has no standard error.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -161.07 6
Fitted model -163.229 2 4.31726 4 0.3648
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-480 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Reduced model -202.816 1 83.4925 5 <.0001
AIC: 330.457
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0893
4 .734
3
000
53
-0.835
2.1400
0.1248
6.738
8
000
54
0.520
7.1400
0.2024
10.725
14
000
53
1.120
15.7000
0.3196
16.937
17
000
53
0. 019
32.9000
0.5056
26.794
22
000
53
-1.317
71.4000
0.7581
40.177
42
000
53
0.585
Chi^2 = 4.30 d.f. = 4 P-value = 0.3672
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 5.67549
BMDL = 4.5323
E.3.38. National Toxicology Program (2006): Liver, Fatty Change, Diffuse
E.3.38.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
2.37
0.67
252.29
4.2E+00
3.2E+00
power restricted >1
logistic
4
15.06
0.00
269.83
1.1E+01
9.3E+00
log-logistic
4
4.96
0.29
255.08
4.7E+00
3.2E+00
slope restricted >1
log-probit
4
5.05
0.28
255.26
4.6E+00
3.2E+00
slope restricted >1
multistage, 2-
degree b
4
2.03
0.73
251.93
4.2E+00
3.2E+00
betas restricted >0
probit
4
14.92
0.00
269.43
1.1E+01
9.1E+00
Weibull
4
2.31
0.68
252.22
4.3E+00
3.2E+00
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-481 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.39. Figure for Selected Model: Multistage, 2-Degree, Betas Restricted >0
Multistage Model with 0.95 Confidence Level
Multistage
.8
0.6
0.4
0.2
0
BMDL BMD
0
10
20
30
40
50
60
70
dose
11:44 11/11 2009
E.3.40. Output File for Selected Model: Multistage, 2-Degree, Betas Restricted >0
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Multistage_BMR2_liver_fatty_change_diff.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Multistage_BMR2_liver_fatty_change_diff.plt
Wed Nov 11 11:44:50 2009
NTP_liver_fatty_change_di ffuse
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2 ) ]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
Maximum number of iterations = 250
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-482 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.02888
Beta(1) = 0.0193083
Beta(2) = 0.000185869
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1) Beta(2)
Beta(1) 1 -0.89
Beta(2) -0.89 1
Parameter Estimates
Variable
Background
Beta(1)
Beta (2)
Estimate
0
0.0248561
9.42857e-005
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-122.992
-123.966
-204.846
Param's Deviance Test d.f.
1.94705
163.708
P-value
0.7455
<.0001
AIC:
251.932
Goodness of Fit
Scaled
Dose
Est
. Prob.
Expected
Observed
Size
Residual
0.0000
0.
0000
0. 000
0
000
53
0
000
2.1400
0.
0522
2.819
2
000
54
-0
501
7.1400
0.
1666
8 831
12
000
53
1
168
15.7000
0.
3387
17 94 9
17
000
53
-0
275
32.9000
0.
6014
31 875
30
000
53
-0
526
71.4000
0.
8952
47.444
48
000
53
0
249
Chi ^2
2 . 03
d.f.
P-value
0.7302
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 4.17277
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-483 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 BMDL = 3.20452
3
4 BMDU = 5.73352
5
6 Taken together, (3.20452, 5.73352) is a 90 % two-sided confidence
7 interval for the BMD
8
9
10 E.3.41. National Toxicology Program (2006): Liver Necrosis
11 E.3.41.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
2
1 P"
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
1.85
0.76
235.58
2.0E+01
1.4E+01
power restricted >1, bound
hit
logistic
4
4.07
0.40
238.31
3.5E+01
2.8E+01
log-logistic
4
1.60
0.81
235.27
1.8E+01
1.2E+01
slope restricted >1, bound
hit
log-probitb
3
1.14
0.77
236.74
1.1E+01
3.5E+00
slope restricted >1
multistage, 2-
degree
4
1.85
0.76
235.58
2.0E+01
1.4E+01
betas restricted >0, bound
hit
probit
4
3.72
0.45
237.89
3.3E+01
2.6E+01
Weibull
4
1.85
0.76
235.58
2.0E+01
1.4E+01
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
12
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-484 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
E.3.41.2. Figure for Selected Model: LogProbit
LogM obt rvoflal ^'omctenca Lerol
nr,
LOCjl "Dbt
0.-1
IIH
u/j
y''
0 1
¦¦¦
y
u
~via.
~IVD
J 1U m M W ijU UJ r'U
dose
12:21 11/20 2010
E.3.41.3. Output File for Selected Model: Log-Probit
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov29\LogProbit_BMR2_liver_necrosis.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 9\LogProbit_BMR2_liver_necrosis.plt
Sun Nov 29 12:24:51 2009
NTP liver necrosis
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope*Log(Dose)),
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-485 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
Default Initial (and Specified) Parameter Values
background = 0.0188679
intercept = -1.98094
slope = 0.316942
Asymptotic Correlation Matrix of Parameter Estimates
background
intercept
slope
background
1
-0.69
0.59
intercept
-0.69
1
-0. 97
slope
0.59
-0. 97
1
Parameter Estimates
Variable
background
intercept
slope
Estimate
0.0228339
-2 .14844
0.367034
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
0.0230818 -0.0224057 0.0680734
0.527256 -3.18184 -1.11503
0.139055 0.0944904 0.639577
Model
Full model
Fitted model
Reduced model
Analysis of Deviance Table
Log(likelihood)
-114 . 813
-115.371
-127.98
Param's Deviance Test d.f.
1.1157
26.3331
P-value
0 .7733
C.0001
236.742
Goodness of Fit
Scaled
Dose
Est
. Prob.
Expected
Observed
Size
Residual
0.0000
0.
0228
1 210
1
000
53
-0
193
2.1400
0.
0529
2 858
4
000
54
0
694
7.1400
0.
0979
5.187
4
000
53
-0
549
15.7000
0.
1475
7.819
8
000
53
0
070
32.9000
0.
2116
11 215
10
000
53
-0
409
71.4000
0.
2968
15.729
17
000
53
0
382
Chi ^2
1.14
d.f.
P-value
0.7678
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 10.6107
BMDL = 3.4 97 91
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-486 DRAFT—DO NOT CITE OR QUOTE
-------
E.3.42. National Toxicology Program (2006): Liver, Pigmentation
E.3.42.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
3
3.05
0.38
197.66
1.5E+00
8.1E-01
power restricted >1
logistic
4
29.12
0.00
203.52
2.3E+00
1.9E+00
log-logistic
3
0.19
0.98
195.60
2.2E+00
1.5E+00
slope restricted >1
log-probitb
3
0.18
0.98
195.45
2.1E+00
1.4E+00
slope restricted >1
multistage, 2-
degree
3
4.53
0.21
199.85
9.4E-01
7.1E-01
betas restricted >0
probit
4
131.22
0.00
210.31
2.3E+00
1.9E+00
Weibull
3
3.75
0.29
198.49
1.3E+00
7.5E-01
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
E.3.42.2. Figure for Selected Model: Log-Probit, Slope Restricted >1
LogProbit Model with 0.95 Confidence Level
LogProbit
0.6
0.4
0.2
3MDL BMD
10
20
30 40
dose
50
60
70
11:59 11/11 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-487 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.42.3. Output File for Selected Model: Log-Probit, Slope Restricted >1
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\LogProbit_BMR2_Pigmentation.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\LogProbit_BMR2_Pigmentation.plt
Wed Nov 11 11:59:31 2009
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope^Log(Dose)
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
User has chosen the log transformed model
Default Initial (and Specified) Parameter Values
background = 0.0754717
intercept = -1.91144
slope = 1.07 385
Asymptotic Correlation Matrix of Parameter Estimates
background intercept slope
background 1 -0.45 0.35
intercept -0.45 1 -0.94
slope 0.35 -0.94 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0.0735956 0.0343284 0.00631316 0.140878
intercept -2.19294 0.400053 -2.97703 -1.40885
slope 1.25068 0.169731 0.918012 1.58335
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -94.6177 6
Fitted model -94.7248 3 0.214232 3 0.9753
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-488 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Reduced model -210.717 1 232.198 5 <.0001
AIC: 195.45
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0
0736
3. 901
4
000
53
0
052
2.1400
0
1729
9.338
9
000
54
-0
122
7.1400
0
6338
33.591
34
000
53
0
117
15.7000
0
9023
47.822
48
000
53
0
082
32.9000
0
9863
52 . 275
52
000
53
-0
325
71.4000
0
9992
52.959
53
000
53
0
202
Chi^2 = 0.18 d.f. = 3 P-value = 0.9801
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2.07241
BMDL = 1.39932
E.3.43. National Toxicology Program (2006): Lung, Alveolar to Bronchiolar Epithelial
Metaplasia (Alveolar Epithelium, Metaplasia, Bronchiolar)
E.3.43.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
4
34.09
0.00
340.13
2.2E+00
1.8E+00
power restricted >1, bound
hit
logistic
4
45.56
0.00
358.35
5.0E+00
4.1E+00
log-logistic
4
3.98
0.41
312.97
6.6E-01
5.0E-01
slope restricted >1, bound
hit
log-probitb
3
1.31
0.73
312.54
3.3E-01
9.0E-02
slope restricted >1
multistage, 2-
degree
4
34.09
0.00
340.13
2.2E+00
1.8E+00
betas restricted >0, bound
hit
probit
4
46.73
0.00
362.18
5.7E+00
4.8E+00
Weibull
4
34.09
0.00
340.13
2.2E+00
1.8E+00
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-489 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.43.2. Figure for Selected Model: Log-Probit, Slope Restricted >1
LogProbit Model with 0.95 Confidence Level
LogProbit
E.3.43.3. Output File for Selected Model: Log-Probit, Slope Restricted >1
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\LogProbit_BMR2_Alv_bronch_epith_metapl.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\LogProbit_BMR2_Alv_bronch_epith_metapl.plt
Wed Nov 11 12:00:22 2009
0
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope*Log(Dose)),
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-490 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
User has chosen the log transformed model
Default Initial (and Specified) Parameter Values
background = 0.0377358
intercept = -0.759264
slope = 0.469642
Asymptotic Correlation Matrix of Parameter Estimates
background
intercept
slope
background
1
-0.24
0.12
intercept
-0.24
1
-0. 9
slope
0.12
-0. 9
1
Parameter Estimates
Variable
background
intercept
slope
Estimate
0.0374101
-0.761678
0.471021
Std. Err.
0.0259232
0.210613
0.0755121
95.0% Wald Confidence Interval
Lower Conf. Limit
-0.0133985
-1.17447
0.32302
Upper Conf. Limit
0.0882186
-0.348885
0. 619022
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Param's Deviance Test d.f.
Log(likelihood)
-152.615
-153.271
-216.802
312.543
1.31226
128.374
P-value
0.7262
<.0001
Goodness of Fit
Est. Prob.
Expected
Observed
Scaled
Residual
0.0000
0
0374
1. 983
2 . 000
53
0. 012
2.1400
0
3679
19.868
19.000
54
-0.245
7.1400
0
5815
30.819
33.000
53
0. 607
15.7000
0
7149
37 .174
35.000
52
-0.668
32.9000
0
8187
43.389
45.000
53
0.574
71.4000
0
8981
46.701
46.000
52
-0.321
Chi'" 2
1. 31
d.f.
P-value
0 .7272
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
0.1
Extra risk
0. 95
0.331636
0.0896842
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-491 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.44. National Toxicology Program (2006): Oval Cell Hyperplasia, 2 Years
2 E.3.44.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
3
7.00
0.07
199.45
9.0E+00
5.5E+00
power restricted >1
logistic
4
8.72
0.07
199.88
9.8E+00
8.2E+00
log-logistic
3
8.38
0.04
202.01
9.7E+00
7.2E+00
slope restricted >1
log-probit
3
7.12
0.07
200.42
1.0E+01
7.8E+00
slope restricted >1
multistage, 2-
degree b
4
6.33
0.18
195.33
5.8E+00
4.0E+00
betas restricted >0
probit
4
7.50
0.11
198.17
9.1E+00
7.7E+00
Weibull
3
6.92
0.07
198.69
7.7E+00
4.7E+00
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
3
4
5 E.3.44.2. Figure for Selected Model: Multistage, 2-Degree, Betas Restricted >0
Multistage Model with 0.95 Confidence Level
1
0.8
£ 0.6
o
5
c
o
"G
55 0.4
0.2
0
0 10 20 30 40 50 60 70
dose
6 11:59 11/11 2009
7
Multistage
BMDL BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-492 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.44.3. Output File for Selected Model: Multistage, 2-Degree, Betas Restricted >0
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AD\Multistage_BMR2_Oval_cell_hyperplasia.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Multistage_BMR2_Oval_cell_hyperplasia.plt
Wed Nov 11 11:59:06 2009
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/Nl-beta2*dose/N2 ) ]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0
Beta(2) = 1.98687e+016
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(1) Beta(2)
Beta(1) 1 -0.89
Beta(2) -0.89 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 0.0133632 * * *
Beta(2) 0.00083535 * * *
- Indicates that this value is not calculated.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-493 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -92.4898 6
Fitted model -95.6669 2 6.35417 4 0.1742
Reduced model -210.191 1 235.402 5 <.0001
AIC: 195.334
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0
0000
0. 000
0
000
53
0
000
2.1400
0
0319
1.723
4
000
54
1
763
7.1400
0
1289
6. 832
3
000
53
-1
571
15.7000
0
3401
18 . 027
20
000
53
0
572
32.9000
0
7392
39.175
38
000
53
-0
368
71.4000
0
9 9 4 6
52 .711
53
000
53
0
539
Chi/N2 = 6.33 d.f. = 4 P-value = 0.1759
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 5.78926
BMDL = 4.04553
BMDU = 9.63861
Taken together, (4.04553, 9.63861) is a 90 % two-sided confidence
interval for the BMD
E.3.45. National Toxicology Program (2006): Toxic Hepatopathy
E.3.45.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma b
4
1.80
0.77
185.63
4.7E+00
3.3E+00
power restricted >1
logistic
4
12.79
0.01
198.45
7.1E+00
5.9E+00
log-logistic
3
3.20
0.36
190.06
5.7E+00
4.0E+00
slope restricted >1
log-probit
3
3.09
0.38
189.86
6.1E+00
4.1E+00
slope restricted >1
multistage, 2-
degree
4
2.89
0.58
186.52
4.2E+00
2.7E+00
betas restricted >0
multistage, 2-
degree
4
2.89
0.58
186.52
4.2E+00
2.7E+00
betas unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-494 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
multistage, 1-
degree
5
10.28
0.07
194.94
2.1E+00
1.8E+00
betas unrestricted
probit
4
11.78
0.02
197.16
6.8E+00
5.7E+00
Weibullc
4
1.95
0.75
185.66
4.5E+00
3.2E+00
power restricted >1
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
c Alternate model also presented in this appendix
E.3.45.2. Figure for Selected Model: Gamma, Power Restricted >1
Gamma Multi-Hit Model with 0.95 Confidence Level
& 0.6
E 0.4
0.2
0 10 20 30 40 50 60 70
dose
16:49 11/20 2009
E.3.45.3. Output file for Selected Model: Gamma, Power Restricted >1
Gamma Model. (Version: 2.13; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Gamma_BMR2_Toxic_hepatopathy.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Gamma_BMR2_Toxic_hepatopathy.plt
Fri Nov 20 16:49:26 2009
0
Gamma Multi-Hit
BMDL BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-495 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the probability function is:
P[response]= background+(1-background)^CumGamma[slope*dose,power],
where CumGamma(.) is the cummulative Gamma distribution function
Dependent variable = DichEff
Independent variable = Dose
Power parameter is restricted as power >=1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial (and Specified) Parameter Values
Background = 0.00925926
Slope = 0.0683125
Power = 1.3
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Slope
Power
Slope
1
0. 94
Power
0. 94
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 NA
Slope 0.105412 0.0237428 0.0588765 0.151947
Power 1.92239 0.361359 1.21414 2.63064
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -89.8076 6
Fitted model -90.8168 2 2.01832 4 0.7324
Reduced model -218.207 1 256.799 5 <.0001
AIC: 185.634
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0000 0.000 0.000 53 0.000
2.1400 0.0265 1.429 2.000 54 0.484
7.1400 0.1926 10.205 8.000 53 -0.768
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-496 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
15.7 000
32.9000
71.4 000
0.517 0
0.8723
(j . 9 9 6 (J
27.4 03 30.000
46.232 45.000
52.788 53.000
53
53
53
0.714
-0.507
0. 4 62
Chi' '2
1.80
d. f.
E'-value
0 . 7 71 f
E'.enchmark Dose Computation
Specified effect = 0 .1
Risk Typ'e = E::tra risk
Cc'nfidence level = 0.95
BMD = 4.66805
BMDL = 3.31743
E.3.45.4. Figure for Unrestricted Model: Weibull, Power Restricted >1
Weibull Model with 0.95 Confidence Level
Weibull
.8
0.6
0.4
0.2
0
BMDL BMD
¦ ¦
10
¦ ' ¦
40
. i .
60
¦ ¦
70
0
20
30
50
dose
16:49 11/20 2009
E.3.45.5. Output File for Unrestricted Model: Weibull, Power Restricted >1
Weibull Model using Weibull Model (Version: 2.12; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Weibull_BMR2_Toxic_hepatopathy.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Weibull_BMR2_Toxic_hepatopathy.plt
Fri Nov 20 16:49:32 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-497 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(-slope*doseApower)]
Dependent variable = DichEff
Independent variable = Dose
Power parameter is restricted as power >=1
Total number of observations = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial (and Specified) Parameter Values
Background = 0.00925926
Slope = 0.0286401
Power = 1.19362
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Slope Power
Slope 1 -0.97
Power -0.97 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 NA
Slope 0.0113696 0.00592954 -0.000252127 0.0229912
Power 1.4905 0.169532 1.15823 1.82278
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -89.8076 6
Fitted model -90.8285 2 2.04181 4 0.7281
Reduced model -218.207 1 256.799 5 <.0001
AIC: 18 5.657
Dose Est. Prob. Expected
0.0000 0.0000 0.000
2.1400 0.0347 1.875
ess of Fit
Scaled
Observed Size Residual
0.000 53 0.000
2.000 54 0.093
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-498 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
7.1400
15.7000
. 9000
.4000
32 .
71.
0.1918
0. 4 97 9
0. 8745
0.9986
10.164
26.
46.
52 .
391
349
927
8 . 000
30.000
45.000
53.000
53
53
53
53
-0.755
0. 992
-0.559
0.270
Chi ^2
1. 95
d. f.
P-value
0.7454
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 4.4538
BMDL = 3.158 8 6
E.3.46. Ohsako et al. (2001): Anogenital PND120
E.3.46.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
0.46
5.02
0.17
185.44
7.5E+02
4.8E+02
nonconstant variance,
power restricted >1
exponential (M3)
3
0.46
5.02
0.17
185.44
7.5E+02
4.8E+02
nonconstant variance,
power restricted >1
exponential (M4)
2
0.46
4.29
0.12
186.72
4.8E+02
1.1E+01
nonconstant variance,
power restricted >1
exponential (M5)
2
0.46
4.29
0.12
186.72
4.8E+02
1.1E+01
nonconstant variance,
power restricted >1
Hill
2
0.46
3.11
0.21
185.54
7.1E+01
1.3E+01
nonconstant variance, n
restricted >1, bound hit
linear
3
0.46
5.09
0.17
185.52
7.6E+02
5.1E+02
nonconstant variance
polynomial
2
0.46
4.45
0.11
186.88
4.8E+02
1.5E+02
nonconstant variance
power
3
0.46
5.09
0.17
185.52
7.6E+02
5.1E+02
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
3
0.46
7.00
0.07
185.90
6.5E+02
4.2E+02
constant variance,
power restricted >1
exponential (M3)
3
0.46
7.00
0.07
185.90
6.5E+02
4.2E+02
constant variance,
power restricted >1
exponential (M4)
2
0.46
3.17
0.20
184.07
4.1E+01
1.2E+01
constant variance,
power restricted >1
exponential (M5)
1
0.46
2.84
0.09
185.74
3.8E+01
1.3E+01
constant variance,
power restricted >1
Hillc
2
0.46
2.74
0.25
183.64
6.0E+01
1.2E+01
constant variance, n
restricted >1, bound
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-499 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
hit
linear
3
0.46
7.12
0.07
186.02
6.6E+02
4.4E+02
constant variance
polynomial
2
0.46
5.55
0.06
186.45
2.7E+02
1.3E+02
constant variance
power
3
0.46
7.12
0.07
186.02
6.6E+02
4.4E+02
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
E.3.46.2. Figure for Selected Model: Hill, Constant Variance, n Restricted >1, Bound Hit
Hill Model with 0.95 Confidence Level
32 ft
Hill
¦
30
28
26
24
22 -
3MDL BMD
0 100
18:59 10/06 2009
¦
200 300 400 500 600
dose
700 800
E.3.46.3. Output File for Selected Model: Hill, Constant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/200S
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-500 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Input Data File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_Anogenital_PND120.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_Anogenital_PND12 0.plt
Tue Oct 06 18:59:02 2009
Figure 7
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 9.96434
rho = 0 Specified
intercept = 28.9146
v = -5.04512
n = 1.44913
k = 35.3408
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha intercept v k
alpha 1 -4.2e-009 -3.6e-008 2.3e-008
intercept -4.2e-009 1 -0.63 -0.52
v -3.6e-008 -0.63 1 -0.13
k 2.3e-008 -0.52 -0.13 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
alpha 9.51224 1.83063 5.92427 13.1002
intercept 28.9963 0.861586 27.3077 30.685
v -4.77893 1.14548 -7.02403 -2.53384
n 1 NA
k 33.2115 32.41 -30.3109 96.7338
NA - Indicates that this parameter has hit a bound
implied by some ineguality constraint and thus
has no standard error.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-501 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
12 . 5
50
200
800
12
10
10
10
12
28 . 9
28 .1
25.3
26.1
23. 9
29
27 . 7
26.1
24 . 9
24 . 4
3.54
2 . 52
3.59
3.59
2 . 36
3. 08
3. 08
3. 08
3. 08
3. 08
-0.0918
0.399
-0.836
1. 2
-0.605
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma^2
Likelihoods of Interest
Model
Log(likelihood)
# Param's
AIC
A1
-86.449919
6
184.899838
A2
-84 . 654549
10
189.309098
A3
-86.449919
6
184.899838
fitted
-87.819648
4
183.639297
R
-95.473923
2
194.947846
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
21. 6387
3.59074
3.59074
2 .73946
0.005631
0. 4642
0. 4642
0.2542
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than
model appears to be appropriate here
.1. A homogeneous variance
The p-value for Test 3 is greater than
to be appropriate here
.1. The modeled variance appears
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-502 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 60.4402
BMDL = 12.154 6
E.3.47. Schantz et al. (1996): Maze Errors Per Block, Female
E.3.47.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linear
1
0.71
3.25
0.07
20.63
6.7E+01
3.8E+01
nonconstant variance
polynomial
1
0.71
3.25
0.07
20.63
6.7E+01
3.8E+01
nonconstant variance
power
1
0.71
3.25
0.07
20.63
6.7E+01
3.8E+01
nonconstant variance,
power restricted >1,
bound hit
linearc
1
0.71
2.77
0.10
18.72
7.1E+01
4.6E+01
constant variance
polynomial
1
0.71
2.77
0.10
18.72
7.1E+01
4.6E+01
constant variance
power
1
0.71
2.77
0.10
18.72
7.1E+01
4.6E+01
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-503 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.47.2. Figure for Selected Model: Linear, Constant Variance
Linear Model with 0.95 Confidence Level
Linear
4
3.5
2
BMDL
BMD
0
20
40
60
80
100
dose
13:40 11/11 2009
E.3.47.3. Output File for Selected Model: Linear, Constant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\AD\LinearConstVar_BMR4_maze_errors.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\LinearConstVar_BMR4_maze_errors.plt
Rel Male Thymus wt, Tbl 2
The form of the response function is:
Y[dose] = beta_0 + beta_l*dose + beta_2*dose^2 + ...
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Signs of the polynomial coefficients are not restricted
A constant variance model is fit
Total number of dose groups = 3
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Wed Nov 11 13:40:58 2009
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-504 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
alpha =
rho =
beta_0 =
beta 1 =
0.569565
0
3.32773
-0.0105912
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha beta 0 beta 1
alpha 1 1.3e-010 3.9e-011
beta_0 1.3e-010 1 -0.7
beta 1 3.9e-011 -0.7 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
beta_0
beta 1
Estimate
0.562168
3.32773
-0.0105912
Std. Err.
0.145151
0.191722
0. 00322157
Lower Conf. Limit
0.277677
2.95196
-0.0169054
Upper Conf. Limit
0. 846659
3.7035
-0.00427705
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 10
25 10
100 10
3.55
2.76
2 . 34
3.33
3. 06
2 . 27
0. 639
0. 806
0. 806
0.75
0.75
0.75
0. 957
-1.28
0.319
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -4.976366 4 17.952732
A2 -4.638353 6 21.276707
A3 -4.976366 4 17.952732
fitted -6.360686 3 18.721371
R -10.975997 2 25.951993
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-505 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
Explanation of Tests
Test 1:
Test 2
Test 3
Test 4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio)
Test df
p-value
Test 1
Test 2
Test 3
Test 4
12 . 6753
0. 676025
0. 676025
2 .76864
0.01298
0.7132
0.7132
0.09613
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1.
model appears to be appropriate here
A homogeneous variance
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 70.7926
BMDL = 45.8305
E.3.48. Shi et al. (2007): Estradiol
E.3.48.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
3
0.05
15.48
0.00
395.70
1.7E+01
9.0E+00
nonconstant variance,
power restricted >1
exponential (M3)
3
0.05
15.48
0.00
395.70
1.7E+01
9.0E+00
nonconstant variance,
power restricted >1
exponential (M4)
C
2
0.05
1.41
0.49
383.64
5.6E-01
2.2E-01
nonconstant variance,
power restricted >1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-506 DRAFT—DO NOT CITE OR QUOTE
-------
exponential (M5)
2
0.05
1.41
0.49
383.64
5.6E-01
2.2E-01
nonconstant variance,
power restricted >1
Hill
2
0.05
0.52
0.77
382.74
4.4E-01
error
nonconstant variance, n
restricted >1, bound hit
linear
3
0.05
17.26
0.00
397.48
2.2E+01
1.5E+01
nonconstant variance
polynomial
2
0.05
7.34
0.03
389.57
5.4E+00
3.6E+00
nonconstant variance
power
3
0.05
17.26
0.00
397.48
2.2E+01
1.5E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
3
0.05
13.33
0.00
396.06
1.2E+01
6.6E+00
constant variance, power
restricted >1
exponential (M3)
3
0.05
13.33
0.00
396.06
1.2E+01
6.6E+00
constant variance, power
restricted >1
exponential (M4)
2
0.05
0.87
0.65
385.59
3.9E-01
1.5E-01
constant variance, power
restricted >1
exponential (M5)
2
0.05
0.87
0.65
385.59
3.9E-01
1.5E-01
constant variance, power
restricted >1
Hill
2
0.05
0.37
0.83
385.09
3.1E-01
1.0E-01
constant variance, n
restricted >1, bound hit
linear
3
0.05
15.40
0.00
398.12
1.9E+01
1.3E+01
constant variance
polynomial
2
0.05
8.10
0.02
392.82
4.5E+00
2.9E+00
constant variance
power
3
0.05
15.40
0.00
398.12
1.9E+01
1.3E+01
constant variance, power
restricted >1, bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meetBMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-507 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.48.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
140
120
100
80
60
40
20
EMDL BMD
Exponential
10
15
dose
20
25
30
19:13 10/06 2009
E.3.48.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\Exp_BMRl_Shi_estradiol_17B_conc_PE9.(d)
Gnuplot Plotting File:
Tue Oct 06 19:13:28 2009
Figure 4 PE9 only
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-508 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
lnalpha
rho
Model 4
2.65881
0.913414
108
0.136287
0.340136
1
Parameter Estimates
Variable Model 4
lnalpha 1.81331
rho 1.12126
a 100.526
b 1.53823
c 0.431796
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 10 102.9 41.41
0.143 10 86.19 19.58
0.714 10 63.33 29.36
7.14 10 48.1 18.82
28.6 10 38.57 22.59
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 100.5 32.83 0.2245
0.143 89.25 30.71 -0.3147
0.714 62.45 25.14 0.1108
7.14 43.41 20.5 0.723
28.6 43.41 20.5 -0.7458
Other models for which
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
likelihoods are calculated:
= Mu(i) + e(ij)
= Sigma^2
= Mu(i) + e(ij)
= Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-509 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -188.3615 6 388.7231
A2 -183.667 10 387.3339
A3 -186.1132 7 386.2263
R -203.3606 2 410.7211
4 -186.8176 5 383.6352
Additive constant for all log-likelihoods = -45.95. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
39.39
9.389
4 . 892
1.409
p-value
< 0.0001
0.05208
0.1798
0.4944
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 0.555948
BMDL = 0.223612
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-510 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.49. Smialowicz et al. (2008): PFC per 10A6 Cells
2 E.3.49.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)c
3
<0.0001
11.58
0.01
890.56
1.1E+02
7.2E+01
nonconstant variance,
power restricted >1
exponential (M3)
2
<0.0001
10.85
0.00
891.83
1.3E+02
7.6E+01
nonconstant variance,
power restricted >1
exponential (M4)
3
<0.0001
11.58
0.01
890.56
1.1E+02
7.2E+01
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
10.85
0.00
891.83
1.3E+02
7.6E+01
nonconstant variance,
power restricted >1
Hill
2
<.0001
10.26
0.01
891.23
1.3E+02
error
nonconstant variance, n
restricted >1, bound hit
linear
3
<.0001
11.79
0.01
890.77
1.8E+02
1.5E+02
nonconstant variance
polynomial
2
<.0001
10.36
0.01
891.34
1.3E+02
8.4E+01
nonconstant variance
power
3
<.0001
11.79
0.01
890.77
1.8E+02
1.5E+02
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
3
<0.0001
7.92
0.05
903.59
8.2E+01
4.8E+01
constant variance,
power restricted >1
exponential (M3)
3
<0.0001
7.92
0.05
903.59
8.2E+01
4.8E+01
constant variance,
power restricted >1
exponential (M4)
2
<0.0001
7.91
0.02
905.58
8.0E+01
6.2E+00
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
7.91
0.02
905.58
8.0E+01
6.2E+00
constant variance,
power restricted >1
Hill
2
<.0001
7.31
0.03
904.98
1.6E+01
2.2E+00
constant variance, n
restricted >1, bound hit
linear
3
<.0001
10.33
0.02
905.99
1.5E+02
1.1E+02
constant variance
polynomial
2
<.0001
8.14
0.02
905.80
8.8E+01
5.5E+01
constant variance
power
3
<.0001
10.33
0.02
905.99
1.5E+02
1.1E+02
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-511 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.49.2. Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
BMDL
50 100 150 200 250 300
dose
19:14 10/06 2009
E.3.49.3. Output File for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\Exp_BMRl_PFC_per_cells.(d)
Gnuplot Plotting File:
Tue Oct 06 19:14:43 2009
Anti Response to SRBCs, PFC per 10^6 cells, Table 4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-512 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
lnalpha 3.29848
rho 1.2578
a 1565.55
b 0.00725727
c 0.00205679
d 1
Parameter Estimates
Variable Model 2
lnalpha 1. 84544
rho 1.53651
a 1195.73
b 0.00560912
c 0
d 1.22053
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 15 1491 716
1.07 14 1129 171
10.7 15 945 516
107 15 677 465
321 8 161 117
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 1232 593.6 1.688
1.07 1224 590.7 -0.6027
10.7 1153 565 -1.428
107 635.7 362 0.442
321 169.2 134.6 -0.1716
Other models for which
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
likelihoods are calculated:
= Mu(i) + e(ij)
= Sigma^2
= Mu(i) + e(ij)
= Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-513 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Model A3:
Yij
(ij ) }
Mu(i) + e(ij)
exp(lalpha + log(mean(
rho )
Model R: Yij
Var{e(ij)}
Mu + e(i
Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -444.8329 6 901.6657
A2 -425.4028 10 870.8057
A3 -435.4894 7 884.9787
R -463.7537 2 931.5074
2 -441.2778 4 890.5555
Additive constant for all log-likelihoods = -61.57. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 76.7 8 < 0.0001
Test 2 38.86 4 < 0.0001
Test 3 20.17 3 0.0001563
Test 4 11.58 3 0.008983
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 4 is less than .1. Model 2 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 106.252
BMDL = 71.9153
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-514 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.50. Smialowicz et al. (2008): PFC per Spleen
2 E.3.50.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)c
3
0.00
5.60
0.13
377.40
1.3E+02
8.4E+01
nonconstant
variance, power
restricted >1
exponential (M3)
3
0.00
5.60
0.13
377.40
1.3E+02
8.4E+01
nonconstant variance,
power restricted >1
exponential (M4)
3
0.00
5.60
0.13
377.40
1.3E+02
8.2E+01
nonconstant variance,
power restricted >1
exponential (M5)
2
0.00
5.60
0.06
379.40
1.3E+02
8.2E+01
nonconstant variance,
power restricted >1
Hill
2
0.00
5.35
0.07
379.15
1.4E+02
error
nonconstant variance,
n restricted >1, bound
hit
linear
3
0.00
8.09
0.04
379.89
2.2E+02
1.7E+02
nonconstant variance
polynomial
2
0.00
5.58
0.06
379.38
1.4E+02
8.9E+01
nonconstant variance
power
3
0.00
8.09
0.04
379.89
2.2E+02
1.7E+02
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
3
0.00
5.58
0.13
392.71
1.0E+02
5.5E+01
constant variance,
power restricted >1
exponential (M3)
3
0.00
5.58
0.13
392.71
1.0E+02
5.5E+01
constant variance,
power restricted >1
exponential (M4)
2
0.00
5.58
0.06
394.71
1.0E+02
6.5E+00
constant variance,
power restricted >1
exponential (M5)
2
0.00
5.58
0.06
394.71
1.0E+02
6.5E+00
constant variance,
power restricted >1
Hill
2
0.00
5.36
0.07
394.49
8.4E+01
1.7E+00
constant variance, n
restricted >1, bound
hit
linear
3
0.00
7.31
0.06
394.44
1.7E+02
1.3E+02
constant variance
polynomial
2
0.00
5.74
0.06
394.87
1.1E+02
6.3E+01
constant variance
power
3
0.00
7.31
0.06
394.44
1.7E+02
1.3E+02
constant variance,
power restricted >1,
bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-515 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.50.2. Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
BMDL
BMD
50
100
150
dose
200
250
300
19:15 10/06 2009
E.3.50.3. Output File for Selected Model: Exponential (M2'), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AniDose\Exp_BMRl_PFC_per_spleen.(d)
Gnuplot Plotting File:
Tue Oct 06 19:15:26 2009
Anti Response to SRBCs - PFC x 10 to the 4 per spleen, Table 4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-516 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
lnalpha
rho
Model 2
0.786146
1. 36372
29.19
0.00907371
0.0513875
1
Parameter Estimates
Variable Model 2
lnalpha 0.525138
rho 1.45988
a 22.9464
b 0.00618274
c 0
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0
15
27 . 8
13. 4
1 . 07
14
21
13. 6
10.7
15
17 . 6
9.4
107
15
12 . 6
8 . 7
321
8
3
3.1
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 22.95 12.8 1.468
1.07 22.8 12.74 -0.5272
10.7 21.48 12.2 -1.231
107 11.84 7.899 0.3719
321 3.153 3.007 -0.1444
Other models for which
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
likelihoods are calculated:
= Mu(i) + e(ij)
= Sigma^2
= Mu(i) + e(ij)
= Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-517 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Model A3:
Yij
(ij ) }
Mu(i) + e(ij)
exp(lalpha + log(mean(
rho )
Model R: Yij
Var{e(ij)}
Mu + e(i
Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -190.565 6 393.13
A2 -181.4763 10 382.9526
A3 -181.9 7 377.8001
R -204.6365 2 413.273
2 -184.6977 4 377.3954
Additive constant for all log-likelihoods = -61.57. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 46.32 8 < 0.0001
Test 2 18.18 4 0.001139
Test 3 0.8475 3 0.8381
Test 4 5.595 3 0.133
The p-value for Test 1 is less than .05. There appears to be
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is greater than .1. Model 2 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 132.016
BMDL = 84.3108
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-518 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
E.3.51. Toth et al. (1978): Amyloidosis
E.3.51.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
multistage, 1-
degree b
1
0.00
1.00
33.16
8.7E-01
4.5E-01
betas restricted >0
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
E.3.51.2. Figure for Selected Model: Multistage, 1-Degree, Betas Restricted >0
Multistage Model with 0.95 Confidence Level
Multistage
0.25
0.2
0.15
0.05
BMDL
BMD
0
0.2
0.4
0.6
0.8
1
dose
19:21 10/06 2009
E.3.51.3. Output File for Selected Model: Multistage, 1-Degree, Betas Restricted >0
Multistage Model. (Version: 3.0; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\mult2_0.l_amyloidosis_lyr.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\mult2_0.1_amyloidosis_lyr.plt
Tue Oct 06 19:21:03 2009
Table 2
The form of the probability function is:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-519 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
P[response] = background + (1-background)*[1-EXP(
-betal^dose^l)]
The parameter betas are restricted to be positive
Dependent variable = DichEff
Independent variable = Dose
Total number of observations = 2
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0.120628
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 0.120628 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -15.5783 2
Fitted model -15.5783 1 1.42109e-014 1 1
Reduced model -18.8308 1 6.50504 1 0.01076
AIC: 33.1565
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0000 0.000 0.000 38 0.000
1.0000 0.1136 5.000 5.000 44 -0.000
Chi/N2 = 0.00 d.f. = 1 P-value = 1.0000
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-520 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 0.873433
BMDL = 0.453242
BMDU = 2.10749
Taken together, (0.453242, 2.10749) is a 90 % two-sided confidence
interval for the BMD
E.3.52. Toth et al. (1978): Skin Lesions
E.3.52.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X1 Test
Statistic
1P-
Value3
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
gamma
2
9.35
0.01
159.22
1.2E+02
8.3E+01
power restricted >1, bound
hit
logistic
2
12.19
0.00
162.97
2.7E+02
2.1E+02
log-logistic
2
7.10
0.03
156.57
6.7E+01
4.1E+01
slope restricted >1, bound
hit
log-probitb
2
1.17
0.56
148.22
1.1E+00
6.8E-02
slope restricted >1
multistage, 2-
degree
2
9.35
0.01
159.22
1.2E+02
8.3E+01
betas restricted >0, bound
hit
probit
2
11.98
0.00
162.68
2.5E+02
2.0E+02
Weibull
2
9.35
0.01
159.22
1.2E+02
8.3E+01
power restricted >1, bound
hit
a Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-521 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.52.2. Figure for Selected Model: Log-Probit, Slope Restricted >1
LogProbit Model with 0.95 Confidence Level
0.7
0.6
0.5
0.4
0.3
0.2
0.1
B i/IDLBMD
LogProbit
200
400 600
dose
800
1000
19:21 10/06 2009
E.3.52.3. Output File for Selected Model: Log-Probit, Slope Restricted >1
Probit Model. (Version: 3.1; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\LogProbit_BMR2_Skin_lesion_lyr.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\LogProbit_BMR2_Skin_lesion_lyr.plt
Tue Oct 06 19:21:47 2009
Table 2
The form of the probability function is:
P[response] = Background
+ (1-Background) * CumNorm(Intercept+Slope*Log(Dose)),
where CumNorm(.) is the cumulative normal distribution function
Dependent variable = DichEff
Independent variable = Dose
Slope parameter is not restricted
Total number of observations = 4
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-522 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
User has chosen the log transformed model
Default Initial (and Specified) Parameter Values
background = 0
intercept = -1.26532
slope = 0.195762
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
intercept slope
intercept 1 -0.87
slope -0.87 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
background 0 NA
intercept -1.30013 0.240943 -1.77237 -0.827887
slope 0.202414 0.0463497 0.111571 0.293258
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -71.5177 4
Fitted model -72.1089 2 1.18249 2 0.553£
Reduced model -95.8498 1 48.6642 3 <.0001
AIC: 148.218
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
38
0. 000
1.0000
0.0968
4 . 258
5. 000
44
0.378
100.0000
0.3564
15.684
13.000
44
-0.845
1000.0000
0.5391
23.180
25.000
43
0.557
Chi ^2 = 1.17 d.f. = 2 P-value = 0.5581
Benchmark Dose Computation
Specified effect = 0.1
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1.09611
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-523 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMDL = 0.0684731
2
3
4 E.3.53. Van Birgelen et al. (1995a): Hepatic Retinol
5 E.3.53.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
4
<0.0001
45.69
<0.0001
164.34
2.9E+02
error
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
45.69
<0.0001
164.34
2.9E+02
error
nonconstant variance,
power restricted >1
exponential (M4)c
3
<0.0001
27.40
<0.0001
148.05
1.2E+02
7.1E+01
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
27.40
<0.0001
148.05
1.2E+02
7.1E+01
nonconstant variance,
power restricted >1
exponential (M5)d
3
<0.0001
27.40
<0.0001
148.05
1.2E+02
7.1E+01
nonconstant variance,
power unrestricted
Hill
3
<.0001
8.11
0.04
128.76
1.3E+01
error
nonconstant variance, n
restricted >1, bound hit
Hilld
2
<.0001
2.62
0.27
125.27
5.6E+00
error
nonconstant variance, n
unrestricted
linear
4
<.0001
60.09
<.0001
178.73
7.8E+02
6.0E+02
nonconstant variance
polynomial
4
<.0001
60.09
<.0001
178.73
7.8E+02
6.0E+02
nonconstant variance
power
4
<.0001
60.09
<.0001
178.73
7.8E+02
6.0E+02
nonconstant variance,
power restricted >1,
bound hit
power d
3
<.0001
9.34
0.03
129.99
4.2E-01
8.5E-03
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
141.80
<0.0001
322.09
error
error
constant variance,
power restricted >1
exponential (M3)
4
<0.0001
141.80
<0.0001
322.09
error
error
constant variance,
power restricted >1
exponential (M4)
3
<0.0001
2.71
0.44
185.03
1.2E+01
7.0E+00
constant variance,
power restricted >1
exponential (M5)
3
<0.0001
2.71
0.44
185.03
1.2E+01
7.0E+00
constant variance,
power restricted >1
exponential (M5)
3
<0.0001
2.71
0.44
185.03
1.2E+01
7.0E+00
constant variance,
power unrestricted
Hill
3
<.0001
1.37
0.71
183.68
8.3E+00
4.2E+00
constant variance, n
restricted >1, bound hit
Hill
2
<.0001
0.89
0.64
185.20
4.5E+00
5.8E-02
constant variance, n
unrestricted
linear
4
<.0001
27.00
<.0001
207.31
5.3E+02
3.9E+02
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-524 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
polynomial
4
<.0001
27.00
<.0001
207.31
5.3E+02
3.9E+02
constant variance
power
4
<.0001
27.00
<.0001
207.31
5.3E+02
3.9E+02
constant variance,
power restricted >1,
bound hit
power
3
<.0001
1.92
0.59
184.23
4.4E-01
6.6E-03
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
1
2
3 E.3.53.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
4 Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
BMDL
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-525 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.3.53.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_hepatic_retinol.(d)
Gnuplot Plotting File:
Fri Nov 20 14:29:52 2009
Tbl3, hepatic retinol
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -1.16065
rho 1.53688
a 15.645
b 0.00625117
c 0.0365247
d 1
Parameter Estimates
Variable Model 4
lnalpha -0.882224
rho 1.82707
a 10.5294
b 0.00720346
c 0.068866
d 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-526 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Stats From Input Data
Obs Mean
Obs Std Dev
0
14
26
47
320
1024
14 . 9
8 . 4
8 . 2
5.1
2 . 2
0.6
8.768
3.394
2 .263
0. 8485
0. 8485
0.5657
Estimated Values of Interest
Dose
0
14
26
47
320
1024
Est Mean
10.53
9.589
8 . 855
7 .714
1. 703
0.7313
Est Std
5.526
5. 073
4 .717
4 .159
1.046
0. 4833
Scaled Residual
2 . 237
-0.6628
-0.3926
-1.778
1. 343
-0.7681
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -87.1567 7 188.3134
A2 -47.28742 12 118.5748
A3 -55.32422 8 126.6484
R -109.967 2 223.934
4 -69.02619 5 148.0524
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Tests of Interest
-2*log(Likelihood Ratio) D. F. p-value
Test
1:
Test
2 :
Test
3:
Test
6a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-527 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Test 1
Test 2
Test 3
Test 6a
125. 4
79.74
16. 07
27 . 4
10
5
4
3
< 0.0 0 01
< 0.0 0 01
0.002 922
< 0.0 0 01
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/o'r variances among the duse
levels, it seems appropriate to' model the 'data.
The p-value tor Test 2 is less than .1. A riO'ri-hO'mogerieO'US
variance model appears t'0 be appropriate.
The p-value tor Test 3 is less than .1. You may want t'0
consider a 'different variance model.
The p-value tor Test 6a is less than .1. Model 4 may no't adequately
describe the data; you may want t'O corisider another model.
E'enchmark Dcse Cc'irputaticris :
Specified Effect = 1.000000
Risk Typ'e = Estimated standard deviatioris from coritro'l
Confidence Level = 0.950000
E'MD = 115.12 8
E'MDL = 7 0.981
E.3.53.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
5
BMDL BMD
200
400
600
800
1000
dose
14:30 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-528 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
E.3.53.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_Unrest_BMRl_hepatic_retinol.(d)
Gnuplot Plotting File:
FriNov 20 14:30:05 2009
Tbl3, hepatic retinol
The form of the response function by Model:
Model 2: Y[dose] = a * exp{sign * b * dose}
Model 3: Y[dose] = a * exp{sign * (b * dose)Ad}
Model 4: Y[dose] = a * [c-(c-l) * exp{-b * dose}]
Model 5: Y[dose] = a * [c-(c-l) * exp{-(b * dose)Ad}]
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -1.16065
rho 1.53688
a 15.645
b 0.00625117
c 0.0365247
d 1
Parameter Estimates
Variable Model 5
lnalpha -0.882224
rho 1.82707
a 10.5294
b 0.00720346
c 0.068866
d 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-529 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
00
o
14.9
8.768
14 8
8.4
3.394
26 8
8.2
2.263
47 8
5.1
0.8485
320 8
2.2
0.8485
1024 *
! 0.6
0.5657
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0
10.53
5.526
2.237
14
9.589
5.073
-0.6628
26
8.855
4.717
-0.3926
47
7.714
4.159
-1.778
320
1.703
1.046
1.343
1024
0.7313
0.4833
-0.7681
Other models for which likelihoods are calculated:
Model Al: Yij = Mu(i) + e(ij)
Var{e(ij)} = SigmaA2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)A2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = SigmaA2
Likelihoods of Interest
Model
Log(likelihood)
DF AIC
Al
-87.1567
7
188.3134
A2
-47.28742
12
118.5748
A3
-55.32422
8
126.6484
R
-109.967
2
223.934
5
-69.02619
5
148.0524
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. Al)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-530 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1
125.4
10
<0.0001
Test 2
79.74
5
<0.0001
Test 3
16.07
4
0.002922
Test 7a
27.4
3
<0.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than . 1. You may want to
consider a different variance model.
The p-value for Test 7a is less than . 1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 115.128
BMDL = 70.981
E.3.53.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
20
15
10
Hill
BMD
200
400
600
800
1000
dose
14:30 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-531 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.3.53.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_hepatic_retinol.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_hepatic_retinol.plt
FriNov 20 14:30:12 2009
Tbl3, hepatic retinol
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 2.76506
rho = 0
intercept = 14.9
v= -14.3
n = 2.92354
k = 29.0484
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -0.78 -0.041 0.015 0.037 0.029
rho -0.78 1 -0.098 0.11 -0.047 -0.046
intercept -0.041 -0.098 1 -0.9 -0.25 -0.8
v 0.015 0.11 -0.9 1 0.63 0.63
n 0.037 -0.047 -0.25 0.63 1 0.15
k 0.029 -0.046 -0.8 0.63 0.15 1
Parameter Estimates
95.0%
Wald Confidence Interval
Variable
Estimate
Std. Err.
Lower Conf. Limit
Upper Conf. Limit
lalpha
-1.16547
0.373814
-1.89813
-0.432809
rho
1.69882
0.185479
1.33529
2.06235
intercept
16.6759
2.07841
12.6023
20.7495
V
-17.4464
2.46627
-22.2801
-12.6126
n
0.570647
0.161383
0.254343
0.886951
k
16.5364
7.36467
2.10191
30.9709
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-532 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
00
o
14.9
16.7
8.77
6.1
-0.824
14 8
8.4
8.37
3.39
3.39
0.0276
26 8
8.2
6.83
2.26
2.86
1.35
47 8
5.1
5.43
0.849
2.35
-0.394
320 8
2.2
1.95
0.849
0.983
0.732
1024 :
3 0.6
0.742
0.566
0.434
-0.929
Model Descriptions for likelihoods calculated
Model Al: Yij = Mu(i) + e(ij)
Var{e(ij)} = SigmaA2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)A2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = SigmaA2
Likelihoods of Interest
Model
Log(likelihood)
# Param's AIC
Al
-87.156698
7
188.313395
A2
-47.287416
12
118.574833
A3
-55.324218
8
126.648436
fitted
-56.636555
6
125.273110
R
-109.967018
2
223.934036
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (Al vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Testdf p-value
Test 1
125.359
10
<.0001
Test 2
79.7386
5
<.0001
Test 3
16.0736
4
0.002922
Test 4
2.62467
2
0.2692
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than . 1. A non-homogeneous variance
model appears to be appropriate
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-533 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
The p-value for Test 3 is less than . 1. You may want to consider a
different variance model
The p-value for Test 4 is greater than . 1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 5.56122
BMDL computation failed.
E.3.53.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
20
15
10
5
0
0
200
400
600
800
1000
dose
14:30 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-534 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
E.3.53.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_hepatic_retinol.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_hepatic_retinol.plt
FriNov 20 14:30:13 2009
Tbl3, hepatic retinol
The form of the response function is:
Y[dose] = control + slope * dose power
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
1 alpha = 2.76506
rho = 0
control = 14.9
slope = -3.78637
power = 0.191713
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho control slope power
lalpha 1 -0.8 -0.047 0.042 0.065
rho -0.8 1 -0.085 -0.0029 -0.11
control -0.047 -0.085 1 -0.95 -0.81
slope 0.042 -0.0029 -0.95 1 0.96
power 0.065 -0.11 -0.81 0.96 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
-1.02622
1.68421
16.9577
-7.19097
0.117935
Std. Err.
0.389164
0.199212
2.21133
1.99708
0.0225396
Lower Conf. Limit
-1.78897
1.29376
12.6235
-11.1052
0.0737578
Upper Conf. Limit
-0.263475
2.07466
21.2918
-3.27676
0.162111
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-535 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
00
o
14.9
17
8.77
6.49
-0.896
14 8
8.4
7.14
3.39
3.13
1.14
26 8
8.2
6.4
2.26
2.86
1.78
47 8
5.1
5.63
0.849
2.57
-0.588
320 8
2.2
2.76
0.849
1.41
-1.12
1024 :
3 0.6
0.672
0.566
0.428
-0.475
Model Descriptions for likelihoods calculated
Model Al: Yij = Mu(i) + e(ij)
Var{e(ij)} = SigmaA2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)A2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = SigmaA2
Likelihoods of Interest
Model
Log(likelihood)
# Param's AIC
Al
-87.156698
7
188.313395
A2
-47.287416
12
118.574833
A3
-55.324218
8
126.648436
fitted
-59.994980
5
129.989960
R
-109.967018
2
223.934036
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (Al vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Testdf p-value
Test 1
125.359
10
<.0001
Test 2
79.7386
5
<.0001
Test 3
16.0736
4
0.002922
Test 4
9.34152
3
0.02508
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than . 1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-536 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
different variance model
The p-value for Test 4 is less than . 1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.420475
BMDL = 0.00850422
E.3.54. Van Birgelen et al. (1995a): Hepatic Retinol Palmitate
E.3.54.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
p-V aluea
J2 Test
Statistic
X2P-
Value"
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
exponential (M2)
4
<0.0001
64.68
<0.0001
467.45
error
error
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
64.68
<0.0001
467.45
error
error
nonconstant variance,
power restricted >1
exponential (M4)
3
<0.0001
49.32
<0.0001
454.09
error
error
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
49.32
<0.0001
454.09
error
error
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
49.32
<0.0001
454.09
error
error
nonconstant variance,
power unrestricted
Hill
3
<.0001
158.81
<.0001
563.58
error
error
nonconstant variance, n
restricted >1
Hilld
3
<.0001
117.56
<.0001
522.32
2.4E-12
2.4E-12
nonconstant variance, n
unrestricted
linearc
4
<0001
85.68
<0001
488.45
1.4E+03
9.9E+02
nonconstant variance
polynomial
4
<.0001
85.68
<.0001
488.45
1.4E+03
9.9E+02
nonconstant variance
power
4
<.0001
85.68
<.0001
488.45
1.4E+03
9.9E+02
nonconstant variance,
power restricted >1,
bound hit
power4
3
<.0001
3.30
0.35
408.06
3.8E-02
1.2E-05
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
140.00
<0.0001
647.15
error
error
constant variance, power
restricted >1
exponential (M3)
4
<0.0001
140.00
<0.0001
647.15
error
error
constant variance, power
restricted >1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-537 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Valuea
J2 Test
Statistic
1P-
Value"
AIC
BMD
(ng/kg-d)
BMDL
(ng/kg-d)
Model Notes
exponential (M4)
3
<0.0001
3.50
0.32
512.61
2.5E+00
9.0E-03
constant variance, power
restricted >1
exponential (M5)
3
<0.0001
3.50
0.32
512.61
2.5E+00
9.0E-03
constant variance, power
restricted >1
exponential (M5)d
3
<0.0001
3.50
0.32
512.61
2.5E+00
9.0E-03
constant variance, power
unrestricted
Hill
3
<.0001
1.33
0.72
510.44
1.3E+00
2.5E-01
constant variance, n
restricted >1, bound hit
Hill
2
<.0001
0.29
0.86
511.40
7.9E-06
7.9E-06
constant variance, n
unrestricted
linear
4
<.0001
44.59
<.0001
551.70
8.7E+02
5.5E+02
constant variance
polynomial
4
<.0001
44.59
<.0001
551.70
8.7E+02
5.5E+02
constant variance
power
4
<.0001
44.59
<.0001
551.70
8.7E+02
5.5E+02
constant variance, power
restricted >1, bound hit
power
3
<.0001
0.29
0.96
509.40
2.2E-08
2.2E-08
constant variance, power
unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meetBMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-538 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.54.2. Figure for Selected Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
700
600
500
400
300
200
100
Hill
B wDLBMD
200
400
600
800
1000
dose
14:31 11/20 2009
E.3.54.3. Output File for Selected Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_hepatic_retinol_palmitate.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\Nov2 0\Hill_Unrest_BMRl_hepatic_retinol_palmitate.pit
Fri Nov 20 14:31:05 2009
Tbl3, hepatic retinol palmitate
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-539 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha
rho
intercept
9.57332
0
472
-469
1.50651
8.68519
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
The model parameter(s) -k
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha
1
-1
-0.43
0.39
-0.014
rho
-1
1
0.44
-0.41
0. 015
intercept
-0.43
0.44
1
-1
0. 027
v
0.39
-0.41
-1
1
-0.026
n
-0.014
0. 015
0. 027
-0.026
1
Parameter Estimates
Variable
lalpha
rho
intercept
k
Estimate
2.09439
1.43616
640.986
-4 95.665
0. 451934
1. 024e-012
Std. Err.
1. 99191
0.388229
167.573
166.074
0.597514
NA
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-1.80969
0. 675242
312 . 548
-821.163
-0.719171
5.99847
2.19707
969.423
-170.167
1.62304
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 472 641
14 8 94 145
26 8 107 145
47 8 74 145
320 8 22 145
1024 8 3 145
272 295 -1.62
67.9 102 -1.43
76.4 102 -1.07
39.6 102 -1.98
22.6 102 -3.43
2.83 102 -3.96
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-540 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -250.554817 7 515.109634
A2 -196.755746 12 417.511491
A3 -197.383174 8 410.766347
fitted -256.161039 5 522.322077
R -276.789644 2 557.579287
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 160.068 10 <.0001
Test 2 107.598 5 <.0001
Test 3 1.25486 4 0.869
Test 4 117.556 3 <.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 2.41754e-012
BMDL = 2.41754e-012
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-541 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.54.4. Figure for Unrestricted Model: Linear, Nonconstant Variance
Linear Model with 0.95 Confidence Level
700
600
500
400
300
200
100
-100
Linear
BMDL
BMEI
200
400
600 800
dose
1000
1200
1400
14:30 11/20 2009
E.3.54.5. Output File for Unrestricted Model: Linear, Nonconstant Variance
Polynomial Model. (Version: 2.13; Date: 04/08/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Linear_BMRl_hepatic_retinol_palmitate.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Linear_BMRl_hepatic_retinol_palmitate.plt
Fri Nov 20 14:30:57 2009
Tbl3, hepatic retinol palmitate
The form of the response function is:
Y[dose] = beta 0 + beta l*dose + beta 2*dose^2 +
Dependent variable = Mean
Independent variable = Dose
Signs of the polynomial coefficients are not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
rho)
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-542 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
lalpha
rho
beta_0
beta 1
9.57332
0
177 .506
-0.204775
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho beta 0 beta 1
lalpha 1 -0.95 -0.017 0.022
rho -0.95 1 0.00019 -0.0048
beta_0 -0.017 0.00019 1 -1
beta_l 0.022 -0.0048 -1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
beta_0
beta 1
Estimate
-0.723216
2 .26615
150.535
-0.143931
Std. Err.
0.638291
0.140196
31.5457
0.0308317
Lower Conf. Limit
-1.97424
1.99137
88.7064
-0.20436
Upper Conf. Limit
0.527811
2 . 54093
212.363
-0.0835018
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
14
26
47
320
1024
472
94
107
74
22
3
151
149
147
144
104
3.15
272
67 . 9
76.4
39.6
22 . 6
2 . 83
204
201
199
194
135
2 . 56
4 .45
-0.766
-0.567
-1. 02
-1.73
-0.166
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Mu(i) + e(ij ;
Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -250.554817 7 515.109634
A2 -196.755746 12 417.511491
A3 -197.383174 8 410.766347
fitted -240.223107 4 488.446215
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-543 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
-276.789644
557.579287
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 160.068 10 <.0001
Test 2 107.598 5 <.0001
Test 3 1.25486 4 0.869
Test 4 85.6799 4 <.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1419.81
BMDL = 988.945
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-544 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.54.6. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
700
600
500
400
300
200
100
Power
Bi/IDLBMD
200
400 600
dose
800
1000
14:31 11/20 2009
E.3.54.7. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_hepatic_retinol_palmitate.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_hepatic_retinol_palmitate.pit
Fri Nov 20 14:31:06 2009
Tbl3, hepatic retinol palmitate
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(
exp(lalpha + log(mean(i)
rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-545 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
lalpha
rho
control
slope
power
9.57332
0
472
-315.054
0.0586881
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 95
0.29
-0.31
-0.3
rho
-0. 95
1
-0.4
0.39
0.29
control
0.29
-0.4
1
-0. 98
-0. 82
slope
-0.31
0.39
-0. 98
1
0. 91
power
-0.3
0.29
-0. 82
0. 91
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
0.0734958
1.80632
465.497
-318.06
0. 0540573
Std. Err.
0. 849559
0.194602
86.914
82.4127
0.0117709
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-1.59161
1.42491
295.149
-479.586
0.0309869
1.7386
2 .18774
635.845
-156.534
0. 0771278
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
14
26
47
320
1024
472
94
107
74
22
3
465
98 . 7
86.2
73.8
31.1
2 .86
272
67 . 9
76.4
39.6
22 . 6
2 . 83
266
65. 6
58 .1
50.5
23.1
2 . 68
0. 069
-0.201
1. 01
0.0086
-1.11
0.145
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma^2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-546 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Model Log(likelihood) # Param's AIC
A1 -250.554817 7 515.109634
A2 -196.755746 12 417.511491
A3 -197.383174 8 410.766347
fitted -199.031154 5 408.062307
R -276.789644 2 557.579287
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 160.068 10 <.0001
Test 2 107.598 5 <.0001
Test 3 1.25486 4 0.869
Test 4 3.29596 3 0.3482
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.0376489
BMDL = 1.207 69e-005
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-547 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.54.8. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
700
600
500
400
300
200
100
B i/lDL
BMD
Exponential
T
200
400 600
dose
800
1000
14:31 11/20 2009
E.3.54.9. Output File for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_CV_Unrest_BMRl_hepatic_retinol_palmitate.(d)
Gnuplot Plotting File:
Fri Nov 20 14:31:06 2009
Tbl3, hepatic retinol palmitate
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-548 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 5
lnalpha
rho(S)
9.43978
0
495. 6
0.00826283
0.00576502
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 5
9.51279
0
470.237
0.126105
0.0898547
1
NC
No Convergence
Table of Stats From Input Data
Obs Mean
Obs Std Dev
0
14
26
47
320
1024
472
94
107
74
22
3
271. 5
67 .88
76.37
39.6
22 . 63
2 . 828
Estimated Values of Interest
Dose
Est Mean
Est Std
Scaled Residual
0
470.2
116.3
0. 04286
14
115.5
116.3
-0.5224
26
58 . 38
116.3
1.182
47
43.39
116.3
0.7442
320
42 . 25
116.3
-0.4924
1024
42 . 25
116.3
-0.9544
Other models for which likelihoods are calculated:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-549 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model A1: Yij = Mu(i) + e(ij)
Va r{e(i j ) } = S i gma'' 2
Model A2 : Yij = Mu(i) + e(ij)
Va r{e(i j ) } = S i gma(i) 112
Model A3: Yij = Mu(i) + e(ij )
Var{e(ij)} = e::p (lalpha + Log (mean ( i ) ) ^ rloo )
Model R: Yij = Mu + e(i)
Va r{e(i j ) } = S i gma112
LikelihO'O'ds O'f Interest
Model Log(likelihood) DF AIC
A1 -250.5548 7 515.1096
A2 -196.7557 12 417.5115
A3 -250.5548 7 515.1096
R -276.7896 2 557.5793
5 -252.3071 4 512.6141
Additive ccristant Lor all Log-likelihoC'ds = -4 4.11. This cunstant added t'O the
abO'Ve values gives the Log-likelihoC'd including the term that dO'es riO't
depend cri the iriO'del parameters.
E::p'lanaticri C'f Tests
Test 1: D'Oes resp"Onse and/o'r variances 'differ amcrig Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately iriC'deled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests O'f Interest
-2^ Log ( Likelihood Rati'O )
160.1
107 . 6
107 . 6
3.504
D. F.
10
5
5
3
P'-value
< 0.0 0 01
< 0.0 0 01
< 0.0 0 01
0.32 02
The p'-value Lor Test 1 is less than .05. There appears t'O be a
difference between resp'Orise and /cr variances amcrig the dcse
levels, it seems app're'P'riate t'O mo'del the data.
The p'-value Lor Test 2 is less than .1. Corisider running
a nO'n-hO'mO'geneO'US variance mo'del.
The p'-value Lor Test 3 is less than .1. Y'OU may want t'O
corisider a different variance mo'del.
The p'-value Lor Test 7a is greater than .1. M'Odel 5 seems
t'O adeguately describe the data.
E'enchmark Dose C'Oirp'Utati'Oris :
Specified Effect = 1.000000
Risk Typ'e = Estimated standard deviations frc'in coritro'l
Confidence Level = 0.950000
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-550 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 BMD = 2.5152
3
4 BMDL = 0.00902578
5
6
7 E.3.55. Van Birgelen et al. (1995a): Plasma FT4
8 E.3.55.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
p-Value3
X1 Test
Statistic
2
1 P"
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)c
4
0.01
3.18
0.53
215.46
3.4E+02
2.2E+02
nonconstant variance,
power restricted >1
exponential (M3)
4
0.01
3.18
0.53
215.46
3.4E+02
2.2E+02
nonconstant variance,
power restricted >1
exponential (M4)
3
0.01
2.13
0.55
216.42
2.2E+02
8.3E+01
nonconstant variance,
power restricted >1
exponential (M5)
3
0.01
2.13
0.55
216.42
2.2E+02
8.3E+01
nonconstant variance,
power restricted >1
exponential (M5)d
3
0.01
2.13
0.55
216.42
2.2E+02
8.3E+01
nonconstant variance,
power unrestricted
Hill
3
0.01
2.02
0.57
216.30
1.9E+02
4.2E+01
nonconstant variance, n
restricted >1, bound hit
Hilld
2
0.01
1.87
0.39
218.16
1.7E+02
3.5E+01
nonconstant variance, n
unrestricted
linear
4
0.01
4.52
0.34
216.81
4.6E+02
3.3E+02
nonconstant variance
polynomial
4
0.01
4.52
0.34
216.81
4.6E+02
3.3E+02
nonconstant variance
power
4
0.01
4.52
0.34
216.81
4.6E+02
3.3E+02
nonconstant variance,
power restricted >1,
bound hit
power d
3
0.01
2.09
0.55
216.38
1.8E+02
3.2E+01
nonconstant variance,
power unrestricted
exponential (M2)
4
0.01
3.77
0.44
214.06
3.1E+02
2.1E+02
constant variance, power
restricted >1
exponential (M3)
4
0.01
3.77
0.44
214.06
3.1E+02
2.1E+02
constant variance, power
restricted >1
exponential (M4)
3
0.01
2.79
0.43
215.08
1.9E+02
6.7E+01
constant variance, power
restricted >1
exponential (M5)
3
0.01
2.79
0.43
215.08
1.9E+02
6.7E+01
constant variance, power
restricted >1
exponential (M5)
3
0.01
2.79
0.43
215.08
1.9E+02
6.7E+01
constant variance, power
unrestricted
Hill
3
0.01
2.58
0.46
214.87
1.6E+02
3.7E+01
constant variance, n
restricted >1, bound hit
Hill
2
0.01
2.29
0.32
216.58
1.4E+02
3.0E+01
constant variance, n
unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-551 DRAFT—DO NOT CITE OR QUOTE
-------
linear
4
0.01
5.11
0.28
215.40
4.3E+02
3.3E+02
constant variance
polynomial
4
0.01
5.11
0.28
215.40
4.3E+02
3.3E+02
constant variance
power
4
0.01
5.11
0.28
215.40
4.3E+02
3.3E+02
constant variance, power
restricted >1, bound hit
power
3
0.01
2.46
0.48
214.75
1.5E+02
2.8E+01
constant variance, power
unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
1
2
3 E.3.55.2. Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
4 Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
30
25
20
15
10
5
BMDL
BMD
0
200
400
600
800
1000
dose
5 14:31 11/20 2009
6
7
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-552 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.3.55.3. Output File for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_plasma_FT4.(d)
Gnuplot Plotting File:
Fri Nov 20 14:31:57 2009
Tbl3, plasma FT4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
lnalpha 4.29134
rho -0.423761
a 25.725
b 0.00336354
c 0.381323
d 1
Parameter Estimates
Variable Model 2
lnalpha 1.552 98
rho 0.59724
a 23.1888
b 0.00232277
c 0.391713
d 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-553 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0
8
23. 4
3.111
14
8
24 . 5
5. 657
26
8
22 . 4
2 . 828
47
8
19.3
9.334
320
8
16.3
4 .243
1024
8
10.3
4 .808
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0
22 . 66
5.554
0.3768
14
22 . 4
5.536
1 . 073
26
22 .18
5.521
0.1131
47
21. 8
5. 495
-1.286
320
17 . 4
5.17
-0.6029
1024
9.735
4 .417
0.3618
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -102.145 7 218.2901
A2 -94.04963 12 212.0993
A3 -102.143 8 220.286
R -117.8175 2 239.635
2 -103.7322 4 215.4645
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 47.54 10 < 0.0001
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-554 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Test 2
Test 3
Test 4
16.19
16.19
3.178
5
4
4
0.0 0 632
0.002778
0.52 8 4
The p-value for Test 1 is less than .05. There appears to be a
difference between resp'Orise andAor variances among the dase
levels, it seems appropriate to' model the 'data.
The p-value tor Test 2 is less than .1. A riO'ri-hO'mogerieO'US
variance model appears to' be appropriate.
The p-value tor Test 3 is less than .1. YU'U may want to'
cunsider a 'different variance model.
The p-value tor Test 4 is greater than .1. Model 2 seems
t'O adeguately describe the data.
E'enchmark DO'Se C'Oirputati'Oris :
Specified Effect = 1.000000
Risk Typ'e = Estimated standard deviations frc'in centre,1
Confidence Level = 0.950000
EMD = 3 40.74 9
EMDL = 217.397
E.3.55.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
30
25
20
15
10
5
BMDL
BMD
0
200
400
600
800
1000
dose
14:32 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-555 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.3.55.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_Unrest_BMRl_plasma_FT4.(d)
Gnuplot Plotting File:
Fri Nov 20 14:32:02 2009
Tbl3, plasma FT4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha 4.29134
rho -0.423761
a 25.725
b 0.00336354
c 0.381323
d 1
Parameter Estimates
Variable Model 5
lnalpha 1.552 98
rho 0.59724
a 23.1888
b 0.00232277
c 0.391713
d 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-556 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0
8
23. 4
3.111
14
8
24 . 5
5. 657
26
8
22 . 4
2 . 828
47
8
19.3
9.334
320
8
16.3
4 .243
1024
8
10.3
4 .808
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 23.19 5.558 0.1075
14 22.74 5.526 0.9022
26 22.36 5.498 0.01946
47 21.73 5.451 -1.261
320 15.79 4.956 0.2904
1024 10.39 4.373 -0.0587
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(1)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -102.145 7 218.2901
A2 -94.04963 12 212.0993
A3 -102.143 8 220.286
R -117.8175 2 239.635
5 -103.2077 5 216.4154
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-557 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Test 1 47.54 10 < 0.0001
Test 2 16.19 5 0.00632
Test 3 16.19 4 0.002778
Test 7a 2.129 3 0.546
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/o'r variances among the duse
levels, it seems appropriate to' model the 'data.
The p-value tor Test 2 is less than .1. A riO'ri-hO'mogerieO'US
variance model appears t'0 be appropriate.
The p-value tor Test 3 is less than .1. You may want t'0
consider a 'different variance model.
The p-value tor Test 7a is greater than .1. Model 5 seems
t'O adeguately describe the data.
E'enchmark Dcse Cc'irputaticris :
Specified Effect = 1.000000
Risk Typ'e = Estimated standard deviations from coritro'l
Confidence Level = 0.950000
E'MD = 215.664
E'MDL = 83.4 225
E.3.55.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
Hill
30
25
20
15
10
5
TBMDL
BMD
0
200
400
600
800
1000
dose
14:32 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-558 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
E.3.55.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_plasma_FT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_Unrest_BMRl_plasma_FT4.plt
Fri Nov 20 14:32:03 2009
Tbl3, plasma FT4
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 3.38957
rho = 0
intercept = 23.4
v = -13.1
n = 0.834965
k = 269.95
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -1 0.2 -0.14 -0.19 0.12
rho -1 1 -0.2 0.14 0.19 -0.12
intercept 0.2 -0.2 1 -0.35 -0.58 0.27
v -0.14 0.14 -0.35 1 0.9 -0.99
n -0.19 0.19 -0.58 0.9 1 -0.89
k 0.12 -0.12 0.27 -0.99 -0.89 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha 1.83708 2.19197 -2.4591 6.13327
rho 0.498362 0.745649 -0.963083 1.95981
intercept 23.7686 1.71713 20.403 27.1341
v -26.4457 39.0446 -102.972 50.0802
n 0.751729 0.57261 -0.370565 1.87402
k 988.089 3688.99 -6242.2 8218.37
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-559 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Data and Estimated Values of Interest
5 N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 23.4 23.8 3.11 5.52 -0.189
14 8 24.5 22.7 5.66 5.46 0.916
26 8 22.4 22.2 2.83 5.42 0.127
47 8 19.3 21.3 9.33 5.37 -1.07
320 8 16.3 15.8 4.24 4.99 0.263
1024 8 10.3 10.4 4.81 4.49 -0.043
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -102.145036 7 218.290071
A2 -94.049629 12 212.099258
A3 -102.143023 8 220.286046
fitted -103.078418 6 218.156836
R -117.817514 2 239.635028
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
47.5358
16.1908
16.1868
1.87079
10
5
4
2
<.0001
0.00632
0.002778
0.3924
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-560 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
different variance model
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 167.752
BMDL = 34.6031
E.3.55.8.
30
25
20
15
10
5
0 200 400 600 800 1000
dose
14:32 11/20 2009
E.3.55.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_plasma_FT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Pwr_Unrest_BMRl_plasma_FT4.plt
Fri Nov 20 14:32:04 2009
Power
Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-561 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Tbl3, plasma FT4
The form of the response function is:
Y[dose] = control + slope 'k doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 3.38957
rho = 0
control = 2 4.5
slope = -0.64474
power = 0.449494
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho control slope power
lalpha 1 -1 0.11 -0.098 -0.083
rho -1 1 -0.12 0.099 0.083
control 0.11 -0.12 1 -0.8 -0.75
slope -0.098 0.099 -0.8 1 0.99
power -0.083 0.083 -0.75 0.99 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
1.96489
0.456282
23.9768
-0.373722
0.52088
Std. Err.
2.14806
0.730608
1.64641
0.513355
0.188688
Lower Conf. Limit
-2.24522
-0.975684
20.7499
-1.37988
0.151059
Upper Conf. Limit
6.175
1.88825
27.2037
0.632435
0.890701
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
8
23. 4
24
3.11
5.51
-0.296
14
8
24 . 5
22 . 5
5. 66
5.43
1. 04
26
8
22 . 4
21. 9
2 . 83
5.4
0.242
47
8
19.3
21. 2
9.33
5.36
-1
320
8
16.3
16.4
4 .24
5. 06
-0.0759
1024
8
10.3
10.2
4 . 81
4 . 53
0.0903
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-562 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -102.145036 7 218.290071
A2 -94.049629 12 212.099258
A3 -102.143023 8 220.286046
fitted -103.188719 5 216.377438
R -117.817514 2 239.635028
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
47.5358
16.1908
16.1868
2.09139
10
5
4
3
<.0001
0.00632
0.002778
0.5537
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 175.43
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-563 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3 BMDL = 32.1986
4
5
6 E.3.56. Van Birgelen et al. (1995a): Plasma TT4
7 E.3.56.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/J-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
4
0.94
10.42
0.03
241.86
4.4E+02
2.8E+02
nonconstant variance,
power restricted >1
exponential (M3)
4
0.94
10.42
0.03
241.86
4.4E+02
2.8E+02
nonconstant variance,
power restricted >1
exponential (M4)
3
0.94
9.83
0.02
243.27
2.8E+02
2.1E+01
nonconstant variance,
power restricted >1
exponential (M5)
3
0.94
9.83
0.02
243.27
2.8E+02
2.0E+01
nonconstant variance,
power restricted >1
exponential (M5)
3
0.94
9.83
0.02
243.27
2.8E+02
2.0E+01
nonconstant variance,
power unrestricted
Hill
3
0.94
5.45
0.14
238.89
4.4E+01
error
nonconstant variance, n
restricted >1, bound hit
Hill
3
0.94
5.45
0.14
238.89
4.4E+01
error
nonconstant variance, n
unrestricted
linear
4
0.94
10.82
0.03
242.26
5.0E+02
3.5E+02
nonconstant variance
polynomial
4
0.94
10.82
0.03
242.26
5.0E+02
3.5E+02
nonconstant variance
power
4
0.94
10.82
0.03
242.26
5.0E+02
3.5E+02
nonconstant variance,
power restricted >1,
bound hit
power
3
0.94
8.70
0.03
242.14
1.6E+02
2.2E+01
nonconstant variance,
power unrestricted
exponential
(M2)c
4
0.94
9.83
0.04
239.86
4.4E+02
3.0E+02
constant variance,
power restricted >1
exponential (M3)
4
0.94
9.83
0.04
239.86
4.4E+02
3.0E+02
constant variance,
power restricted >1
exponential (M4)
3
0.94
9.24
0.03
241.27
2.8E+02
2.6E+01
constant variance,
power restricted >1
exponential (M5)
3
0.94
9.24
0.03
241.27
2.8E+02
2.3E+01
constant variance,
power restricted >1
exponential (M5)
d
3
0.94
9.24
0.03
241.27
2.8E+02
2.3E+01
constant variance,
power unrestricted
Hill
3
0.94
6.31
0.10
238.33
4.5E+01
error
constant variance, n
restricted >1, bound hit
Hilld
3
0.94
6.31
0.10
238.33
4.5E+01
error
constant variance, n
unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-564 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
linear
4
0.94
10.23
0.04
240.26
5.0E+02
3.7E+02
constant variance
polynomial
4
0.94
10.23
0.04
240.26
5.0E+02
3.7E+02
constant variance
power
4
0.94
10.23
0.04
240.26
5.0E+02
3.7E+02
constant variance,
power restricted >1,
bound hit
power d
3
0.94
8.14
0.04
240.16
1.7E+02
2.4E+01
constant variance,
power unrestricted
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
d Alternate model also presented in this appendix
1
2
3 E.3.56.2. Figure for Selected Model: Exponential (M2), Constant Variance, Power Restricted
4 >1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
45
40
35
30
25
20
BMD
BMDI
0
200
400
600
800
1000
dose
5 14:32 11/20 2009
6
7
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-565 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
E.3.56.3. Output File for Selected Model: Exponential (M2), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_CV_BMRl_plasma_TT4.(d)
Gnuplot Plotting File:
Fri Nov 20 14:32:54 2009
Tbl3, plasma TT4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
Specified
lnalpha 3.66719
rho(S) 0
a 43.47
b 0.00268876
c 0.558678
d 1
Parameter Estimates
Variable Model 2
lnalpha 3.85975
rho 0
a 39.9223
b 0.00192618
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-566 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
c 0.587293
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 40.9 6.788
14 8 41.4 5.374
26 8 41.4 6.505
47 8 32.3 7.354
320 8 33.6 6.223
1024 8 25.5 7.637
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 39.36 6.931 0.6265
14 39.12 6.931 0.9302
26 38.91 6.931 1.015
47 38.55 6.931 -2.551
320 34.15 6.931 -0.2227
1024 24.97 6.931 0.2158
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma'-2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(1)
Var{e(ij)} = Sigma'-2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -112.0125 7 238.025
A2 -111.4015 12 246.8029
A3 -112.0125 7 238.025
R -127.4455 2 258.891
2 -116.929 3 239.858
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adeguately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-567 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Test
-2*log(Likelihood Ratio)
D. F.
p-value
Test 1
Test 2
Test 3
Test 4
32 .09
1. 222
1. 222
9. 833
10
5
5
4
0.0003871
0.9427
0.9427
0.04334
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is less than .1. Model 2 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD
435.731
BMDL
296.489
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-568 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E.3.56.4. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
45
40
35
30
25
20
BMDL
BMD
0
200
400
600
800
1000
dose
14:33 11/20 2009
E.3.56.5. Output File for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_CV_Unrest_BMRl_plasma_TT4.(d)
Gnuplot Plotting File:
Fri Nov 20 14:33:03 2009
Tbl3, plasma TT4
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-569 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 5
lnalpha
rho(S)
3.66719
0
43.47
0.00268876
0.558678
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 5
3. 85975
0
39.9223
0. 00192618
0.587293
1
Table of Stats From Input Data
0
14
26
47
320
1024
Obs Mean
40. 9
41. 4
41. 4
32 . 3
33. 6
25.5
Obs Std Dev
6.788
5.374
6.505
7 . 354
6.223
7 . 637
Estimated Values of Interest
Dose
Est Mean
Est Std
Scaled Residual
0
39. 92
6
889
0.4014
14
39.48
6
889
0.7867
26
39.12
6
889
0.9372
47
38 . 5
6
889
-2.544
320
32 . 34
6
889
0.5167
1024
25.74
6
889
-0.09785
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-570 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Var{e(ij)} = Sigma'-2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -112.0125 7 238.025
A2 -111.4015 12 246.8029
A3 -112.0125 7 238.025
R -127.4455 2 258.891
5 -116.634 4 241.268
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
32 .09
1. 222
1. 222
9.243
D. F.
10
5
5
3
p-value
0.0003871
0.9427
0.9427
0.02623
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is less than .1. Model 5 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 281.101
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-571 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
E'.MDL =
23.4709
E.3.56.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model
45
40
25
20
35 -
30
200
400
600
800
1000
dose
14:33 11/20 2009
E.3.56.7. Output File for Unrestricted Model: Hill, constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_CV_Unrest_BMRl_plasma_TT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_CV_Unrest_BMRl_plasma_TT4.plt
Fri Nov 20 14:33:05 2009
Tbl3, plasma TT4
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-572 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
rho
intercept
44 . 7333
0
40. 9
-15. 4
2.59801
44.9231
Specified
Asymptotic Correlation Matrix of Parameter Estimates
alpha
intercept
The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
1
1.6e-0 0 9
-le-009
8.3e-008
intercept
1. 6e-009
1
-0. 63
-0.12
v
-le-009
-0. 63
1
-0.29
k
S. 3e-008
-0.12
-0.29
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
intercept
k
Estimate
44.637
41.2336
-11.6836
18
44 . 0222
Std. Err.
9.11149
1. 36385
2 .15625
NA
3.14538
Lower Conf. Limit
26.7788
38.5605
-15.9098
37 . 8573
Upper Conf. Limit
62.4952
43.9067
-7 .45747
50.187
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
8
40. 9
41. 2
6.79
CO
-0.141
14
8
41. 4
41. 2
5.37
6.68
0.0704
26
8
41. 4
41. 2
6.51
CO
0.0708
47
8
32 3
32 . 3
7 . 35
CO
-3.05e-005
320
8
33. 6
29.5
6.22
CO
1.71
1024
8
25.5
29.5
7 . 64
CO
-1 .71
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-573 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'"2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -112.012501 7 238.025002
A2 -111.401462 12 246.802924
A3 -112.012501 7 238.025002
fitted -115.165512 4 238.331023
R -127.445484 2 258.890968
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 32.088 10 0.0003871
Test 2 1.22208 5 0.9427
Test 3 1.22208 5 0.9427
Test 4 6.30602 3 0.09763
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 44.7355
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-574 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.56.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
45
40
35
30
25
20
BMD
BMDI
0
200
400
600
800
1000
dose
14:33 11/20 2009
E.3.56.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_CV_Unrest_BMRl_plasma_TT4.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Pwr_CV_Unrest_BMRl_plasma_TT4.plt
Fri Nov 20 14:33:06 2009
Tbl3, plasma TT4
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-575 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
alpha =
rho =
control =
slope =
power =
44 . 7333
0
41. 4
-4 .42652
0.155038
Specified
Asymptotic Correlation Matrix of Parameter Estimates
alpha
control
The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
1
5. 3e-011
control
5 . 3e-011
1
slope
3 . 5e-010
-0. 81
power
5. 9e-010
-0.76
slope 3.5e-010 -0.81 1 0.99
power 5.9e-010 -0.76 0.99 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
46.3718
41.441
-0.626626
0. 464532
Std. Err.
9. 4656
2 .1824
0. 890104
0.195802
Lower Conf. Limit
27 . 8195
37 .1636
-2.3712
0.0807676
Upper Conf. Limit
64.924
45.7184
1.11795
0.848296
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 40.9 41.4
14 8 41.4 39.3
26 8 41.4 38.6
47 8 32.3 37.7
320 8 33.6 32.3
1024 8 25.5 25.8
6.79 6.81 -0.225
5.37 6.81 0.87
6.51 6.81 1.17
7.35 6.81 -2.24
6.22 6.81 0.538
7.64 6.81 -0.108
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-576 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
Model Log(likelihood) # Param's AIC
A1 -112.012501 7 238.025002
A2 -111.401462 12 246.802924
A3 -112.012501 7 238.025002
fitted -116.080583 4 240.161165
R -127.445484 2 258.890968
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2^log(Likelihood Ratio) Test df p-value
Test 1 32.088 10 0.0003871
Test 2 1.22208 5 0.9427
Test 3 1.22208 5 0.9427
Test 4 8.13616 3 0.04328
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 17 0.004
BMDL = 24.0807
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-577 DRAFT—DO NOT CITE OR QUOTE
-------
1 E.3.57. White et al. (1986): CH50
2 E.3.57.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/7-Value
a
X1 Test
Statistic
12P-
Value"
AIC
BMD
(ng/kg-
d)
BMDL
(ng/kg-
d)
Model Notes
exponential (M2)
5
0.09
20.99
0.00
391.47
4.5E+02
2.8E+02
nonconstant variance,
power restricted >1
exponential (M3)
5
0.09
20.99
0.00
391.47
4.5E+02
2.8E+02
nonconstant variance,
power restricted >1
exponential (M4)
4
0.09
19.65
0.00
392.13
3.1E+02
1.1E+02
nonconstant variance,
power restricted >1
exponential (M5)
4
0.09
19.65
0.00
392.13
3.1E+02
1.1E+02
nonconstant variance,
power restricted >1
Hill
4
0.09
18.75
0.00
391.22
2.0E+02
3.6E+01
nonconstant variance, n
restricted >1, bound hit
linear
5
0.09
25.95
<.0001
396.43
8.1E+02
5.9E+02
nonconstant variance
polynomial
5
0.09
25.95
<.0001
396.43
8.1E+02
5.9E+02
nonconstant variance
power
5
0.09
25.95
<.0001
396.43
8.1E+02
5.9E+02
nonconstant variance,
power restricted >1, bound
hit
exponential (M2)
5
0.09
21.77
0.00
390.45
4.0E+02
2.6E+02
constant variance, power
restricted >1
exponential (M3)
5
0.09
21.77
0.00
390.45
4.0E+02
2.6E+02
constant variance, power
restricted >1
exponential (M4)
4
0.09
20.51
0.00
391.19
2.7E+02
7.2E+01
constant variance, power
restricted >1
exponential (M5)
4
0.09
20.51
0.00
391.19
2.7E+02
7.2E+01
constant variance, power
restricted >1
Hillc
4
0.09
19.30
0.00
389.98
1.3E+02
2.9E+01
constant variance, n
restricted >1, bound hit
linear
5
0.09
26.50
<.0001
395.18
7.3E+02
5.7E+02
constant variance
polynomial
5
0.09
26.50
<.0001
395.18
7.3E+02
5.7E+02
constant variance
power
5
0.09
26.50
<.0001
395.18
7.3E+02
5.7E+02
constant variance, power
restricted >1, bound hit
a Values <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
b Values <0.1 fail to meet BMDS goodness-of-fit criteria
c Best-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
3
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-578 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
E.3.57.2. Figure for Selected Model: Hill, Constant Variance, n Restricted >1, Bound Hit
Hill Model with 0.95 Confidence Level
Hill
100
80
60
40
20
0
ISMDI
IMD
0
500
1000
1500
2000
dose
19:52 10/06 2009
E.3.57.3. Output File for Selected Model: Hill, Constant Variance, n Restricted >1, Bound
Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_CH50.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AniDose\HillConstVar_BMRl_CH50.plt
Tue Oct 06 19:52:50 2009
[insert study notes]
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 7
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-579 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Default Initial Parameter Values
alpha
rho
intercept
273.143
0
91
-74
0.0969998
10
Specified
Asymptotic Correlation Matrix of Parameter Estimates
alpha
intercept
The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
1
-1. 4e-008
-3.3e-008
6.7 e-0 0 9
intercept
-1.4e-008
1
0.38
-0.8
v
-3.3e-008
0.38
1
-0. 81
k
5.7 e-00 9
-0.8
-0. 81
1
Parameter Estimates
Variable
alpha
intercept
k
Estimate
337.326
73.1945
-58.2543
1
289.939
Std. Err.
63.7486
6.21329
12.308
NA
354.891
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
212.381
61.0167
-82.3776
-405.635
462.271
85.3723
-34.131
985.512
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
8
91
73.2
14 .1
18
4
2.74
10
8
54
71. 3
8.49
18
4
-2 . 66
50
8
63
64 . 6
11. 3
18
4
-0.25
100
8
56
58 . 3
25.5
18
4
-0.347
500
8
41
36.3
17
18
4
0.72
1000
8
32
28
17
18
4
0. 611
2000
8
17
22 . 3
17
18
4
-0.819
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-580 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -181.340979 8 378.681959
A2 -175.820265 14 379.640529
A3 -181.340979 8 378.681959
fitted -190.989397 4 389.978794
R -212.367055 2 428.734109
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 73.0936 12 <.0001
Test 2 11.0414 6 0.0871
Test 3 11.0414 6 0.0871
Test 4 19.2968 4 0.0006871
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. Consider running a
non-homogeneous variance model
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 133.503
BMDL = 28.903
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-581 DRAFT—DO NOT CITE OR QUOTE
-------
E.4. REFERENCES
Amin, S; Moore, RW; Peterson, RE; et al. (2000) Gestational and lactational exposure to TCDD
or coplanar PCBs alters adult expression of saccharin preference behavior in female rats.
Neurotoxicol Teratol 22(5):675-682.
Bell, DR; Clode, S; Fan, MQ; et al. (2007a) Toxicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin in
the developing male Wistar(Han) rat. II: Chronic dosing causes developmental delay. Toxicol
Sci 99(l):224-233.
Bell, DR; Clode, S; Fan, MQ; et al. (2007b) Relationships between tissue levels of 2,3,7,8-
tetrachlorodibenzo-p-dioxin (TCDD), mRNAs, and toxicity in the developing male Wistar (Han)
rat. Toxicol Sci 99(2):591-604.
Cantoni, L; Salmona, M; Rizzardini, M. (1981) Porphyrogenic effect of chronic treatment with
2,3,7,8-tetrachlorodibenzo-p-dioxin in female rates. Dose-effect relationship following urinary
excretion of porphyrins. Toxicol Appl Pharmacol 57:156-157.
Crofton, KM; Craft, ES; Hedge, JM; et al. (2005) Thyroid-hormone-disrupting chemicals:
evidence for dose-dependent additivity or synergism. Environ Health Perspect 113(11): 1549—
1554.
DeCaprio, AP; McMartin, DN; O'Keefe, PE; et al. (1986) Subchronic oral toxicity of 2,3,7,8-
tetrachlorodibenzo-p-dioxin in the guinea pig: comparisons with a PCB-containing transformer
fluid pyrolysate. Fund Appl Toxicol 6:454-463.
Hojo, R; Stern, S; Zareba, G; et al. (2002) Sexually dimorphic behavioral responses to prenatal
dioxin exposure. Environ Health Perspect 110(3):247-254.
Kattainen, H; Tuukanan, J; Simanainen, U; et al. (2001) In utero/lactational 2,3,7,8-
tetrachlorodibenzo-p-dioxin exposure impairs molar tooth development in rats. Toxicol Appl
Pharmacol 17:216-224.
Keller, JM; Huet-Hudson, YM; Leamy, LJ. (2007) Qualitative effects of dioxin on molars vary
among inbred mouse strains. Arch Oral Biol 52:450-454.
Keller, JM; Zelditch, ML; Huet, YM; et al. (2008a) Genetic differences in sensitivity to
alterations of mandible structure caused by the teratogen 2,3,7,8-tetrachlorodibenzo-p-dioxin.
Toxicol Pathol 36:1006-1013.
Keller, JM; Huet-Hudson, Y; Leamy, LJ. (2008b) Effects of 2,3,7,8-tetrachlorodibenzo-p-dioxin
on molar development among non-resistant inbred strains of mice: a geometric morphometric
analysis. Growth Devel Aging 71:3-16.
Kociba, RJ; Keyes, DG; Beyer, JE; et al. (1978) Results of a two-year chronic toxicity and
oncogenicity study of 2,3,7,8-tetrachlorodibenzo-p-dioxin in rats. Toxicol Appl Pharmacol
46(2):279-303.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-582 DRAFT—DO NOT CITE OR QUOTE
-------
Latchoumycandane, C; Mathur, PP. (2002) Effects of vitamin E on reactive oxygen species-
mediated 2,3,7,8-tetrachlorodi-benzo-p-dioxin toxicity in rat testis. J Appl Toxicol 22(5):345-
351.
Li, B; Liu, H-Y; Dai, L-J; et al. (2006) The early embryo loss caused by 2,3,7,8-
tetrachl orodibenzo-/>di oxi n may be related to the accumulation of this compound in the uterus.
Reprod Toxicol 21:301-306.
Markowski, VP; Zareba, G; Stern, S; et al. (2001) Altered operant responding for motor
reinforcement and the determination of benchmark doses following perinatal exposure to low-
level 2,3,7,8-tetrachlorodibenzo-p-dioxin. Environ Health Perspect 109(6):621-627.
Miettinen, HM; Sorvari, R; Alaluusua, S; et al. (2006) The Effect of perinatal TCDD exposure
on caries susceptibility in rats. Toxicol Sci 91(2):568-575.
NTP (National Toxicology Program). (1982) NTP Technical Report on carcinogenesis bioassay
of 2,3,7,8-tetrachlorodibenzo-p-dioxin in Osborne-Mendel rats and B6C3F1 mice (gavage
study). Public Health Service, U.S. Department of Health and Human Services; NTP TR 209.
Available from the National Institute of Environmental Health Sciences, Research Triangle Park,
NC.
NTP (National Toxicology Program). (2006) Toxicology and carcinogenesis studies of a mixture
of 2,3,7,8-tetrachlorodibenzo-/?-dioxin (TCDD) (CAS No. 1746-01-6), 2,3,4,7,8-
pentachlorodibenzofuran (PeCDF) (CAS No. 57117-31-4), and 3,3',4,4',5-pentachlorobiphenyl
(PCB 126) (CAS No. 57465-28-8) in female Harlan Sprague-Dawley Rats (gavage studies).
Public Health Service, U.S. Department of Health and Human Services; NTP TR 526. Available
from the National Institute of Environmental Health Sciences, Research Triangle Park, NC.
Available at: http://ntp.niehs.nih.gov/index.cfm?objectid=070B7300-0E62-BF 12-
F4C3E3B5B645A92B
Ohsako, S; Miyabara, Y; Nishimura, N; et al. (2001) Maternal exposure to a low dose of 2,3,7,8-
tetrachl orodibenzo-p-di oxi n (TCDD) suppressed the development of reproductive organs of male
rats: dose-dependent increase of mRNA levels of 5a-reductase type 2 in contrast to decrease of
androgren receptor in the pubertal ventral prostate. Toxicol Sci 60:132-143.
Schantz, SL; Seo, BW; Moshtaghian, J; et al. (1996) Effects of gestational and lactational
exposure to TCDD or coplanar PCBs on spatial learning. Neurotoxicol Teratol 18(3):305—313.
Shi, Z; Valdez, KE; Ying, AY; et al. (2007) Ovarian endocrine disruption underlies premature
reproduction senescence following environmentally relevant chronic exposure to aryl
hydrocarbon receptor agonist 2,3,7,8-tetrachlorodibenzo-p-dioxin. Biol Reprod 30(4):293-342.
Smialowicz, RJ; DeVito, MJ; Williams, WC; et al. (2008) Relative potency based on hepatic
enzyme induction predicts immunosuppressive effects of a mixture of PCDDS/PCDFS and
PCBS. Toxicol Appl Pharmacol 227:477-484.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-583 DRAFT—DO NOT CITE OR QUOTE
-------
Toth, KJ; Sugar, S; Somfai-Relle, S; et al. (1978) Carcinogenic bioassay of the herbicide 2,4,5-
trichlorphenoxy ethanol (TCPE) with Swiss mice. Prog Biochem Pharmacol 14:82-93.
Toth, L; Somfai-Relle, S; Sugar, J; et al. (1979) Carcinogenicity testing of herbicide 2,4,5-
trichlorophenoxyethanol containing dioxin and of pure dioxin in Swiss mice. Nature
278:548-549.
Van Birgelen, AP; Van der Kolk, J; Fase, KM; et al. (1995) Subchronic dose-response study of
2,3,7,8-tetrachlorodibenzo-p-dioxin in female Sprague-Dawley rats. Toxicol Appl Pharmacol
132:1-13.
White, KL, Jr; Lysy, HH; McCay, JA; et al. (1986) Modulation of serum complement levels
following exposure to polychlorinated dibenzo-p-dioxins. Toxicol Appl Pharmacol 84:209-219.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 E-584 DRAFT—DO NOT CITE OR QUOTE
-------
DRAFT
DO NOT CITE OR QUOTE
January 2010
Agency/Interagency Review Draft
APPENDIX F
Cancer Benchmark Dose Modeling
NOTICE
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
CONTENTS—APPENDIX F: Cancer Benchmark Dose Modeling
APPENDIX F. CANCER BENCHMARK DOSE MODELING F-l
F.l. BLOOD SERUM BMDS RESULTS F-l
F.l.l. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma
of Hard Palate or Nasal Turbinates F-l
F.l. 1.1. Summary Table of BMDS Modeling Results F-l
F.l. 1.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-2
F.l.1.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-2
F.1.2. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma
of Tongue F-4
F.l.2.1. Summary Table of BMDS Modeling Results F-4
F. 1.2.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-4
F. 1.2.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-5
F.1.3. Kociba et al. (1978): Female, Adenoma of Adrenal Cortex F-6
F.1.3.1. Summary Table of BMDS Modeling Results F-6
F. 1.3.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-7
F.1.3.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-7
F.1.4. Kociba et al. (1978): Female, Hepatocellular Adenoma(s) or
Carcinoma(s) F-9
F.l.4.1. Summary Table of BMDS Modeling Results F-9
F. 1.4.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-9
F. 1.4.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-10
F.1.5. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma
of Hard Palate or Nasal Turbinates F-l 1
F.l.5.1. Summary Table of BMDS Modeling Results F-l 1
F. 1.5.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-12
F.1.5.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-12
F.1.6. Kociba et al. (1978): Female, Keratinizing Squamous Cell
Carcinoma of Lung F-14
F.l.6.1. Summary Table of BMDS Modeling Results F-14
F. 1.6.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-15
F. 1.6.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-l5
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-ii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
F.1.7. National Toxicology Program (1982): Female Rat, Subcutaneous
Tissue. Fibrosarcoma F-17
F. 1.7.1. Summary Table ofBMDS Modeling Results F-17
F. 1.7.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-17
F. 1.7.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-18
F.1.8. National Toxicology Program (1982): Female Rat, Liver, Neoplastic
Nodule or Hepatocellular Carcinoma F-19
F. 1.8.1. Summary Table ofBMDS Modeling Results F-19
F. 1.8.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-20
F.1.8.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-20
F.1.9. National Toxicology Program (1982): Female Rat, Adrenal, Cortical
Adenoma, or Carcinoma or Adenoma, NOS F-22
F. 1.9.1. Summary Table ofBMDS Modeling Results F-22
F. 1.9.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-22
F. 1.9.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-23
F.1.10. National Toxicology Program (1982): Female Rat, Thyroid,
Follicular-Cell Adenoma F-24
F. 1.10.1. Summary Table ofBMDS Modeling Results F-24
F.1.10.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-25
F.1.10.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-25
F. 1.11. National Toxicology Program (1982): Male Rat, Liver, Neoplastic
Nodule or Hepatocellular Carcinoma F-27
F. 1.11.1. Summary Table ofBMDS Modeling Results F-27
F. 1.11.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-28
F.1.11.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-28
F. 1.12. National Toxicology Program (1982): Male Ra, Thyroid,
Follicular-Cell Adenoma or Carcinoma F-30
F. 1.12.1. Summary Table ofBMDS Modeling Results F-30
F. 1.12.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-31
F.1.12.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-31
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-iii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
F.1.13. National Toxicology Program (1982): Male Rat, Adrenal Cortex,
Adenoma F-33
F.1.13.1. Summary Table ofBMDS Modeling Results F-33
F.1.13.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-34
F.1.13.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-34
F.1.14. National Toxicology Program (1982): Female Mice, Subcutaneous
Tissue, Fibrosarcoma F-36
F. 1.14.1. Summary Table ofBMDS Modeling Results F-36
F.1.14.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-37
F.1.14.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-37
F.1.15. National Toxicology Program (1982): Female Mice, Hematopoietic
System, Lymphoma F-39
F.1.15.1. Summary Table ofBMDS Modeling Results F-39
F.1.15.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-40
F.1.15.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-40
F.1.16. National Toxicology Program (1982): Female Mice, Liver,
Hepatocellular Adenoma or Carcinoma F-42
F. 1.16.1. Summary Table ofBMDS Modeling Results F-42
F.1.16.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-43
F.1.16.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-43
F.1.17. National Toxicology Program (1982): Female Mice, Thyroid,
Follicular-Cell Adenoma F-45
F. 1.17.1. Summary Table ofBMDS Modeling Results F-45
F.1.17.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-46
F.1.17.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-46
F. 1.18. National Toxicology Program (1982): Male Mice, Lung,
Alveolar/Bronchiolar Adenoma or Carcinoma F-48
F.1.18.1. Summary Table ofBMDS Modeling Results F-48
F.1.18.2. Figure for Selected Model: Multistage Cancer, 2-Degree,
Betas Restricted >0 F-49
F.1.18.3. Output File for Selected Model: Multistage Cancer,
2-Degree, Betas Restricted >0 F-49
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-iv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
F. 1.19. National Toxicology Program (1982): Male Mice, Liver,
Hepatocellular Adenoma or Carcinoma F-51
F. 1.19.1. Summary Table of BMDS Modeling Results F-51
F. 1.19.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-52
F.1.19.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-52
F.1.20. National Toxicology Program (2006): Liver, Cholangiocarcinoma F-54
F. 1.20.1. Summary Table of BMDS Modeling Results F-54
F. 1.20.2. Figure for Selected Model: Multistage Cancer, 3-Degree,
Betas Restricted >0 F-55
F.1.20.3. Output File for Selected Model: Multistage Cancer,
3-Degree, Betas Restricted >0 F-55
F. 1.21. National Toxicology Program (2006): Liver, Hepatocellular
Adenoma F-5 7
F. 1.21.1. Summary Table of BMDS Modeling Results F-57
F. 1.21.2. Figure for Selected Model: Multistage Cancer, 3-Degree,
Betas Restricted >0 F-58
F.1.21.3. Output File for Selected Model: Multistage Cancer,
3-Degree, Betas Restricted >0 F-58
F.1.22. National Toxicology Program (2006): Oral mucosa, Squamous Cell
Carcinoma F-60
F. 1.22.1. Summary Table of BMDS Modeling Results F-60
F. 1.22.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0, Bound Hit F-61
F. 1.22.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0, Bound Hit F-61
F.1.23. National Toxicology Program (2006): Pancreas, Adenoma or
Carcinoma F-63
F.1.23.1. Summary Table of BMDS Modeling Results F-63
F. 1.23.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-64
F.1.23.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-64
F.1.24. National Toxicology Program (2006): Lung, Cystic Keratinizing
Epithelioma F-66
F. 1.24.1. Summary Table of BMDS Modeling Results F-66
F. 1.24.2. Figure for Selected Model: Multistage Cancer, 2-Degree,
Betas Restricted >0 F-67
F. 1.24.3. Output File for Selected Model: Multistage Cancer,
2-Degree, Betas Restricted >0 F-67
F.1.25. Toth etal. (1978): 1YR, Liver, Tumors F-69
F. 1.25.1. Summary Table of BMDS Modeling Results F-69
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-v DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
F. 1.25.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-70
F.1.25.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-70
F.2. ADMINISTERED DOSE BMDS RESULTS F-72
F.2.1. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma
of Hard Palate or Nasal Turbinates F-72
F.2.1.1. Summary Table of BMDS Modeling Results F-72
F.2.1.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-73
F.2.1.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-73
F.2.2. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma
of Tongue F-75
F.2.2.1. Summary Table of BMDS Modeling Results F-75
F.2.2.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-76
F.2.2.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-76
F.2.3. Kociba et al. (1978): Female. Adenoma of Adrenal Cortex F-78
F.2.3.1. Summary Table of BMDS Modeling Results F-78
F.2.3.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-79
F.2.3.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-79
F.2.4. Kociba et al. (1978): Female, Hepatocellular Adenoma(S) or
Carcinoma(s) F-81
F.2.4.1. Summary Table of BMDS Modeling Results F-81
F.2.4.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-82
F.2.4.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-82
F.2.5. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma
of Hard Palate or Nasal Turbinates F-84
F.2.5.1. Summary Table of BMDS Modeling Results F-84
F.2.5.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-85
F.2.5.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-85
F.2.6. Kociba et al. (1978): Female, Keratinizing Squamous Cell
Carcinoma of Lung F-87
F.2.6.1. Summary Table of BMDS Modeling Results F-87
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-vi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
F.2.6.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-88
F.2.6.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-88
F.2.7. National Toxicology Program (1982): Female Rats, Subcutaneous
Tissue, Fibrosarcoma F-90
F.2.7.1. Summary Table of BMDS Modeling Results F-90
F.2.7.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-91
F.2.7.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-91
F.2.8. National Toxicology Program (1982): Female Rats, Liver,
Neoplastic Nodule or Hepatocellular Carcinoma F-93
F.2.8.1. Summary Table of BMDS Modeling Results F-93
F.2.8.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-94
F.2.8.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-94
F.2.9. National Toxicology Program (1982): Female Rats, Adrenal,
Cortical Adenoma, or Carcinoma or Adenoma, NOS F-96
F.2.9.1. Summary Table of BMDS Modeling Results F-96
F.2.9.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-97
F.2.9.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-97
F.2.10. National Toxicology Program (1982): Female Rats, Thyroid,
Follicular-Cell Adenoma F-99
F.2.10.1. Summary Table of BMDS Modeling Results F-99
F.2.10.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-100
F.2.10.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-100
F.2.11. National Toxicology Program (1982): Male Rats, Liver, Neoplastic
Nodule or Hepatocellular Carcinoma F-102
F.2.11.1. Summary Table of BMDS Modeling Results F-102
F.2.11.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-103
F.2.11.3. Output File for selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-103
F.2.12. National Toxicology Program (1982): Male Rats, Thyroid,
Follicular-Cell Adenoma or Carcinoma F-105
F.2.12.1. Summary Table of BMDS Modeling Results F-105
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-vii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
F.2.12.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-106
F.2.12.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-106
F.2.13. National Toxicology Program (1982): Male Rats, Adrenal cortex,
Adenoma F-108
F.2.13.1. Summary Table ofBMDS Modeling Results F-108
F.2.13.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-109
F.2.13.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-109
F.2.14. National Toxicology Program (1982): Female Mice, Subcutaneous
Tissue, Fibrosarcoma F-l 11
F.2.14.1. Summary Table ofBMDS Modeling Results F-l 11
F.2.14.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-l 12
F.2.14.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-l 12
F.2.15. National Toxicology Program (1982): Female Mice, Hematopoietic
System, Lymphoma F-l 14
F.2.15.1. Summary Table ofBMDS Modeling Results F-114
F.2.15.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-l 15
F.2.15.3. Output File for Selected Model: Multistage Cancer,
1 -Degree, Betas Restricted >0 F-l 15
F.2.16. National Toxicology Program (1982): Female Mice, Liver,
Hepatocellular Adenoma or Carcinoma F-l 17
F.2.16.1. Summary Table ofBMDS Modeling Results F-l 17
F.2.16.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-l 18
F.2.16.3. Output File for Selected Model: Multistage Cancer,
1 -Degree, Betas Restricted >0 F-l 18
F.2.17. National Toxicology Program (1982): Female Mice, Thyroid
Follicular Cell Adenoma F-120
F.2.17.1. Summary Table ofBMDS Modeling Results F-120
F.2.17.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-121
F.2.17.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-121
F.2.18. National Toxicology Program (1982): Male Mice, Lung,
Alveolar/Bronchiolar Adenoma or Carcinoma F-123
F.2.18.1. Summary Table ofBMDS Modeling Results F-123
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-viii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
F.2.18.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-124
F.2.18.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-124
F.2.19. National Toxicology Program (1982): Male Mice, Liver,
Hepatocellular Adenoma or Carcinoma F-126
F.2.19.1. Summary Table ofBMDS Modeling Results F-126
F.2.19.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-127
F.2.19.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-127
F.2.20. National Toxicology Program (2006): Liver, Cholangiocarcinoma F-129
F.2.20.1. Summary Table ofBMDS Modeling Results F-129
F.2.20.2. Figure for Selected Model: Multistage Cancer, 3-Degree,
Betas Restricted >0 F-130
F.2.20.3. Output File for Selected Model: Multistage Cancer,
3-Degree, Betas Restricted >0 F-130
F.2.21. National Toxicology Program (2006): Liver, Hepatocellular
Adenoma F-132
F.2.21.1. Summary Table ofBMDS Modeling Results F-132
F.2.21.2. Figure for Selected Model: Multistage Cancer, 3-Degree,
Betas Restricted >0 F-133
F.2.21.3. Output File for Selected Model: Multistage Cancer,
3-Degree, Betas Restricted >0 F-133
F.2.22. National Toxicology Program (2006): Oral Mucosa, Squamous Cell
Carcinoma F-135
F.2.22.1. Summary Table ofBMDS Modeling Results F-135
F.2.22.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0, Bound Hit F-136
F.2.22.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0, Bound Hit F-136
F.2.23. National Toxicology Program (2006): Pancreas, Adenoma or
Carcinoma F-138
F.2.23.1. Summary Table ofBMDS Modeling Results F-138
F.2.23.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-139
F.2.23.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-139
F.2.24. National Toxicology Program (2006): Lung, Cystic Keratinizing
Epitheli oma F-141
F.2.24.1. Summary Table ofBMDS Modeling Results F-141
F.2.24.2. Figure for Selected Model: Multistage Cancer, 2-Degree,
Betas Restricted >0 F-142
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-ix DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
F.2.24.3. Output File for Selected Model: Multistage Cancer,
2-Degree, Betas Restricted >0 F-142
F.2.25. Toth et al. (1978): 1YR, Liver, Tumors F-144
F.2.25.1. Summary Table ofBMDS Modeling Results F-144
F.2.25.2. Figure for Selected Model: Multistage Cancer, 1-Degree,
Betas Restricted >0 F-145
F.2.25.3. Output File for Selected Model: Multistage Cancer,
1-Degree, Betas Restricted >0 F-145
F.3. HUMAN EQUIVALENT DOSES FOR 1, 5, AND 10% EXTRA RISK F-148
F.4. REFERENCES F-149
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-x DRAFT—DO NOT CITE OR QUOTE
-------
APPENDIX F.
CANCER BENCHMARK DOSE MODELING
F.l. BLOOD SERUM BMDS RESULTS
F.l.l. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma of Hard Palate
or Nasal Turbinates
F. 1.1.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.94
0.82
31.56
3.2E+03
1.5E+03
betas restricted >0
Multistage
cancer,
2-degree
3
0.15
0.99
30.17
7.5E+03
1.9E+03
betas restricted >0
Multistage
cancer,
3-degree
3
0.03
1.00
29.93
1.1E+04
2.0E+03
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
F.l.1.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.2
0.15
0.05
0
BMDL
BMD
0 5000 10000 15000 20000
dose
15:42 11/23 2009
F.l.1.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_palate_nasal.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_palate_nasal.plt
Mon Nov 23 15:42:08 2009
Source - Table 4
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 5
Total number of records with missing values = 1
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 4 .1047e-006
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(l) 3.16499e-006 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-13.9385
-14.782
-20.2589
Param's
4
1
1
Deviance Test d.f.
1.68697
12.6409
P-value
0. 6398
0. 005481
AIC:
31. 5639
Est. Prob.
Goodness of Fit
Expected Observed
Scaled
Residual
0.0000
860.4590
3944.9299
21334.0000
0.0000
0.0027
0.0124
0.0653
0. 000
0.136
0. 620
3.265
0. 000
0. 000
0. 000
4 . 000
85
50
50
50
0. 000
-0.369
-0.793
0. 421
Chi'" 2
0. 94
d.f.
P-value
0. 8153
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 3175.47
BMDL = 1539.87
BMDU = 8231.22
Taken together, (1539.87, 8231.22) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-3 draft—do not cite or quote
-------
1
2 Multistage Cancer Slope Factor = 6.4 94 06e-006
3
4
5 F.1.2. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma of Tongue
6 F. 1.2.1. Summary Table of BMDS Modeling Results
l
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
1.50
0.47
47.93
3.4E+03
1.4E+03
betas restricted >0
Multistage
cancer,
2-degree
2
1.50
0.47
47.93
3.4E+03
1.4E+03
betas restricted >0
Multistage
cancer,
3-degree
2
1.50
0.47
47.93
3.4E+03
1.4E+03
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
8
9
10 F.l.2.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
11
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.15
0.1
0.05
0
BMD
BMDI
0
5000
10000
15000
20000
dose
15:42 11/23 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
F. 1.2.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov2 3\Blood\mscl_ngkgd_tongue.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_tongue.plt
Mon Nov 23 15:42:27 2009
Source - Table 4
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal^dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 5
Total number of records with missing values = 1
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.00925136
Beta(1) = 2.49061e-006
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.58
Beta(1) -0.58 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.00510493 * * *
Beta(l) 2.99496e-006 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -21.1523 4
Fitted model -21.9667 2 1.6288 2 0.4429
Reduced model -24.1972 1 6.08976 3 0.1073
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-5 draft—do not cite or quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
AIC:
47 .9334
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0051
0. 434
0. 000
85
-0.660
860.4590
0.0077
0.383
1. 000
50
1. 000
3944.9299
0.0168
0.840
1. 000
50
0.177
21334.0000
0.0667
3.334
3. 000
50
-0.189
Chi/S2 = 1.50
d. f.
= 2
P-
-value = 0.4716
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 3355.7 5
BMDL = 14 32.78
BMDU = 19112.8
Taken together, (1432.78, 19112.8) is a 90 % two-sided confidence
interval for the BMD
Multistage Cancer Slope Factor = 6.97946e-006
F.1.3. Kociba et al. (1978): Female, Adenoma of Adrenal Cortex
F. 1.3.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
1.09
0.78
52.49
1.8E+03
1.0E+03
betas restricted >0
Multistage
cancer,
2-degree
3
1.09
0.78
52.49
1.8E+03
1.0E+03
betas restricted >0
Multistage
cancer,
3-degree
3
1.09
0.78
52.49
1.8E+03
1.0E+03
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F. 1.3.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.2
0.15
0.05
BMDL BMD
J i ¦ I l_L I L.
0
J I L_
5000
i .
15000
_j i .
20000
10000
dose
15:42 11/23 2009
F. 1.3.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_adre_adenoma.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_adre_adenoma.plt
Mon Nov 23 15:42:49 2009
Source - Table 5
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 5
Total number of records with missing values = 1
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.00493749
Beta(l) = 4.83499e-GG6
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 5.60622e-006 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -24.6514 4
Fitted model -25.2438 1 1.18487 3 0.7566
Reduced model -31.4904 1 13.6781 3 0.00337E
AIC: 52.4876
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
85
0. 000
860.4590
0.0048
0.241
0. 000
50
-0.492
3944.9299
0.0219
1.094
2 . 000
50
0. 876
21334.0000
0.1127
5. 636
5. 000
50
-0.285
Chi'" 2 = 1.09 d.f. = 3 P-value = 0.7793
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 17 92.71
BMDL = 1020.18
BMDU = 3628.63
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-8 draft—do not cite or quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
Taken together, (10U0.18, J6^8.6j) is a 90
interval for the EMD
% two'-sided oO'Tifidence
Multistage Cancer Slope Factor = 9. 80ui:e-006
F.1.4. Kociba et al. (1978): Female, Hepatocellular Adenoma(s) or Carcinoma(s)
F.l.4.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
2.81
0.24
143.26
3.9E+02
2.8E+02
betas restricted >0
Multistage
cancer,
2-degree
2
2.81
0.24
143.26
3.9E+02
2.8E+02
betas restricted >0
Multistage
cancer,
3-degree
2
2.81
0.24
143.26
3.9E+02
2.8E+02
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
F.l.4.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
0.4
0.3
0.2
Multistage Cancer
Linear extrapolation
HMDL BMD
0 5000 10000 15000 20000
This document is a draft for review purposes ofofa and does not constitute Agency policy.
1^-4*5 11/0-5 onnq
1/15/10 F-9 draft—do not cite or quote
Multistage Cancer Model with 0.95 Confidence Level
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
F.l.4.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov2 3\Blood\mscl_ngkgd_liver_ad_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_liver_ad_carc.plt
Mon Nov 23 15:43:10 2009
Source - Table 1 in Goodman and Sauer 1992
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal^dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0400267
Beta(1) = 2.26421e-005
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.51
Beta(1) -0.51 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.022147 * * *
Beta (1) 2.60216e-005 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -68.2561 4
Fitted model -69.6304 2 2.74863 2 0.253
Reduced model -89.1983 1 41.8843 3 <.0001
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
AIC:
143.261
Goodness of Fit
Dose
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
0.0221
1. 905
2 . 000
CO
0. 070
852.5169
0.0436
2 .180
1. 000
50
-0.817
3941.9464
0.1175
5. 874
9. 000
50
1. 373
21246.0000
0.4374
19.685
18 . 000
45
-0.506
Chi/S2 = 2.81
d. f. =
2 P-
-value = 0.2449
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
386.23
276.228
577 .635
Taken together, (276.228, 577.635) is a 90
interval for the BMD
two-sided confidence
Multistage Cancer Slope Factor
3.6202e-005
F.1.5. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma of Hard Palate
or Nasal Turbinates
F.l.5.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.94
0.82
31.56
3.2E+03
1.5E+03
betas restricted >0
Multistage
cancer,
2-degree
3
0.15
0.99
30.17
7.5E+03
1.9E+03
betas restricted >0
Multistage
cancer,
3-degree
3
0.03
1.00
29.93
1.1E+04
2.0E+03
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.l.5.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.2
0.15
0.05
BMDL
BMD
0
5000
10000
15000
20000
dose
15:43 11/23 2009
F.l.5.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_nasal.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_nasal.plt
Mon Nov 23 15:43:31 2009
Source - Table 5
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial
Background =
Beta (1) =
Parameter Values
7.1158 9e-GG5
4.0351e-006
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 4.04 63e-006 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -18.7562 4
Fitted model -18.9547 1 0.397016 3 0.9409
Reduced model -24.1972 1 10.882 3 0.01238
AIC: 39.9093
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
CO
0. 000
852.5169
0.0034
0.172
0. 000
50
-0.416
3941.9464
0.0158
0.791
1. 000
50
0.237
21246.0000
0.0824
4 . 036
4 . 000
49
-0.019
Chi/S2 = 0.23 d.f. = 3 P-value = 0.9728
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2 4 83. 8 4
BMDL = 12 8 9.34
BMDU = 5762.51
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-13 draft—do not cite or quote
-------
1
2 Taken together, (1289.34, 5762.51) is a 90 % two-sided confidence
3 interval for the BMD
4
5 Multistage Cancer Slope Factor = 7.7559e-006
6
7
8 F.1.6. Kociba et al. (1978): Female, Keratinizing Squamous Cell Carcinoma of Lung
9 F.l.6.1. Summary Table of BMDS Modeling Results
10
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
1.75
0.63
45.30
1.7E+03
9.8E+02
betas restricted >0
Multistage
cancer,
2-degree
3
0.28
0.96
42.74
5.5E+03
1.5E+03
betas restricted >0
Multistage
cancer,
3-degree
3
0.05
1.00
42.29
8.6E+03
1.7E+03
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
11
12
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-14 draft—do not cite or quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.l.6.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
0.25
0.2
0.15
0.1
0.05
0 5000 10000 15000 20000
dose
15:43 11/23 2009
F.l.6.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_kera_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_kera_carc.plt
Mon Nov 23 15:43:52 2009
Source - Table 5
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
BMDL BMD
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 7.61927e-GG6
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(l) 5.80969e-006 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -20.0957 4
Fitted model -21.6489 1 3.10635 3 0.3755
Reduced model -31.4904 1 22.7894 3 <.0001
AIC: 45.2978
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0.000 0.000
CO
0. 000
852.5169
0.0049
0.247 0.000
50
-0.498
3941.9464
0.0226
1.132 0.000
50
-1.076
21246.0000
0.1161
5.690 7.000
49
0.584
Chi'" 2 = 1.75
d.f.
= 3
P-value = 0.62 63
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1729.93
BMDL = 984.302
BMDU = 3461.69
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-16 draft—do not cite or quote
-------
1 Taken together, (984.302, 3461.69) is a 90 % two-sided confidence
2 interval for the BMD
3
4 Multistage Cancer Slope Factor = 1.01595e-005
5
6
7 F.1.7. National Toxicology Program (1982): Female Rat, Subcutaneous Tissue.
8 Fibrosarcoma
9 F.l.7.1. Summary Table of BMDS Modeling Results
10
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
Multistage
cancer,
2-degree
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
Multistage
cancer,
3-degree
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
11
12
13 F.l.7.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
14
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.2
0.15
0.05
-A-
BMDI
BMD
This document is a drVift foPW vie \40
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
F.l.7.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov2 3\Blood\mscl_ngkgd_sub_fibro.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_sub_fibro.plt
Mon Nov 23 15:44:12 2009
Source - Table 10
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal^dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.026791
Beta(1) = 3.88561e-006
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.63
Beta(1) -0.63 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.0149169 * * *
Beta(1) 5.91146e-006 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -33.5998 4
Fitted model -35.6885 2 4.17734 2 0.1239
Reduced model -37.7465 1 8.29346 3 0.04032
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-18 draft—do not cite or quote
-------
1
2
3
4
AIC:
75.3769
5
Goodness of Fit
6
Scaled
7
8
Dose
Est. Prob
Expected Observed
Size
Residual
9
0.0000
0. 0149
1.119 0.000
75
-1.066
10
1072.2652
0.0211
1.057 2.000
50
0. 927
11
3111.2349
0.0329
1.643 3.000
50
1. 076
12
13
16207.0000
0.1049
5.141 4.000
49
-0.532
14
15
16
Chi'"'2 = 3
.44 d.f.
= 2 P-value = 0.1795
17
18
Benchmark Dose Computation
19
20
Specified 1
effect =
0. 01
21
22
Risk Type
=
Extra risk
23
24
Confidence
level =
0. 95
25
26
BMD =
1700.14
27
28
BMDL =
751.001
29
30
BMDU = 1.
77581e+009
31
Taken together, (751.001, 1.77581e+009) is a 90
% two-sided
conf iden<
32
interval for the BMD
33
34 Multistage Cancer Slope Factor = 1.33156e-005
35
36
37 F.1.8. National Toxicology Program (1982): Female Rat, Liver, Neoplastic Nodule or
38 Hepatocellular Carcinoma
39 F.l.8.1. Summary Table of BMDS Modeling Results
40
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.80
0.22
135.20
6.4E+02
4.0E+02
betas restricted
>0
Multistage
cancer,
2-degree
3
0.13
0.49
133.45
3.0E+03
4.8E+02
betas restricted >0
Multistage
cancer,
3-degree
3
0.02
0.24
135.44
3.9E+03
4.8E+02
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
41
42
43
44
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.l.8.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.15
0.05
0
BMDL
BMD
0 2000 4000 6000 8000 10000 12000 14000 16000
dose
15:45 11/23 2009
F.l.8.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_liver_nod.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_liver_nod.plt
Mon Nov 23 15:45:38 2009
Source - Table 9
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-20 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 4.03747e-006
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta (1) 3.004 92e-006 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -11.3484 4
Fitted model -12.0545 1 1.41226 3 0.7027
Reduced model -15.9189 1 9.14109 3 0.02747
AIC: 26.109
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
74
0. 000
1071.8576
0.0032
0.161
0. 000
50
-0.402
3115.7313
0.0093
0. 466
0. 000
50
CO
o
16272.0000
0.0477
2 . 386
3. 000
50
0. 407
Chi'" 2 = 0.80 d.f. = 3 P-value = 0.8501
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 3344.63
BMDL = 1472.42
BMDU = 10322 . 4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-21 draft—do not cite or quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Taken together, (147U.4U, 10JUU.4) is a 90
interval for the EMD
% two'-sided confidence
Multistage Cancer Slope Factor = 6.7 9156e-006
F.1.9. National Toxicology Program (1982): Female Rat, Adrenal, Cortical Adenoma, or
Carcinoma or Adenoma, NOS
F.l.9.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
2.18
0.34
203.83
8.8E+02
4.4E+02
betas restricted
>0
Multistage
cancer,
2-degree
2
1.51
0.47
203.03
3.6E+03
4.9E+02
betas restricted >0
Multistage
cancer,
3-degree
2
1.37
0.51
202.87
5.9E+03
5.0E+02
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
F.l.9.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.5
0.4
0.3
0.2
0.1
Multistage Cancer
Linear extrapolation
BMDL BMD
1/15/10
rr, ¦ , , • 0 r. j2000 .4000 6000 8000 10000 12000 14000 , 16000
1 his document is a araft jor review purposes owy anaaoes not constitute Agency policy.
15:44 11/23 2009
F-22
DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
F.l.9.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov2 3\Blood\mscl_ngkgd_adre_cort_ad_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_adre_cort_ad_carc.plt
Mon Nov 23 15:44:55 2009
Source - Table 10
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal^dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.134119
Beta(1) = 1.27888e-005
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.54
Beta(1) -0.54 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.139831 * * *
Beta(1) 1.14475e-005 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -98.7282 4
Fitted model -99.9133 2 2.37035 2 0.3057
Reduced model -102.201 1 6.94636 3 0.07363
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
AIC:
203.827
Goodness of Fit
Dose
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
0.1398
10.208
11.000
73
0.267
1072.2652
0.1503
7 . 366
9. 000
49
0. 653
3111.2349
0.1699
8 . 326
5. 000
49
-1.265
16207.0000
0.2855
13.132
14.000
46
0.283
Chi/S2 = 2.18
d. f. =
2 P-
-value = 0.3363
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
877.947
443.554
.93624e+008
Taken together, (443.554, 6.93624e+008) is a 90
interval for the BMD
Multistage Cancer Slope Factor = 2.25452e-005
two-sided confidence
F.1.10. National Toxicology Program (1982): Female Rat, Thyroid, Follicular-Cell
Adenoma
F. 1.10.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
1.13
0.57
92.41
1.8E+03
8.5E+02
betas restricted >0
Multistage
cancer,
2-degree
2
0.62
0.74
91.75
5.2E+03
9.2E+02
betas restricted >0
Multistage
cancer,
3-degree
2
0.52
0.77
91.63
7.5E+03
9.4E+02
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F. 1.10.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.25
0.2
0.15
0.1
0.05
Multistage Cancer
Linear extrapolation
BMDL BMD
0 2000 4000 6000 8000 10000 12000 14000 16000
dose
15:45 11/23 2009
F. 1.10.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_thy_ad.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_thy_ad.plt
Mon Nov 23 15:45:17 2009
Source - Table 10
The form of the probability function is:
background + (1-background)*[1-EXP(
-betal^dose^l)]
P[response]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0282954
Beta(l) = 6.36609e-006
Background
Beta (1)
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
1 -0.54
-0.54 1
Variable
Background
Beta (1)
Parameter Estimates
Estimate
0. 0332349
5. 46313e-006
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Deviance
Log(likelihood)
-43.5264
-44.2066
-46.2299
92.4132
Param's
4
2
1
Test d.f.
1. 36031
5 . 4 0699
P-value
0.5065
0.1443
Goodness of Fit
Scaled
Dose
Est. Prob.
Expected
Observed
Size
Residual
0.0000
0.0332
2 .426
3. 000
73
0.375
1072.2652
0.0389
1.750
2 . 000
45
0.193
3111.2349
0.0495
2 .427
1. 000
49
-0.939
16207.0000
0.1152
5. 412
6. 000
47
0.269
Chi/S2
1.13
d. f.
P-value
0.5677
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
1839.67
846.279
11586.6
Taken together, (846.279, 11586.6) is a 90 % two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-26 DRAFT—DO NOT CITE OR QUOTE
-------
1 interval for the BMD
2
3 Multistage Cancer Slope Factor = 1.18164e-005
4
5
6 F.l.ll. National Toxicology Program (1982): Male Rat, Liver, Neoplastic Nodule or
7 Hepatocellular Carcinoma
8 F.l.11.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.80
0.22
135.20
6.4E+02
4.0E+02
betas restricted >0
Multistage
cancer,
2-degree
3
0.13
0.49
133.45
3.0E+03
4.8E+02
betas restricted >0
0
11 Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F. 1.11.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
¦
Multistage Cancer
Linear extrapolation
0.15
0.1
0.05
BMDL BMD
0 2000 4000 6000 8000 10000 12000 14000 16000
dose
15:45 11/23 2009
F. 1.11.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_liver_nod.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_liver_nod.plt
Mon Nov 23 15:45:38 2009
Source - Table 9
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 4.03747e-006
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta (1) 3.004 92e-006 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -11.3484 4
Fitted model -12.0545 1 1.41226 3 0.7027
Reduced model -15.9189 1 9.14109 3 0.02747
AIC: 26.109
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
74
0. 000
1071.8576
0.0032
0.161
0. 000
50
-0.402
3115.7313
0.0093
0. 466
0. 000
50
CO
o
16272.0000
0.0477
2 . 386
3. 000
50
0. 407
Chi'" 2 = 0.80 d.f. = 3 P-value = 0.8501
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 3344.63
BMDL = 1472.42
BMDU = 10322 . 4
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Taken together, (1472.42, 10322.4) is a 90 % two-sided confidence
3 interval for the BMD
4
5 Multistage Cancer Slope Factor = 6.79156e-006
6
7
8 F.1.12. National Toxicology Program (1982): Male Ra, Thyroid, Follicular-Cell Adenoma
9 or Carcinoma
10 F.l.12.1. Summary Table of BMDS Modeling Results
11
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
5.73
0.06
149.25
6.6E+02
3.8E+02
betas restricted >0
Multistage
cancer,
2-degree
2
5.73
0.06
149.25
6.6E+02
3.8E+02
betas restricted >0
Multistage
cancer,
3-degree
2
5.73
0.06
149.25
6.6E+02
3.8E+02
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
12
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.l.12.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
Multistage Cancer
Linear extrapolation
BMDL BMD
0 2000 4000 6000 8000 10000 12000 14000 16000
dose
15:45 11/23 2009
F.l.12.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_thyroid.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_thyroid.plt
Mon Nov 23 15:45:59 2009
Source - Table 9
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0767848
Beta(1) = 1.11362e-005
Background
Beta (1)
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
1 -0.62
-0.62 1
Variable
Background
Beta (1)
Parameter Estimates
Estimate
0. 0527729
1.52 871e-005
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Log(likelihood)
-69.5946
Param'
4
Deviance Test d.f.
P-value
Fitted
model -7 2.
6245
2
6. 05993
2
0
. 04832
Reduced
model -7 7 .
5267
1
15.8643
3
0.
001209
AIC: 149
.249
Goodness of
Fit
Scaled
Dose
Est. Prob.
Expected
Observed
Size
Residual
0.0000
0.0528
3. 641
1. 000
69
-1.422
1071.8576
0.0682
3.272
5. 000
48
0. 990
3115.7313
0.0968
4 .842
8 . 000
50
1. 510
16272.0000
0.2614
13.069
11.000
50
-0.6 6 6
Chi/S2
5.73
d.f.
P-value
0.0571
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
657.439
380.166
1571.51
Taken together, (380.166, 1571.51) is a 90 % two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-32 DRAFT—DO NOT CITE OR QUOTE
-------
1 interval for the BMD
2
3 Multistage Cancer Slope Factor = 2.63043e-005
4
5
6 F.1.13. National Toxicology Program (1982): Male Rat, Adrenal Cortex, Adenoma
7 F.l.13.1. Summary Table of BMDS Modeling Results
8
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
5.55
0.06
199.31
2.2E+03
6.7E+02
betas restricted >0
Multistage
cancer,
2-degree
2
5.55
0.06
199.31
2.2E+03
6.7E+02
betas restricted >0
Multistage
cancer,
3-degree
2
5.55
0.06
199.31
2.2E+03
6.7E+02
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-33 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.l.13.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.4
0.35
0.3
0.25
0.2
0.15
0.05
BMDI
IMD
0
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
15:46 11/23 2009
F.l.13.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_adre_cort.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_adre_cort.plt
Mon Nov 23 15:46:20 2009
Source - Table 9
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-34 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.16365
Beta(1) = 2.66257e-GG6
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.61
Beta(1) -0.61 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.146024 * * *
Beta(l) 4.6499e-006 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -94.8672 4
Fitted model -97.6531 2 5.57181 2 0.06167
Reduced model -98.0432 1 6.35197 3 0.09569
AIC: 199.306
Goodness of Fit
Dose
Est. Prob.
Expected
Observed
Size
Scalei
Residu;
0.0000
0.1460
10.514
6. 000
72
-1. 506
1071.8576
0.1503
7 . 513
9. 000
50
0.588
3115.7313
0.1583
7 .757
12.000
49
1. 661
16272.0000
0.2083
10.204
9. 000
49
-0.424
Chi'A2 = 5.55 d.f. = 2 P-value = 0.0623
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2161. 41
BMDL = 665.411
BMDU did not converge for BMR = 0.010000
BMDU calculation failed
BMDU = Inf
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-35 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3 F.1.14. National Toxicology Program (1982): Female Mice, Subcutaneous Tissue,
4 Fibrosarcoma
5 F. 1.14.1. Summary Table of BMDS Modeling Results
6
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistag
e cancer,
l-degreeb
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
Multistage
cancer,
2-degree
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
Multistage
cancer,
3-degree
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
7
8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-36 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F. 1.14.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.25
0.2
0.15
0.1
0.05
0 2000 4000 6000 8000 10000 12000 14000 16000 18000
dose
15:46 11/23 2009
F. 1.14.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_subcu_fibro.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_subcu_fibro.plt
Mon Nov 23 15:46:40 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Multistage Cancer
Linear extrapolation
BMDL
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-37 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0104441
Beta(1) = 5.78849e-006
Background
Beta (1)
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
1 -0.55
-0.55 1
Variable
Background
Beta (1)
Parameter Estimates
Estimate
0. 0124215
5. 43453e-006
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Deviance Test d.f.
Log(likelihood) # Param'
-30.9876 4
-31.0699 2
-34.3291 1
66.1398
0.16463
6. 68308
P-value
0. 921
0.08272
Goodness of Fit
Scaled
Dose
Est. Prob.
Expected
Observed
Size
Residual
0.0000
0.0124
0. 919
1. 000
74
0. 085
1063.6377
0.0181
0. 906
1. 000
50
0.100
3184.3353
0.0294
1.410
1. 000
48
-0.350
17406.0000
0.1016
4 .773
5. 000
47
0.110
Chi/S2
0.15
d.f.
P-value
0.9269
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
1849.35
916.028
6164.32
Taken together, (916.028, 6164.32) is a 90 % two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-38 DRAFT—DO NOT CITE OR QUOTE
-------
1 interval for the BMD
2
3 Multistage Cancer Slope Factor = 1.09167e-005
4
5
6 F.1.15. National Toxicology Program (1982): Female Mice, Hematopoietic System,
7 Lymphoma
8 F. 1.15.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistag
e cancer,
l-degreeb
2
0.05
0.98
261.45
6.2E+02
3.3E+02
betas restricted >0
Multistage
cancer,
2-degree
1
0.03
0.87
263.43
9.3E+02
3.3E+02
betas restricted >0
Multistage
cancer,
3-degree
1
0.03
0.87
263.43
9.3E+02
3.3E+02
betas restricted >0
a Values <0.1 fail to meet BMDS goodness-of-fit criteria.
b Best-fitting model as assessed by lowest-AIC criterion, bolded.
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F. 1.15.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.6
0.5
0.4
0.3
0.2
0.1
Multistage Cancer
Linear extrapolation
3MDL BMD
2000
4000
6000
8000 10000
dose
12000 14000 16000 18000
15:47 11/23 2009
F. 1.15.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_mice_f_lymphoma.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_mice_f_lymphoma.plt
Mon Nov 23 15:47:00 2009
Table 15 page 64 Hematopoietic System Lymphoma or Leukemia
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-40 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.234156
Beta(l) = 1.645e-005
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.54
-0.54
1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0.236108
1. 61681e-005
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Log(likelihood) # Param'
-128.699 4
-128.723 2
-131.412 1
261.446
Deviance Test d.f.
0.0471776
5. 42487
P-value
0. 9767
0.1432
Dose
Goodness of Fit
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
1063.6377
3184.3353
17406.0000
Chi/S2
0.2361
0.2491
0.2744
0. 4235
0. 05
d.f.
17 .472
12.457
13.173
19.904
18 . 000
12.000
13.000
20.000
P-value
0.9767
74
50
48
47
0.145
-0.149
-0.056
0. 028
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
621.617
330.742
2332 . 7
Taken together, (330.742, 2332.7 ) is a 90 % two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-41 DRAFT—DO NOT CITE OR QUOTE
-------
1 interval for the BMD
2
3 Multistage Cancer Slope Factor = 3.0235e-005
4
5
6 F.1.16. National Toxicology Program (1982): Female Mice, Liver, Hepatocellular
7 Adenoma or Carcinoma
8 F. 1.16.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
2.15
0.34
155.21
8.1E+02
4.5E+02
betas restricted >0
Multistage
cancer,
2-degree
2
2.15
0.34
155.21
8.1E+02
4.5E+02
betas restricted >0
Multistage
cancer,
3-degree
2
2.15
0.34
155.21
8.1E+02
4.5E+02
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-42 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F. 1.16.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.4
¦
¦1 ¦
Multistage Cancer
Linear extrapolation
i i | i
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
BMDL
3MD
0 2000 4000 6000 8000 10000 12000 14000 16000 18000
dose
15:47 11/23 2009
F. 1.16.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_mice_f_liv_aden_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_mice_f_liv_aden_carc.plt
Mon Nov 23 15:47:20 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-43 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0808715
Beta(1) = 1.07 435e-GG5
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.57
Beta(1) -0.57 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.0691337 * * *
Beta (1) 1.24516e-005 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -74.5177 4
Fitted model -75.603 2 2.17074 2 0.3378
Reduced model -79.6703 1 10.3053 3 0.01614
AIC: 155.206
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0691 5.047 3.000 73 -0.944
1063.6377 0.0814 4.069 6.000 50 0.999
3184.3353 0.1053 5.055 6.000 48 0.444
17406.0000 0.2505 11.774 11.000 47 -0.261
Chi ^2 = 2.15 d.f. = 2 P-value = 0.3405
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 807.155
BMDL = 4 4 8.599
BMDU = 2161.58
Taken together, (448.599, 2161.58) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-44 DRAFT—DO NOT CITE OR QUOTE
-------
1 Multistage Cancer Slope Factor = 2.22916e-005
2
3
4 F.1.17. National Toxicology Program (1982): Female Mice, Thyroid, Follicular-Cell
5 Adenoma
6 F. 1.17.1. Summary Table of BMDS Modeling Results
l
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
Multistage
cancer,
2-degree
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
Multistage
cancer,
3-degree
2
3.44
0.18
75.38
1.7E+03
7.5E+02
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
8
9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-45 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F. 1.17.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.25
0.2
0.15
0.1
0.05
¦ ¦
Multistage Cancer
Linear extrapolation
BMDL BMD
0 2000 4000 6000 8000 10000 12000 14000 16000 18000
dose
15:47 11/23 2009
F. 1.17.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_mice_f_thyroid_aden.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_mice_f_thyroid_aden.plt
Mon Nov 23 15:47:40 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-46 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0202008
Beta(1) = 5.39488e-006
Background
Beta (1)
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
1 -0.58
-0.58 1
Variable
Background
Beta (1)
Parameter Estimates
Estimate
0. 0152512
). 07986e-006
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Analysis of Deviance Table
Deviance Test d.f.
Log(likelihood) # Param'
-32.0017 4
-34.3878 2
4 .77223
P-value
0.09199
Reduced
model
-37
.2405 1
10.
4776
o
CO
AIC:
72
.7756
Goodness
of Fit
Scaled
Dose
Est. Prob
Expected Observed
Size
Residual
0.0000
0.0153
1.052 0.
000
69
-1.034
1063.6377
0.0216
1.080 3.
000
50
1.868
3184.3353
0.0341
1.604 1.
000
47
-0.485
17406.0000
0.1141
5.250 5.
000
46
-0.116
Chi/S2 = 4
. 81
d.f.
=
2 P-value
= 0.0904
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
1653.05
778.784
6460.82
Taken together, (778.784, 6460.82) is a 90 % two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-47 DRAFT—DO NOT CITE OR QUOTE
-------
1 interval for the BMD
2
3 Multistage Cancer Slope Factor = 1.28405e-005
4
5
6 F.1.18. National Toxicology Program (1982): Male Mice, Lung, Alveolar/Bronchiolar
7 Adenoma or Carcinoma
8 F. 1.18.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
1-degree
2
4.87
0.09
168.35
3.5E+02
1.9E+02
betas restricted >0
Multistag
e cancer,
2-degreeb
2
3.58
0.17
166.95
1.4E+03
2.3E+02
betas restricted >0
Multistage
cancer,
3-degree
2
3.41
0.18
166.80
2.3E+03
2.3E+02
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F. 1.18.2. Figure for Selected Model: Multistage Cancer, 2-Degree, Betas Restricted >0
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
BMDL
BMD
1000
2000
3000
dose
4000
5000
6000
15:48 11/23 2009
F. 1.18.3. Output File for Selected Model: Multistage Cancer, 2-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\msc2_ngkgd_lung_aden_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\msc2_ngkgd_lung_aden_carc.plt
Mon Nov 23 15:48:02 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2 ) ]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-49 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.086839
Beta(1) = 0
Beta(2) = 5.59843e-009
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Beta(l)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Background Beta(2)
Background 1 -0.46
Beta(2) -0.46 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Estimate
0.0942375
0
5.31152e-009
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood) # Param'
-79.5959 4
-81.4737 2
-85.3351 1
Deviance Test d.f.
3.75561
11.4782
P-value
0.1529
0. 009402
166.947
Dose
Est. Prob.
Goodness of Fit
Expected Observed
Scaled
Residual
0.0000
420.0366
1239.6134
6117.5662
0.0942
0.0951
0.1016
0.2575
6. 691
4 . 564
4 . 877
12.876
10.000
2 . 000
4 . 000
13.000
71
48
48
50
1.344
-1.262
-0.419
0. 040
Chi ^2
3.58
d.f.
P-value
0.1673
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 137 5.56
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-50 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 BMDL = 225.385
3
4 BMDU = 2284.92
5
6 Taken together, (225.385, 2284.92) is a 90 % two-sided confidence
7 interval for the BMD
8
9 Multistage Cancer Slope Factor = 4.43686e-005
10
11
12 F.1.19. National Toxicology Program (1982): Male Mice, Liver, Hepatocellular Adenoma
13 or Carcinoma
14 F. 1.19.1. Summary Table of BMDS Modeling Results
15
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
0.15
0.93
258.55
1.1E+02
7.5E+01
betas restricted >0
Multistage
cancer,
2-degree
1
0.08
0.78
260.48
1.7E+02
7.5E+01
betas restricted >0
Multistage
cancer,
3-degree
1
0.07
0.79
260.47
1.6E+02
7.5E+01
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
16
17
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-51 draft—do not cite or quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F. 1.19.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.7
0.6
0.5
0.4
0.3
ISMDL BMD
1000
2000
3000
dose
4000
5000
6000
15:48 11/23 2009
Multistage Cancer
Linear extrapolation
F. 1.19.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_mice_m_liver_aden_carc.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_mice_m_liver_aden_carc.pit
Mon Nov 23 15:48:23 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-52 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.201516
Beta(1) = 8.94392e-005
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.53
-0.53
1
Parameter Estimates
Variable
Background
Beta(1)
Estimate
0.204153
]. 75513e-005
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood) # Param'
-127.199 4
-127.274 2
-135.589 1
Deviance Test d.f.
0.151061
16.7801
P-value
0. 9273
0.0007843
AIC:
258.549
Dose
Est. Prob.
Goodness of Fit
Expected
Observed
Scaled
Residual
0.0000
420.0366
1239.6134
6117.5662
2042
2329
2860
5342
14.903
11.412
14.014
26.709
15.
12 .
13.
27 .
000
000
000
000
73
49
49
50
0. 028
0.199
-0.321
0. 083
Chi'" 2
0.15
d.f.
P-value
0. 9278
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
114.794
74 . 9717
208.915
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-53 DRAFT—DO NOT CITE OR QUOTE
-------
1 Taken together, (74.9717, 208.915) is a 90 % two-sided confidence
2 interval for the BMD
3
4 Multistage Cancer Slope Factor = 0.000133384
5
6
7 F.1.20. National Toxicology Program (2006): Liver, Cholangiocarcinoma
8 F. 1.20.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
1-degree
5
20.87
0.00
138.46
5.2E+02
3.9E+02
betas restricted >0
Multistage
cancer,
2-degree
5
5.09
0.40
119.37
2.3E+03
1.6E+03
betas restricted >0
Multistage
cancer,
3-degreeb
5
0.47
0.99
113.51
4.2E+03
2.3E+03
betas restricted >0
" Values <0.1 fail to meet BMDS goodness-of-fit criteria
b Best-fitting model as assessed by lowest-AIC criterion, bolded
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-54 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F. 1.20.2. Figure for Selected Model: Multistage Cancer, 3-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.6
0.5
0.4
0.3
0.2
0.1
0 2000 4000 6000 8000 10000 12000 14000 16000
dose
15:48 11/23 2009
F. 1.20.3. Output File for Selected Model: Multistage Cancer, 3-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\msc3_ngkgd_liv_cho-carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\msc3_ngkgd_liv_cho-carc.pit
Mon Nov 23 15:48:43 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2-beta3*dose/N3) ]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 4
Total number of specified parameters = 0
Degree of polynomial = 3
BMDL
Multistage Cancer
Linear extrapolation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-55 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0
Beta(2) = 0
Beta(3) = 1.46324e-013
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background -Beta(l) -Beta(2)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(3)
Beta(3) 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Beta(3)
Estimate
0
0
0
1. 382 96e-013
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-55.408
-55.7539
-96.9934
Param's Deviance Test d.f.
0.691685
83.1708
P-value
0.9834
C.0001
113.508
Dose
Est. Prob.
Goodness of Fit
Expected
Observed
Size
Scaled
Residual
0.0000
0
0000
0
000
0
000
49
0
000
1408.4504
0
0004
0
019
0
000
48
-0
136
3137.0446
0
0043
0
196
0
000
46
-0
444
5392.9593
0
0215
1
073
1
000
50
-0
071
9128.8027
0
0 9 9 9
4
893
4
000
49
-0
426
16361.0000
0
4543
24
078
25
000
53
0
254
Chi/S2
0.47
d.f.
P-value
0.9933
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-56 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMD = 417 3.08
2
3 BMDL = 2211.lb
4
5 BMDU = 4 642.18
6
7 Taken together, (2277.15, 4642.18) is a 90 % two-sided confidence
8 interval for the BMD
9
10 Multistage Cancer Slope Factor = 4.39145e-006
11
12
13 F.1.21. National Toxicology Program (2006): Liver, Hepatocellular Adenoma
14 F. 1.21.1. Summary Table of BMDS Modeling Results
15
Model
Degrees of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model
Notes
Multistage
cancer,
1-degree
5
12.73
0.03
87.02
1.2E+03
8.0E+02
betas
restricted
>0
Multistage
cancer,
2-degree
5
4.29
0.51
76.98
3.6E+03
2.4E+03
betas
restricted
>0
Multistag
e cancer,
3-degreeb
5
1.32
0.93
72.78
5.6E+03
3.6E+03
betas
restricted
>0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
16
17
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-57 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F. 1.21.2. Figure for Selected Model: Multistage Cancer, 3-Degree, Betas Restricted >0
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
BMDL
BMD
2000
4000
6000
8000
dose
10000 12000 14000 16000
15:49 11/23 2009
F. 1.21.3. Output File for Selected Model: Multistage Cancer, 3-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\msc3_ngkgd_liv_hepat_ad.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\msc3_ngkgd_liv_hepat_ad.plt
Mon Nov 23 15:49:03 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2-beta3*dose/N3) ]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 4
Total number of specified parameters = 0
Degree of polynomial = 3
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-58 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0
Beta(2) = 0
Beta(3) = 6.51095e-014
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background -Beta(l) -Beta(2)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(3)
Beta(3) 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Beta(3)
Estimate
0
0
0
5.627 66e-014
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-34.4075
-35.3908
-56.3333
Param's Deviance Test d.f.
1.96651
43.8515
P-value
0.8538
C.0001
72.7815
Dose
Est. Prob.
Goodness of Fit
Expected
Observed
Size
Scaled
Residual
0.0000
0
0000
0
000
0
000
49
0
000
1408.4504
0
0002
0
008
0
000
48
-0
087
3137.0446
0
0017
0
080
0
000
46
-0
283
5392.9593
0
0088
0
439
0
000
50
-0
6 6 6
9128.8027
0
0419
2
054
1
000
49
-0
751
16361.0000
0
2184
11
577
13
000
53
0
473
Chi/S2
1. 32
d.f.
P-value
0.9330
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-59 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMD = 5631.42
2
3 BMDL = 3596.03
4
5 BMDU = 6597.97
6
7 Taken together, (3596.03, 6597.97) is a 90 % two-sided confidence
8 interval for the BMD
9
10 Multistage Cancer Slope Factor = 2.78084e-006
11
12
13 F.1.22. National Toxicology Program (2006): Oral mucosa, Squamous Cell Carcinoma
14 F. 1.22.1. Summary Table of BMDS Modeling Results
15
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
Multistage
cancer, 1-degree
4
4.15
0.39
125.48
4.8E+00
3.0E+00
betas restricted >0
Multistage cancer,
2-degreeb
4
2.83
0.59
123.25
1.6E+01
3.8E+00
betas restricted >0,
bound hit
Multistage cancer,
3-degree
4
2.83
0.59
123.25
1.6E+01
3.8E+00
betas restricted >0,
bound hit
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
16
17
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-60 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F. 1.22.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0,
Bound Hit
Multistage Cancer Model with 0.95 Confidence Level
0.35
Multistage Cancer
Linear extrapolation
0.3
0.25
0.2
0.15
0.05
BMDL BMD
0
10
20
30
40
50
60
70
dose
15:11 11/23 2009
F. 1.22.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0,
Bound Hit
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_oral_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_oral_carc.pit
Mon Nov 23 15:11:37 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-61 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.00607545
Beta(1) = 0.00265195
Background
Beta(1)
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
1 -0.6
-0.6 1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0. 0171416
0. 00211536
Std. Err.
- Indicates that this value is not calculated.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Param's Deviance Test d.f.
Log(likelihood)
-57.5353
-60.7418
-67.7782
125.484
6.41293
20.4858
P-value
0.1704
0. 001013
Dose
Goodness of Fit
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
0.0171
0
840
1
000
49
0
176
2.1400
0.0216
1
036
2
000
48
0
958
7.1400
0.0319
1
466
1
000
46
-0
391
15.7000
0.0492
2
462
0
000
50
-1
609
32.9000
0.0832
4
078
4
000
49
-0
040
71.4000
0.1549
8
211
10
000
53
0
679
Chi ^2
4 .15
d.f.
P-value
0.3855
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 4 .75111
BMDL = 2 . 9556
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-62 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 BMDU = 9.194 54
3
4 Taken together, (2.9556 , 9.19454) is a 90 % two-sided confidence
5 interval for the BMD
6
7 Multistage Cancer Slope Factor = 0.0033834
8
9
10 F.1.23. National Toxicology Program (2006): Pancreas, Adenoma or Carcinoma
11 F. 1.23.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
5
3.39
0.64
29.37
5.8E+03
2.6E+03
betas restricted >0
Multistage
cancer,
2-degree
5
1.36
0.93
27.06
8.0E+03
4.0E+03
betas restricted >0
Multistage
cancer,
3-degree
5
0.64
0.99
25.97
9.6E+03
5.2E+03
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
13
14
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-63 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F. 1.23.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.15
0.1
0.05
'I I I I 1
Multistage Cancer
Linear extrapolation
BMD
¦ i ¦ i ¦ ¦ ¦ ¦ i ¦ ¦ i ¦
2000 4000 6000 8000 10000 12000 14000 16000
dose
15:49 11/23 2009
F. 1.23.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_panc_ad_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_panc_ad_carc.plt
Mon Nov 23 15:49:45 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-64 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(l) = 3.46905e-006
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
Variable
Background
Beta(1)
Estimate
0
1.73461e-006
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-11.4096
-13.6863
-16.7086
Param's Deviance Test d.f.
P-value
4 . 55338
10.598
0.4728
0.05996
AIC:
29.3726
Est. Prob.
Goodness of Fit
Expected Observed
Scaled
Residual
0.0000
0
0000
0
000
0
000
48
0
000
1408.4504
0
0024
0
117
0
000
48
-0
343
3137.0446
0
0054
0
250
0
000
46
-0
501
5392.9593
0
0093
0
466
0
000
50
-0
686
9128.8027
0
0157
0
754
0
000
48
-0
875
16361.0000
0
0280
1
427
3
000
51
1
336
Chi/N2 = 3.39 d.f. = 5 P-value = 0.6404
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 57 94
BMDL = 2550.9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-65 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMDU = 18101.1
2
3 Taken together, (2550.9 , 18101.1) is a 90 % two-sided confidence
4 interval for the BMD
5
6 Multistage Cancer Slope Factor = 3.92019e-006
7
8
9 F.1.24. National Toxicology Program (2006): Lung, Cystic Keratinizing Epithelioma
10 F.l.24.1. Summary Table of BMDS Modeling Results
11
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
1-degree
5
10.52
0.06
64.03
1.9E+03
1.1E+03
betas restricted >0
Multistage
cancer,
2-degreeb
5
4.30
0.51
56.94
4.6E+03
2.9E+03
betas restricted
>0
Multistage
cancer,
3-degree
5
2.03
0.84
53.56
6.6E+03
4.3E+03
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
12
13
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-66 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F. 1.24.2. Figure for Selected Model: Multistage Cancer, 2-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.3
0.25
0.2
0.15
0.1
0.05
Multistage Cancer
Linear extrapolation
BMDL
BMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
15:50 11/23 2009
F. 1.24.3. Output File for Selected Model: Multistage Cancer, 2-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\msc2_ngkgd_lung_epith.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\msc2_ngkgd_lung_epith.plt
Mon Nov 23 15:50:07 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2 ) ]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-67 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0
Beta(2) = 7.12912e-010
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background -Beta(l)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(2)
Beta(2) 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Estimate
0
0
4 . 80115e-010
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-23.958
-27 .4714
-40.2069
Param's Deviance Test d.f.
P-value
7.02665
32.4976
0.2187
<.0001
AIC:
56.9427
Dose
Est. Prob.
Goodness of Fit
Expected Observed Size
Scaled
Residual
0.0000
0
0000
0
000
0
000
49
0
000
1408.4504
0
0010
0
046
0
000
48
-0
214
3137.0446
0
0047
0
217
0
000
46
-0
467
5392.9593
0
0139
0
679
0
000
49
-0
830
9128.8027
0
0392
1
922
0
000
49
-1
414
16361.0000
0
1206
6
271
9
000
52
1
162
Chi/N2 = 4.30 d.f. = 5 P-value = 0.5067
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 4575.28
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-68 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMDL = 288 9.7 9
2
3 BMDU = 6187.18
4
5 Taken together, (2889.79, 6187.18) is a 90 % two-sided confidence
6 interval for the BMD
7
8 Multistage Cancer Slope Factor = 3.46046e-006
9
10
11 F.1.25. Toth et al. (1978): 1YR, Liver, Tumors
12 F. 1.25.1. Summary Table of BMDS Modeling Results
13
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
1
1.10
0.29
155.74
2.0E+02
1.2E+02
betas restricted >0
Multistage
cancer,
2-degree
1
1.10
0.29
155.74
2.0E+02
1.2E+02
betas restricted >0
Multistage
cancer,
0-degree
0.29
-999.00
error
error
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
14
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-69 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F. 1.25.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.6
0.5
0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_adr_cor_lyr.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\mscl_ngkgd_adr_cor_lyr.plt
Mon Nov 23 17:29:16 2009
Table 1
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 3
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
:
Multistage Cancer
Linear extrapolation
¦—FT
MDL
BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-70 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.234944
Beta(l) = 4.90901e-005
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.55
Beta(1) -0.55 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.235288 * * *
Beta(l) 4.96192e-005 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -75.3127 3
Fitted model -75.8701 2 1.11477 1 0.291
Reduced model -79.4897 1 8.35401 2 0.01534
AIC: 155.74
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.2353 8.941 7.000 38 -0.742
315.4949 0.2472 10.875 13.000 44 0.743
7814.0188 0.4811 21.167 21.000 44 -0.050
Chi^2 = 1.10 d.f. = 1 P-value = 0.2932
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 202.54 9
BMDL = 115.257
BMDU = 555.60 9
Taken together, (115.257, 555.609) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-71 draft—do not cite or quote
-------
1
2 Multistage Cancer Slope Factor = 8.67623e-005
3
4
5 F.2. ADMINISTERED DOSE BMDS RESULTS
6 F.2.1. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma of Hard Palate
7 or Nasal Turbinates
8 F.2.1.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.46
0.93
30.75
1.3E+01
6.5E+00
betas restricted >0
Multistage
cancer,
2-degree
3
0.04
1.00
29.96
3.5E+01
7.2E+00
betas restricted >0
Multistage
cancer,
3-degree
3
0.00
1.00
29.89
4.9E+01
7.3E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-72 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.1.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.2
0.15
0.1
0.05
0
20 40 60 80 100
dose
15:04 11/23 2009
F.2.1.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_palate_nasal.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_palate_nasal.plt
Mon Nov 23 15:04:28 2009
Source - Table 4
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
BMDL
I-'I-1I"II"II"II"I—T—TI-1I"II"'I"II"II"II"II"II"I—T—T—I"'I"II"II"II"II"II"II"I—T—I"
Multistage Cancer -
Linear extrapolation -
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-73 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0.000858074
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta (1) 0.00074801 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -13.9385 4
Fitted model -14.3726 1 0.868297 3 0.8331
Reduced model -20.2589 1 12.6409 3 0.005481
AIC: 30.7452
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
85
0. 000
1.0000
0.0007
0. 037
0. 000
50
-0.193
10.0000
0.0075
0.373
0. 000
50
-0.613
100.0000
0.0721
3. 604
4 . 000
50
0.217
Chi'A2 = 0.46 d.f. = 3 P-value = 0.9276
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 13.4361
BMDL = 6.51522
BMDU = 34.82 9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-74 DRAFT—DO NOT CITE OR QUOTE
-------
1 Taken together, (6.51522, 34.829 ) is a 90 % two-sided confidence
2 interval for the BMD
3
4 Multistage Cancer Slope Factor = 0.00153487
5
6
7 F.2.2. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma of Tongue
8 F.2.2.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
1 degreeb
2
1.59
0.45
48.37
1.7E+01
7.1E+00
betas restricted >0
Multistage
cancer,
2-degree
2
1.59
0.45
48.37
1.7E+01
7.1E+00
betas restricted >0
Multistage
cancer,
3-degree
2
1.59
0.45
48.37
1.7E+01
7.1E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-75 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.2.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.15
0.1
0.05
Multistage Cancer
Linear extrapolation
BMDL
15:04 11/23 2009
F.2.2.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_tongue.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_tongue.plt
Mon Nov 23 15:04:49 2009
Source - Table 4
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-76 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0113883
Beta(1) = 0.000508703
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.52
Beta(1) -0.52 1
Parameter Estimates
Variable
Background
Beta (1)
Estimate
0.00809154
0.000576915
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
Analysis of Deviance Table
Log(likelihood)
-21.1523
-22 .1838
-24 .1972
Param's
4
2
1
Deviance Test d.f.
2.06309 2
6.08976 3
P-value
0.3565
0.1073
AIC:
48.3677
Dose
Goodness of Fit
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
1.0000
10.0000
100.0000
Chi'" 2
1.59
0.0081
0.0087
0.0138
0.0637
d.f.
0. 688
0. 433
0. 690
3.185
0. 000
1. 000
1. 000
3. 000
P-value
0.450^
85
50
50
50
-0.833
0. 865
0.376
-0.107
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
17 .4208
7 .14637
3 . 20359e + 006
Taken together, (7.14637, 3.20359e+006) is a 90
interval for the BMD
two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-77 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.00139931
3
4
5 F.2.3. Kociba et al. (1978): Female. Adenoma of Adrenal Cortex
6 F.2.3.1. Summary Table of BMDS Modeling Results
l
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
3.11
0.38
53.52
7.6E+00
4.3E+00
betas restricted >0
Multistage
cancer,
2-degree
3
3.11
0.38
53.52
7.6E+00
4.3E+00
betas restricted >0
Multistage
cancer,
3-degree
3
3.11
0.38
53.52
7.6E+00
4.3E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
8
9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-78 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.3.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.2
0.15
0.05
BMDL BMD
¦ ¦ ¦ -i ¦ '¦ ¦ ¦
0
¦ ¦
20
¦ ¦
40
¦ ¦
60
80
100
dose
15:05 11/23 2009
F.2.3.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_adre_adenoma.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_adre_adenoma.plt
Mon Nov 23 15:05:10 2009
Source - Table 5
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-79 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.00927818
Beta(1) = 0.00098105
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 0.00132464 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -24.6514 4
Fitted model -25.759 1 2.2152 3 0.529
Reduced model -31.4904 1 13.6781 3 0.003378
AIC: 53.5179
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
85
0. 000
1.0000
0.0013
0. 066
0. 000
50
-0.257
10.0000
0.0132
0. 658
2 . 000
50
1. 666
100.0000
0.1241
6.203
5. 000
50
-0.516
Chi'" 2 = 3.11 d.f. = 3 P-value = 0.3755
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 7.587 22
BMDL = 4.31737
BMDU = 17.638
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-80 DRAFT—DO NOT CITE OR QUOTE
-------
1 Taken together, (4.31737, 17.638 ) is a 90 % two-sided confidence
2 interval for the BMD
3
4 Multistage Cancer Slope Factor = 0.00231623
5
6
7 F.2.4. Kociba et al. (1978): Female, Hepatocellular Adenoma(S) or Carcinoma(s)
8 F.2.4.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
6.77
0.03
146.20
1.8E+00
1.2E+00
betas restricted >0
Multistage
cancer,
2-degree
2
6.77
0.03
146.20
1.8E+00
1.2E+00
betas restricted >0
Multistage
cancer,
3-degree
2
6.77
0.03
146.20
1.8E+00
1.2E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-81 draft—do not cite or quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.4.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
0.5
0.4
0.3
0.2
0.1
ISMDL BMD
20 40 60 80 100
dose
15:05 11/23 2009
Multistage Cancer Model with 0.95 Confidence Level
i—ii—ii—ii—ii—ii—ii—ii—ii—ii—Ii—'|—ii—ii—ii—ii—ii—ii—ii—ii—Ii—'|—ii—ii—ii—ii—ii—ii—ii—ii—Ir
Multistage Cancer
Linear extrapolation
F.2.4.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_liver_ad_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_liver_ad_carc.plt
Mon Nov 23 15:05:31 2009
Source - Table 1 in Goodman and Sauer 1992
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-82 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0591902
Beta(1) = 0.00458516
Background
Beta (1)
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
1 -0.47
-0.47 1
Variable
Background
Beta (1)
Parameter Estimates
Estimate
0.0328755
0. 00568299
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Log(likelihood)
-68.2561
Param'
4
Deviance Test d.f.
P-value
Fitted
model
-71.
0993 2
5. 68634
2
0.
Reduced
model
-89.
1983 1
41.8843
3
<.
AIC:
146
.199
Goodness
of Fit
Scaled
Dose
Est. Prob.
Expected Observed Si
Residual
0.0000
0.0329
2.827 2.
000
86
-0.500
1.0000
0.0384
1. 918 1.
000
50
-0.676
10.0000
0.0863
4.315 9.
000
50
2 . 359
100.0000
0.4521
20.346 18.
000
45
-0.703
Chi'" 2 = 6
.77
d.f. = 2
P-value
= 0.0339
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
1.7685
1.22517
2.77641
Taken together, (1.22517, 2.77641) is a 90
interval for the BMD
two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-83 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.00816214
3
4
5 F.2.5. Kociba et al. (1978): Female, Stratified Squamous Cell Carcinoma of Hard Palate
6 or Nasal Turbinates
7 F.2.5.1. Summary Table of BMDS Modeling Results
8
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.46
0.93
30.75
1.3E+01
6.5E+00
betas restricted >0
Multistage
cancer,
2-degree
3
0.04
1.00
29.96
3.5E+01
7.2E+00
betas restricted >0
Multistage
cancer,
3-degree
3
0.00
1.00
29.89
4.9E+01
7.3E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-84 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.2.5.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.2
0.15
0.05
BMDL
BMD
¦ ¦
20
I L
40
I L
60
0
80
100
dose
15:05 11/23 2009
F.2.5.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_nasal.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_nasal.plt
Mon Nov 23 15:05:50 2009
Source - Table 5
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-85 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.00343283
Beta(1) = 0.000825276
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 0.000953868 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -18.7562 4
Fitted model -19.0532 1 0.594034 3 0.8978
Reduced model -24.1972 1 10.882 3 0.01238
AIC: 40.1064
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
CO
0. 000
1.0000
0.0010
0. 048
0. 000
50
-0.218
10.0000
0.0095
0.475
1. 000
50
0.766
100.0000
0.0910
4 .458
4 . 000
49
-0.227
Chi'" 2 = 0.69 d.f. = 3 P-value = 0.8764
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 10.5364
BMDL = 5.46907
BMDU = 25.8 64
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-86 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Taken together, (5.46907, 25.864 ) is a 90 % two-sided confidence
3 interval for the BMD
4
5 Multistage Cancer Slope Factor = 0.00182846
6
7
8 F.2.6. Kociba et al. (1978): Female, Keratinizing Squamous Cell Carcinoma of Lung
9 F.2.6.1. Summary Table of BMDS Modeling Results
10
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.85
0.84
43.79
7.3E+00
4.2E+00
betas restricted >0
Multistage
cancer,
2-degree
3
0.08
0.99
42.35
2.6E+01
4.9E+00
betas restricted >0
Multistage
cancer,
3-degree
3
0.01
1.00
42.21
4.0E+01
5.0E+00
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
11
12
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-87 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.6.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.3
Multistage Cancer
Linear extrapolation
0.25
0.2
0.15
0.05
BMDL BMD
0
20
40
60
80
100
dose
15:06 11/23 2009
F.2.6.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_kera_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_kera_carc.pit
Mon Nov 23 15:06:12 2009
Source - Table 5
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-88 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0.00158635
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta(1) 0.0013747 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -20.0957 4
Fitted model -20.8959 1 1.60041 3 0.6593
Reduced model -31.4904 1 22.7894 3 <.0001
AIC: 43.7918
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
CO
0. 000
1.0000
0.0014
0. 069
0. 000
50
-0.262
10.0000
0.0137
0. 683
0. 000
50
-0.832
100.0000
0.1284
6.294
7 . 000
49
0.302
Chi'A2 = 0.85 d.f. = 3 P-value = 0.8370
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 7.31091
BMDL = 4.15929
BMDU = 14.6306
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-89 DRAFT—DO NOT CITE OR QUOTE
-------
1 Taken together, (4.15929, 14.6306) is a 90 % two-sided confidence
2 interval for the BMD
3
4 Multistage Cancer Slope Factor = 0.00240426
5
6
7 F.2.7. National Toxicology Program (1982): Female Rats, Subcutaneous Tissue,
8 Fibrosarcoma
9 F.2.7.1. Summary Table of BMDS Modeling Results
10
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
Multistage
cancer,
2-degree
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
Multistage
cancer,
3-degree
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
11
12
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-90 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.7.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.2
0.15
0.05
BMDI
IMD
0
10
20
30
40
50
60
70
dose
15:06 11/23 2009
F.2.7.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_sub_fibro.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_sub_fibro.plt
Mon Nov 23 15:06:33 2009
Source - Table 10
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-91 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.030595
Beta(1) = 0.000799545
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.54
-0.54
1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0. 0231556
0. 00102962
Std. Err.
- Indicates that this value is not calculated.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Log(likelihood)
-33.5998
-36.1883
-37.7465
76.3766
Param's
4
2
1
Deviance Test d.f.
5.17698
8.2 934 £
P-value
0.07513
0.04032
Dose
Goodness of Fit
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
1.4000
7.1000
71.0000
Chi'" 2
3.84
0.0232
0.0246
0.0303
0.0920
d.f.
1.737
1. 228
1. 514
4 . 509
0. 000
2 . 000
3. 000
4 . 000
P-value
0.1463
75
50
50
49
-1.333
0.705
1. 227
-0.252
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
9.76124
3.96354
1.03301e+006
Taken together, (3.96354, 1.03301e+006) is a 90
interval for the BMD
two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-92 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.002523
3
4
5 F.2.8. National Toxicology Program (1982): Female Rats, Liver, Neoplastic Nodule or
6 Hepatocellular Carcinoma
7 F.2.8.1. Summary Table of BMDS Modeling Results
8
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.37
0.40
133.83
2.6E+00
1.6E+00
betas restricted >0
Multistage
cancer,
2-degree
3
0.03
0.50
133.44
1.3E+01
1.7E+00
betas restricted >0
Multistage
cancer,
3-degree
3
0.00
0.50
133.44
1.3E+01
1.7E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-93 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.2.8.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.15
0.1
0.05
0 10 20 30 40 50 60 70
dose
15:07 11/23 2009
F.2.8.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_liver_nod.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_liver_nod.plt
Mon Nov 23 15:07:55 2009
Source - Table 9
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Multistage Cancer
Linear extrapolation
BMDL
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-94 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0.000900399
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta (1) 0.000775683 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -11.3484 4
Fitted model -11.6976 1 0.698469 3 0.8736
Reduced model -15.9189 1 9.14109 3 0.02747
AIC: 25.3952
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
74
0. 000
1.4000
0.0011
0. 054
0. 000
50
-0.233
7.1000
0.0055
0.275
0. 000
50
-0.525
71.0000
0.0536
2.679
3. 000
50
0.201
Chi/S2 = 0.37 d.f. = 3 P-value = 0.9462
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 12.9568
BMDL = 5.70369
BMDU = 3 9.987 8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-95 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Taken together, (5.70369, 39.9878) is a 90 % two-sided confidence
3 interval for the BMD
4
5 Multistage Cancer Slope Factor = 0.00175325
6
7
8 F.2.9. National Toxicology Program (1982): Female Rats, Adrenal, Cortical Adenoma, or
9 Carcinoma or Adenoma, NOS
10 F.2.9.1. Summary Table of BMDS Modeling Results
11
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
1.81
0.40
203.38
3.7E+00
1.9E+00
betas restricted >0
Multistage
cancer,
2-degree
2
1.38
0.50
202.89
1.6E+01
2.0E+00
betas restricted >0
Multistage
cancer,
3-degree
2
1.33
0.51
202.83
2.6E+01
2.0E+00
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
12
13
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-96 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.9.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
0.5
0.4
0.3
0.2
0.1
BMDL BMD
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
10
20
30 40
dose
50
60
70
15:07 11/23 2009
F.2.9.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_adre_cort_ad_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_adre_cort_ad_carc.pit
Mon Nov 23 15:07:15 2009
Source - Table 10
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-97 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.140663
Beta(1) = 0.00289845
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.48
-0.48
1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0.143284
0. 00273674
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Log(likelihood)
-98.7282
-99.6898
-102.201
203.38
Param's
4
2
1
Deviance Test d.f.
1.92318
6.94 636
P-value
0.3823
0. 07363
Goodness of Fit
Est. Prob.
Expected
Observed
Scaled
Residual
0.0000
1.4000
7.1000
71.0000
Chi'" 2
0.1433
0.1466
0.1598
0.2946
1. 81
d.f.
10.460
7 .181
7.829
13.551
11.000
9. 000
5. 000
14.000
P-value
0. 4046
73
49
49
46
0.180
0.735
-1.103
0.145
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
3. 67237
1. 87133
15.4002
Taken together, (1.87133, 15.4002) is a 90
interval for the BMD
two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-98 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.00534381
3
4
5 F.2.10. National Toxicology Program (1982): Female Rats, Thyroid, Follicular-Cell
6 Adenoma
7 F.2.10.1. Summary Table of BMDS Modeling Results
8
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
0.83
0.66
92.02
7.6E+00
3.5E+00
betas restricted >0
Multistage
cancer,
2-degree
2
0.53
0.77
91.64
2.3E+01
3.7E+00
betas restricted >0
Multistage
cancer,
3-degree
2
0.49
0.78
91.60
3.3E+01
3.7E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-99 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.10.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.25
0.2
0.15
0.1
0.05
0
Multistage Cancer
Linear extrapolation
BMDL
10 20 30 40 50 60 70
dose
15:07 11/23 2009
F.2.10.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_thy_ad.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_thy_ad.plt
Mon Nov 23 15:07:34 2009
Source - Table 10
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-100 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.032089
Beta(1) = 0.00143599
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.5
-0.5
1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0. 0345958
0. 00132742
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Deviance
Log(likelihood)
-43.5264
-44.0098
-46.2299
92.0196
Param's
4
2
1
Test d.f.
0. 966786
5 . 4 0699
P-value
0.6167
0.1443
Goodness of Fit
Scaled
Dose
Est. Prob.
Expected
Observed
Size
Residual
0.0000
0. 0346
2 . 525
3. 000
73
0.304
1.4000
0.0364
1. 637
2 . 000
45
0.289
7.1000
0.0437
2 .139
1. 000
49
-0.796
71.0000
0.1214
5.707
6. 000
47
0.131
d. f.
P-value
0.6614
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
7.57131
3. 48815
964541
Taken together, (3.48815, 964541 ) is a 90
interval for the BMD
two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-101 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.00286685
3
4
5 F.2.11. National Toxicology Program (1982): Male Rats, Liver, Neoplastic Nodule or
6 Hepatocellular Carcinoma
7 F.2.11.1. Summary Table of BMDS Modeling Results
8
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
3
0.37
0.40
133.83
2.6E+00
1.6E+00
betas restricted >0
Multistage
cancer,
2-degree
3
0.03
0.50
133.44
1.3E+01
1.7E+00
betas restricted >0
Multistage
cancer,
3-degree
3
0.00
0.50
133.44
1.3E+01
1.7E+00
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-102 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.11.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
¦
Multistage Cancer
Linear extrapolation
0.15
0.1
0.05
BMDL BMD
0 10 20
30 40
dose
50 60 70
15:07 11/23 2009
F.2.11.3. Output File for selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_liver_nod.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_liver_nod.plt
Mon Nov 23 15:07:55 2009
Source - Table 9
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-103 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0.000900399
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0 * * *
Beta (1) 0.000775683 * * *
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -11.3484 4
Fitted model -11.6976 1 0.698469 3 0.8736
Reduced model -15.9189 1 9.14109 3 0.02747
AIC: 25.3952
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000
0.0000
0. 000
0. 000
74
0. 000
1.4000
0.0011
0. 054
0. 000
50
-0.233
7.1000
0.0055
0.275
0. 000
50
-0.525
71.0000
0.0536
2.679
3. 000
50
0.201
Chi'" 2 = 0.37 d.f. = 3 P-value = 0.9462
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 12.9568
BMDL = 5.70369
BMDU = 3 9.987 8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-104 DRAFT—DO NOT CITE OR QUOTE
-------
1 Taken together, (5.70369, 39.9878) is a 90 % two-sided confidence
2 interval for the BMD
3
4 Multistage Cancer Slope Factor = 0.00175325
5
6
7 F.2.12. National Toxicology Program (1982): Male Rats, Thyroid, Follicular-Cell
8 Adenoma or Carcinoma
9 F.2.12.1. Summary Table of BMDS Modeling Results
10
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
7.14
0.03
151.22
3.5E+00
1.9E+00
betas restricted >0
Multistage
cancer,
2-degree
2
7.14
0.03
151.22
3.5E+00
1.9E+00
betas restricted >0
Multistage
cancer,
3-degree
2
7.14
0.03
151.22
3.5E+00
1.9E+00
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
11
12
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-105 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.2.12.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0 10 20 30 40 50 60 70
dose
15:08 11/23 2009
F.2.12.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_thyroid.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_thyroid.plt
Mon Nov 23 15:08:16 2009
Source - Table 9
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
8MDL BMD
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-106 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0867382
Beta(1) = 0.00232055
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.53
Beta(1) -0.53 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.0704713 * * *
Beta(1) 0.00285481 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -69.5946 4
Fitted model -73.6119 2 8.03468 2 0.018
Reduced model -77.5267 1 15.8643 3 0.001209
AIC: 151.224
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0705 4.863 1.000 69 -1.817
1.4000 0.0742 3.561 5.000 48 0.793
7.1000 0.0891 4.456 8.000 50 1.759
71.0000 0.2410 12.051 11.000 50 -0.347
Chi^2 = 7.14 d.f. = 2 P-value = 0.0281
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 3 . 5205
BMDL = 1 . 91558
BMDU = 9.7 6663
Taken together, (1.91558, 9.76663) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-107 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.00522034
3
4
5 F.2.13. National Toxicology Program (1982): Male Rats, Adrenal cortex, Adenoma
6 F.2.13.1. Summary Table of BMDS Modeling Results
l
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
5.83
0.05
199.67
1.4E+01
3.4E+00
betas restricted >0
Multistage
cancer,
2-degree
2
5.83
0.05
199.67
1.4E+01
3.4E+00
betas restricted >0
Multistage
cancer,
3-degree
2
5.83
0.05
199.67
1.4E+01
3.4E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
8
9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-108 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.13.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.4
0.35
0.3
0.25
0.2
0.15
0.05
BMDI
BMD
0
10
20
30
40
50
60
70
dose
15:08 11/23 2009
F.2.13.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_adre_cort.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_adre_cort.plt
Mon Nov 23 15:08:35 2009
Source - Table 9
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-109 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.168444
Beta(1) = 0.000395949
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.53
Beta(1) -0.53 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.153096 * * *
Beta(1) 0.000718012
*
*
- Indicates that this value is
not
calculated.
Analysis
of
Deviance Table
Model Log(likelihood
) #
Param's Deviance
Test
d.f. P-value
Full model -94.8672
4
Fitted model -97.8359
2 5.93732
2 0.05137
Reduced model -98.0432
1 6.35197
3 0.09569
AIC: 199.672
Goodness of Fit
Scaled
Dose Est. Prob. Expected
Observed Si
ze
Residual
0.0000 0.1531 11.
023
6. 000
72
-1.644
1.4000 0.1539 7.
697
9. 000
50
0.510
7.1000 0.1574 7.
713
12.000
49
1. 682
71.0000 0.1952 9.
564
9. 000
49
-0.203
Chi'" 2 = 5. 83 d. f. = 2
P
-value = 0.0541
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 13.9974
BMDL = 3.4 4 43
BMDU did not converge for BMR = 0.010000
BMDU calculation failed
BMDU = Inf
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3 F.2.14. National Toxicology Program (1982): Female Mice, Subcutaneous Tissue,
4 Fibrosarcoma
5 F.2.14.1. Summary Table of BMDS Modeling Results
6
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
Multistage
cancer,
2-degree
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
Multistage
cancer,
3-degree
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
7
8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 11 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
F.2.14.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.25
0.2
0.15
0.1
0.05
BMDL BMD
50
Multistage Cancer
Linear extrapolation
100
150
dose
200
250
300
15:08 11/23 2009
F.2.14.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_subcu_fibro.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_subcu_fibro.plt
Mon Nov 23 15:08:56 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Maximum number of iterations = 250
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0143554
Beta(1) = 0.000341874
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.5
Beta(1) -0.5 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.0145028 * * *
Beta(1) 0.000338561 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -30.9876 4
Fitted model -31.0199 2 0.0645971 2 0.9682
Reduced model -34.3291 1 6.68308 3 0.08272
AIC: 6 6.0397
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0145 1.073 1.000 74 -0.071
5.7000 0.0164 0.820 1.000 50 0.200
28.6000 0.0240 1.152 1.000 48 -0.143
286.0000 0.1055 4.956 5.000 47 0.021
Chi/'2 = 0.07 d.f. = 2 P-value = 0.9675
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 29.6855
BMDL = 14.3524
BMDU = 100.382
Taken together, (14.3524, 100.382) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-113 DRAFT—DO NOT CITE OR QUOTE
-------
1 Multistage Cancer Slope Factor = 0.000696747
2
3
4 F.2.15. National Toxicology Program (1982): Female Mice, Hematopoietic System,
5 Lymphoma
6 F.2.15.1. Summary Table of BMDS Modeling Results
l
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
0.03
0.99
261.43
1.0E+01
5.5E+00
betas restricted >0
Multistage
cancer,
2-degree
2
0.03
0.99
261.43
1.0E+01
5.5E+00
betas restricted >0
Multistage
cancer,
3-degree
2
0.03
0.99
261.43
1.0E+01
5.5E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
8
9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.15.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
0.6
0.5
0.4
0.3
0.2
3MDL
BMD
0
50
100
150
200
250
300
dose
15:09 11/23 2009
F.2.15.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_mice_f_lymphoma.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_mice_f_lymphoma.plt
Mon Nov 23 15:09:17 2009
Table 15 page 64 Hematopoietic System Lymphoma or Leukemia
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.242959
Beta(1) = 0.000967723
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.48
-0.48
1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0.242712
0.000971954
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Log(likelihood) # Param'
-128.699 4
-128.712 2
-131.412 1
261.425
Deviance Test d.f.
0.0264819
5. 42487
P-value
0.9868
0.1432
Goodness of Fit
Est. Prob.
Expected
Observed
Scaled
Residual
0.
5.
28 .
286.
Chi'" 2
0000
7000
6000
0000
0. 03
2427
2469
2635
4265
d.f.
17.961
12 . 345
12.647
20.045
18 .
12 .
13.
20.
P-value
000
000
000
000
74
50
48
47
0. 011
-0.113
0.116
-0.013
0.9868
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
10.3403
5. 45599
38 . 9139
Taken together, (5.45599, 38.9139) is a 90
interval for the BMD
two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.00183285
3
4
5 F.2.16. National Toxicology Program (1982): Female Mice, Liver, Hepatocellular
6 Adenoma or Carcinoma
7 F.2.16.1. Summary Table of BMDS Modeling Results
8
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
2.82
0.24
156.00
1.5E+01
7.8E+00
betas restricted >0
Multistage
cancer,
2-degree
2
2.82
0.24
156.00
1.5E+01
7.8E+00
betas restricted >0
Multistage
cancer,
3-degree
2
2.82
0.24
156.00
1.5E+01
7.8E+00
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria.
'Best-fitting model as assessed by lowest-AIC criterion, bolded.
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.2.16.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
BMDL BMD
50
100
150
dose
200
250
300
15:09 11/23 2009
F.2.16.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_mice_f_liv_aden_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_mice_f_liv_aden_carc.plt
Mon Nov 23 15:09:36 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-118 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0888873
Beta(1) = 0.000616931
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.5
-0.5
1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0. 0788077
0.000689385
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Log(likelihood)
-74.5177
Param'
4
Deviance Test d.f.
P-value
Fitted
model
-76.
0006 2 2.96597
2
0.227
Reduced
model
-79.
6703 1 10.
3053
3
0.
. 01614
AIC:
156
. 001
Goodness of Fit
Scaled
Dose
Est. Prob.
Expected Observed
Size
Residual
0.0000
0.0788
5.753 3.000
73
-1.196
5.7000
0.0824
4.121 6.000
50
0.966
28.6000
0.0968
4.646 6.000
48
0. 661
286.0000
0.2436
11.452 11.000
47
-0.153
Chi/S2 = 2
. 82
d.f. = 2
P-value = 0.2436
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
14.5787
7.82902
42.4536
Taken together, (7.82902, 42.4536) is a 90 % two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-l 19 DRAFT—DO NOT CITE OR QUOTE
-------
1 interval for the BMD
2
3 Multistage Cancer Slope Factor = 0.0012773
4
5
6 F.2.17. National Toxicology Program (1982): Female Mice, Thyroid Follicular Cell
7 Adenoma
8 F.2.17.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
Multistage
cancer,
2-degree
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
Multistage
cancer,
3-degree
2
3.84
0.15
76.38
9.8E+00
4.0E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-120 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.17.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.25
0.2
0.15
0.1
0.05
0 50 100 150 200 250 300
dose
15:09 11/23 2009
F.2.17.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_mice_f_thyroid_aden.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_mice_f_thyroid_aden.plt
Mon Nov 23 15:09:56 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
Multistage Cancer
Linear extrapolation
BMDL
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-121 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.02405
Beta(1) = 0.000315564
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.51
Beta(1) -0.51 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.0207192 * * *
Beta(1) 0.000331835 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -32.0017 4
Fitted model -34.6122 2 5.22112 2 0.07349
Reduced model -37.2405 1 10.4776 3 0.01491
AIC: 73.2245
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.0207 1.430 0.000 69 -1.208
5.7000 0.0226 1.128 3.000 50 1.782
28.6000 0.0300 1.409 1.000 47 -0.350
286.0000 0.1094 5.032 5.000 46 -0.015
Chi^2 = 4.76 d.f. = 2 P-value = 0.0927
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 30.2871
BMDL = 13.993
BMDU = 130.014
Taken together, (13.993 , 130.014) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-122 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.000714641
3
4
5 F.2.18. National Toxicology Program (1982): Male Mice, Lung, Alveolar/Bronchiolar
6 Adenoma or Carcinoma
7 F.2.18.1. Summary Table of BMDS Modeling Results
8
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
3.97
0.14
167.34
3.7E+00
2.0E+00
betas restricted >0
Multistage
cancer,
2-degree
2
3.41
0.18
166.81
1.6E+01
2.1E+00
betas restricted >0
Multistage
cancer,
3-degree
2
3.38
0.18
166.78
2.6E+01
2.1E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-123 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
F.2.18.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
Multistage Cancer Model with 0.95 Confidence Level
Multistage Cancer
Linear extrapolation
BMDL BMD
10
20
30
40
50
60
70
dose
15:10 11/23 2009
F.2.18.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_lung_aden_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_lung_aden_carc.plt
Mon Nov 23 15:10:17 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-124 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.0827179
Beta(1) = 0.00298266
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.49
-0.49
1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0. 0925449
0. 00271189
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Log(likelihood) # Param'
-79.5959 4
-81.6704 2
-85.3351 1
167.341
Deviance Test d.f.
4 .14885
11.4782
P-value
0.1256
0. 009402
Dose
Goodness of Fit
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
1.4000
7.1000
71.0000
Chi/S2
0.0925
0.0960
0.1099
0.2515
3. 97
6.571
4 . 607
5.273
12 . 574
10.000
2 . 000
4 . 000
13.000
d.f.
P-value
0.1375
71
48
48
50
1.404
-1.278
-0.588
0.139
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
3.70603
2.0263
10.562
Taken together, (2.0263 , 10.562 ) is a 90 % two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-125 DRAFT—DO NOT CITE OR QUOTE
-------
1 interval for the BMD
2
3 Multistage Cancer Slope Factor = 0.0049351
4
5
6 F.2.19. National Toxicology Program (1982): Male Mice, Liver, Hepatocellular Adenoma
7 or Carcinoma
8 F.2.19.1. Summary Table of BMDS Modeling Results
9
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
2
0.17
0.92
258.57
1.3E+00
8.6E-01
betas restricted >0
Multistage
cancer,
2-degree
2
0.17
0.92
258.57
1.3E+00
8.6E-01
betas restricted >0
Multistage
cancer,
3-degree
2
0.17
0.92
258.57
1.3E+00
8.6E-01
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
10
11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-126 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.19.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.7
0.6
0.5
0.4
0.3
liMDL BMD
30 40
dose
15:10 11/23 2009
Multistage Cancer
Linear extrapolation
F.2.19.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_mice_m_liver_aden_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_mice_m_liver_aden_carc.plt
Mon Nov 23 15:10:37 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 4
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-127 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.22264
Beta(1) = 0.0074005
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background
Beta(1)
1
-0.46
-0.46
1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0.219315
0. 00750879
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Log(likelihood) # Param'
-127.199 4
-127.286 2
-135.589 1
258.572
Deviance Test d.f.
0.174343
16.7801
P-value
0.9165
0.0007843
Goodness of Fit
Est. Prob.
Expected
Observed
Scaled
Residual
0.
1.
7 .
71.
Chi'" 2
0000
4000
1000
0000
0.2193
0.2275
0.2598
0.5419
16.010
11.146
12 .732
27.096
0.17
d.f.
15.
12 .
13.
27 .
P-value
000
000
000
000
73
49
49
50
-0.286
0.291
0. 087
-0.027
0.9164
Benchmark Dose Computation
Specified effect
Risk Type
Confidence level
BMD
BMDL
BMDU
0. 01
Extra risk
0. 95
1.33848
0. 861975
2 .4671
Taken together, (0.861975, 2.4671
interval for the BMD
is a 90
two-sided confidence
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-128 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Multistage Cancer Slope Factor = 0.0116013
3
4
5 F.2.20. National Toxicology Program (2006): Liver, Cholangiocarcinoma
6 F.2.20.1. Summary Table of BMDS Modeling Results
l
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
1-degree
5
12.91
0.02
129.07
1.9E+00
1.4E+00
betas restricted >0
Multistage
cancer,
2-degree
5
1.18
0.95
114.35
9.4E+00
5.3E+00
betas restricted >0
Multistag
e cancer,
3-degreeb
4
0.22
0.99
115.16
1.3E+01
4.5E+00
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
8
9
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-129 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.20.2. Figure for Selected Model: Multistage Cancer, 3-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.6
0.5
0.4
0.3
0.2
0.1
Multistage Cancer
Linear extrapolation
BMDL
BMD
10
20
30 40
dose
50
60
70
15:10 11/23 2009
F.2.20.3. Output File for Selected Model: Multistage Cancer, 3-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\msc3_ngkgd_liv_cho-carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\msc3_ngkgd_liv_cho-carc.plt
Mon Nov 23 15:10:57 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2-beta3*dose/N3) ]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 4
Total number of specified parameters = 0
Degree of polynomial = 3
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-130 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0.000561481
Beta(2) = 1.74365e-005
Beta(3) = 1.40248e-006
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background -Beta(l)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(2) Beta(3)
Beta(2) 1 -0.99
Beta(3) -0.99 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Beta(3)
Estimate
0
0
4 . 35 92 7e-0 05
1.14186e-006
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-55.408
-55.5789
-96.9934
Param's Deviance Test d.f.
0.34181
83.1708
P-value
0. 987
:. 0001
AIC:
115.158
Goodness of Fit
Scaled
Dose
Est
. Prob.
Expected
Observed
Size
Residual
0.0000
0.
0000
0. 000
0
000
49
0
000
2.1400
0.
0002
0. 010
0
000
48
-0
101
7.1400
0.
0026
0.121
0
000
46
-0
349
15.7000
0.
0150
0.752
1
000
50
0
288
32.9000
0.
0841
4 121
4
000
49
-0
062
71.4000
0.
4716
24.994
25
000
53
0
002
Chi/S2
0.22
d.f.
P-value
0.9945
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-131 draft—do not cite or quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Confidence level =
0. 95
BMD = 13.1014
BMDL = 4 . 46755
BMDU = 19.17 83
Taken together, (4.46755, 19.1783) is a 90 % two-sided confidence
interval for the BMD
Multistage Cancer Slope Factor = 0.00223836
F.2.21. National Toxicology Program (2006): Liver, Hepatocellular Adenoma
F.2.21.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
1-degree
5
8.50
0.13
82.31
4.4E+00
2.9E+00
betas restricted >0
Multistage
cancer,
2-degree
5
1.94
0.86
73.66
1.5E+01
8.6E+00
betas restricted >0
Multistag
e cancer,
3-degreeb
5
0.24
1.00
71.22
2.4E+01
1.2E+01
betas restricted >0
"Values <0.1 fail to meet BMDS goodness-of-fit criteria
bBest-fitting model as assessed by lowest-AIC criterion, bolded
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-132 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.21.2. Figure for Selected Model: Multistage Cancer, 3-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0 10 20 30 40 50 60 70
dose
15:11 11/23 2009
F.2.21.3. Output File for Selected Model: Multistage Cancer, 3-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\msc3_ngkgd_liv_hepat_ad.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\msc3_ngkgd_liv_hepat_ad.plt
Mon Nov 23 15:11:17 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2-beta3*dose/N3) ]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 4
Total number of specified parameters = 0
Degree of polynomial = 3
Multistage Cancer
Linear extrapolation
BMDLj
BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-133 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0
Beta(2) = 0
Beta(3) = 7.77141e-007
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background -Beta(l) -Beta(2)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(3)
Beta(3) 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Beta(3)
Estimate
0
0
0
7 . 46408e-007
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-34.4075
-34.6078
-56.3333
Param's Deviance Test d.f.
0.40065
43.8515
P-value
0.9953
C.0001
71. 2156
Dose
Est. Prob.
Goodness of Fit
Expected
Observed
Size
Scaled
Residual
0.0000
0
0000
0
000
0
000
49
0
000
2.1400
0
0000
0
000
0
000
48
-0
019
7.1400
0
0003
0
012
0
000
46
-0
112
15.7000
0
0029
0
144
0
000
50
-0
380
32.9000
0
0262
1
285
1
000
49
-0
255
71.4000
0
2379
12
609
13
000
53
0
126
Chi/S2
0.24
d.f.
P-value
0.9986
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-134 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMD = 23.7 904
2
3 BMDL = 11.5343
4
5 BMDU = 27.8755
6
7 Taken together, (11.5343, 27.8755) is a 90 % two-sided confidence
8 interval for the BMD
9
10 Multistage Cancer Slope Factor = 0.000866978
11
12
13 F.2.22. National Toxicology Program (2006): Oral Mucosa, Squamous Cell Carcinoma
14 F.2.22.1. Summary Table of BMDS Modeling Results
15
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Value3
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
multistage
cancer,
1-degree
4
4.15
0.39
125.48
4.8E+00
3.0E+00
betas restricted >0
Multistage
cancer,
2-degreeb
4
2.83
0.59
123.25
1.6E+01
3.8E+00
betas restricted >0,
bound hit
Multistage
cancer,
3-degree
4
2.83
0.59
123.25
1.6E+01
3.8E+00
betas restricted >0,
bound hit
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
b Best-fitting model as assessed by lowest-AIC criterion, bolded.
16
17
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-135 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.22.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0,
Bound Hit
Multistage Cancer Model with 0.95 Confidence Level
0.35
Multistage Cancer
Linear extrapolation
0.3
0.25
0.2
0.15
0.05
BMDL BMD
0
10
20
30
40
50
60
70
dose
15:11 11/23 2009
F.2.22.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0,
Bound Hit
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_oral_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_oral_carc.pit
Mon Nov 23 15:11:37 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-136 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Degree of polynomial = 1
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.00607545
Beta(1) = 0.00265195
Background
Beta(1)
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
1 -0.6
-0.6 1
Variable
Background
Beta(1)
Parameter Estimates
Estimate
0. 0171416
0. 00211536
Std. Err.
- Indicates that this value is not calculated.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
Model
Full model
Fitted model
Reduced model
AIC:
Analysis of Deviance Table
Param's Deviance Test d.f.
Log(likelihood)
-57.5353
-60.7418
-67.7782
125.484
6.41293
20.4858
P-value
0.1704
0. 001013
Dose
Goodness of Fit
Est. Prob.
Expected
Observed
Size
Scaled
Residual
0.0000
0.0171
0
840
1
000
49
0
176
2.1400
0.0216
1
036
2
000
48
0
958
7.1400
0.0319
1
466
1
000
46
-0
391
15.7000
0.0492
2
462
0
000
50
-1
609
32.9000
0.0832
4
078
4
000
49
-0
040
71.4000
0.1549
8
211
10
000
53
0
679
Chi ^2
4 .15
d.f.
P-value
0.3855
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 4 .75111
BMDL = 2 . 9556
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-137 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 BMDU = 9.194 54
3
4 Taken together, (2.9556 , 9.19454) is a 90 % two-sided confidence
5 interval for the BMD
6
7 Multistage Cancer Slope Factor = 0.0033834
8
9
10 F.2.23. National Toxicology Program (2006): Pancreas, Adenoma or Carcinoma
11 F.2.23.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
l-degreeb
5
2.37
0.80
28.32
2.1E+01
9.3E+00
betas restricted >0
Multistage
cancer,
2-degree
5
0.80
0.98
26.23
3.3E+01
1.4E+01
betas restricted >0
Multistage
cancer,
3-degree
5
0.32
1.00
25.43
4.1E+01
1.8E+01
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
13
14
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-138 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.23.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.15
0.1
0.05
0
Multistage Cancer
Linear extrapolation
o
o
BMDI
BMD
10
20
30
40
50
60
70
dose
15:11 11/23 2009
F.2.23.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_panc_ad_carc.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_panc_ad_carc.pit
Mon Nov 23 15:11:58 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-139 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0.000817541
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -Background
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta (1)
Beta(1) 1
Parameter Estimates
Variable
Background
Beta(1)
Estimate
0
0.000474004
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
- Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-11.4096
-13.1581
-16.7086
Param's Deviance Test d.f.
P-value
3.49702
10.598
0.6238
0.05996
AIC:
28.3163
Est. Prob.
Goodness of Fit
Expected Observed
Scaled
Residual
0.0000
0.0000
0
000
0
000
48
0
000
2.1400
0.0010
0
049
0
000
48
-0
221
7.1400
0.0034
0
155
0
000
46
-0
395
15.7000
0.0074
0
371
0
000
50
-0
611
32.9000
0.0155
0
743
0
000
48
-0
869
71.4000
0.0333
1
697
3
000
51
1
017
Chi ^2 = 2.37 d.f. = 5 P-value = 0.7964
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 21.2031
BMDL = 9.33481
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-140 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMDU = 65.4351
2
3 Taken together, (9.33481, 65.4351) is a 90 % two-sided confidence
4 interval for the BMD
5
6 Multistage Cancer Slope Factor = 0.00107126
7
8
9 F.2.24. National Toxicology Program (2006): Lung, Cystic Keratinizing Epithelioma
10 F.2.24.1. Summary Table of BMDS Modeling Results
11
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
Multistage
cancer,
1-degree
5
7.42
0.19
60.81
6.9E+00
4.2E+00
betas restricted >0
Multistage
cancer,
2-degreeb
5
2.54
0.77
54.36
1.9E+01
1.1E+01
betas restricted >0
Multistage
cancer,
3-degree
5
1.02
0.96
51.85
2.8E+01
1.6E+01
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
bBest-fitting model as assessed by lowest-AIC criterion, bolded.
12
13
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-141 draft—do not cite or quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.24.2. Figure for Selected Model: Multistage Cancer, 2-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.3
0.25
0.2
0.15
0.1
0.05
Multistage Cancer
Linear extrapolation
BMDL
BMD
10
20
30
40
50
60
70
dose
15:12 11/23 2009
F.2.24.3. Output File for Selected Model: Multistage Cancer, 2-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\msc2_ngkgd_lung_epith.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\msc2_ngkgd_lung_epith.plt
Mon Nov 23 15:12:20 2009
0
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose/sl-beta2*dose/N2 ) ]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 6
Total number of records with missing values = 0
Total number of parameters in model = 3
Total number of specified parameters = 0
Degree of polynomial = 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-142 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0
Beta(1) = 0
Beta(2) = 3.77591e-005
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -Background -Beta(l)
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
Beta(2)
Beta(2) 1
Parameter Estimates
Variable
Background
Beta(1)
Beta(2)
Estimate
0
0
2.91011e-005
Std. Err.
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model
Full model
Fitted model
Reduced model
Log(likelihood)
-23.958
-26.1815
-40.2069
Param's Deviance Test d.f.
P-value
4 .44693
32.4976
0. 487
<.0001
AIC:
54.363
Dose
Est. Prob.
Goodness of Fit
Expected Observed Size
Scaled
Residual
0.0000
0.0000
0
000
0
000
49
0
000
2.1400
0.0001
0
006
0
000
48
-0
080
7.1400
0.0015
0
068
0
000
46
-0
261
15.7000
0.0071
0
350
0
000
49
-0
594
32.9000
0.0310
1
519
0
000
49
-1
252
71.4000
0.1379
7
170
9
000
52
0
736
Chi/N2 = 2.54 d.f. = 5 P-value = 0.7708
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 18.5839
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-143 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMDL = 10.687 8
2
3 BMDU = 25.1324
4
5 Taken together, (10.6878, 25.1324) is a 90 % two-sided confidence
6 interval for the BMD
7
8 Multistage Cancer Slope Factor = 0.000935646
9
10
11 F.2.25. Toth et al. (1978): 1YR, Liver, Tumors
12 F.2.25.1. Summary Table of BMDS Modeling Results
13
Model
Degrees
of
Freedom
X2 Test
Statistic
x2
p-Valuea
AIC
BMD
(ng/kg-day)
BMDL
(ng/kg-day)
Model Notes
multistage
cancer,
l-degreeb
1
1.30
0.25
155.95
2.7E+00
1.5E+00
betas restricted >0
multistage
cancer,
2-degree
1
1.30
0.25
155.95
2.7E+00
1.5E+00
betas restricted >0
multistage
cancer,
0-degree
0.25
-999.00
error
error
betas restricted >0
aValues <0.1 fail to meet BMDS goodness-of-fit criteria.
''Best-fitting model as assessed by lowest-AIC criterion, bolded.
14
15
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-144 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
F.2.25.2. Figure for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model with 0.95 Confidence Level
0.6
0.5
0.4
0.3
0.2
0.1
Multistage Cancer
Linear extrapolation
SMDL BMD
20
40 60
dose
80
100
17:30 11/23 2009
F.2.25.3. Output File for Selected Model: Multistage Cancer, 1-Degree, Betas Restricted >0
Multistage Cancer Model. (Version: 1.7; Date: 05/16/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_adr_cor_lyr.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\mscl_ngkgd_adr_cor_lyr.plt
Mon Nov 23 17:30:17 2009
Table 1
The form of the probability function is:
P[response] = background + (1-background)*[1-EXP(
-betal*dose^l)]
The parameter betas are restricted to be positive
Dependent variable = Mean
Independent variable = Dose
Total number of observations = 3
Total number of records with missing values = 0
Total number of parameters in model = 2
Total number of specified parameters = 0
Degree of polynomial = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-145 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
Background = 0.240176
Beta(1) = 0.00374745
Asymptotic Correlation Matrix of Parameter Estimates
Background Beta(1)
Background 1 -0.53
Beta(1) -0.53 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
Background 0.2418 * * *
Beta(1) 0.00373791 * * *
* - Indicates that this value is not calculated.
Analysis of Deviance Table
Model Log(likelihood) # Param's Deviance Test d.f. P-value
Full model -75.3127 3
Fitted model -75.9728 2 1.3201 1 0.2506
Reduced model -79.4897 1 8.35401 2 0.01534
AIC: 155.946
Goodness of Fit
Scaled
Dose Est. Prob. Expected Observed Size Residual
0.0000 0.2418 9.188 7.000 38 -0.829
1.0000 0.2446 10.764 13.000 44 0.784
100.0000 0.4783 21.044 21.000 44 -0.013
Chi ^2 = 1.30 d.f. = 1 P-value = 0.2537
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2.68876
BMDL = 1.52183
BMDU = 7 . 54263
Taken together, (1.52183, 7.54263) is a 90 % two-sided confidence
interval for the BMD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-146 DRAFT—DO NOT CITE OR QUOTE
-------
1 Multistage Cancer Slope Factor =
2
0. 00657103
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-147 DRAFT—DO NOT CITE OR QUOTE
-------
S"4
>3*
&
to
S
S
to
s
>;*
a
a,
Sf
>i
>i
to
TO'
*
^5
«
s
TO
>;
0
s
1
a,
§•
TO
>;
s
o
^•k
TO
O
s
2! »
O ^
H r?
<*^
hh Oq
H to
W|
o
o
H
W
F.3. HUMAN EQUIVALENT DOSES FOR 1, 5, AND 10% EXTRA RISK
Comparison of Human Equivalent Doses from Benchmark Dose Modeling Assuming 1%, 5%, and 10% Extra Risk
BMDoi
BMDL01
BMD0S
BMDL0S
BMD10
BMDL10
HED
HED
HED
HED
HED
HED
Study
Species
Sex
Morphology: topography
(ng/kg-
day)
(ng/kg-
day)
(ng/kg-
day)
(ng/kg-
day)
(ng/kg-
day)
(ng/kg-
day)
Kociba,
Rat
Male
Stratified squamous cell carcinoma of hard palate or nasal
4.7E-01
1.6E-01
4.6E+00
1.7E+00
1.1E+01
4.6E+00
1978
turbinates
Stratified squamous cell carcinoma of tongue
5.1E-01
1.4E-01
4.9E+00
1.6E+00
1.2E+01
4.1E+00
Adenoma of adrenal cortex
2.0E-01
8.5E-02
2.1E+00
9.7E-01
5.5E+00
2.6E+00
Female
Hepatocellular adenoma(s) or carcinoma(s)
1.9E-02
1.2E-02
2.3E-01
1.4E-01
6.7E-01
4.1E-01
Stratified squamous cell carcinoma of hard palate or nasal
3.3E-01
1.2E-01
3.3E+00
1.3E+00
8.3E+00
3.6E+00
turbinates
Keratinizing squamous cell carcinoma of lung
1.9E-01
8.0E-02
2.0E+00
9.2E-01
5.3E+00
2.5E+00
NTP, 1982
Rat
Female
Subcutaneous tissue: fibrosarcoma
1.8E-01
5.3E-02
2.0E+00
6.2E-01
5.2E+00
1.7E+00
Liver: neoplastic nodule or hepatocellular carcinoma
4.2E-02
2.1E-02
4.9E-01
2.4E-01
1.4E+00
7.1E-01
Adrenal: cortical adenoma, or carcinoma or adenoma, NOS
6.8E-02
2.4E-02
7.8E-01
2.8E-01
2.2E+00
8.2E-01
Thyroid: follicular-cell adenoma
2.1E-01
6.4E-02
2.2E+00
7.4E-01
5.7E+00
2.0E+00
Male
Liver: neoplastic nodule or hepatocellular carcinoma
5.1E-01
1.5E-01
4.9E+00
1.6E+00
1.2E+01
4.3E+00
Thyroid: follicular-cell adenoma or carcinoma
4.3E-02
1.9E-02
5.1E-01
2.3E-01
1.4E+00
6.6E-01
Adrenal cortex: adenoma
2.6E-01
4.4E-02
2.8E+00
5.2E-01
7.0E+00
1.5E+00
Mouse
Female
Subcutaneous tissue: fibrosarcoma
2.1E-01
7.2E-02
2.2E+00
8.3E-01
5.8E+00
2.3E+00
Hematopoietic system: lymphoma or leukemia
4.0E-02
1.5E-02
4.7E-01
1.8E-01
1.3E+00
5.3E-01
Liver: hepatocellular adenoma or carcinoma
5.9E-02
2.4E-02
6.9E-01
2.9E-01
1.9E+00
8.3E-01
Thyroid: follicular-cell adenoma
1.8E-01
5.6E-02
1.9E+00
6.5E-01
5.0E+00
1.8E+00
Male
Lung: alveolar/bronchiolar adenoma or carcinoma
1.3E-01
8.6E-03
4.6E-01
1.0E-01
7.7E-01
3.0E-01
Liver: hepatocellular adenoma or carcinoma
3.1E-03
1.7E-03
3.7E-02
1.9E-02
1.1E-01
5.7E-02
NTP, 2006
Rat
Female
Liver: cholangiocarcinoma
7.0E-01
2.9E-01
1.5E+00
1.2E+00
2.1E+00
1.8E+00
Liver: hepatocellular adenoma
1.1E+00
5.6E-01
2.3E+00
1.8E+00
3.2E+00
2.7E+00
Oral mucosa: squamous cell carcinoma
1.1E-01
5.5E-02
1.2E+00
6.4E-01
3.3E+00
1.8E+00
Pancreas: adenoma or carcinoma
1.1E+00
3.4E-01
5.3E+00
3.1E+00
8.3E+00
5.0E+00
Lung: cystic keratinizing epithelioma
8.0E-01
4.1E-01
2.5E+00
1.8E+00
4.1E+00
2.9E+00
Toth, 1979
Mouse
Male
Liver: tumors
5.1E-03
1.9E-03
6.7E-02
2.7E-02
2.0E-01
8.5E-02
r1
00
o
O
-------
F.4. REFERENCES
1 Goodman, DG; Sauer, RM. (1992) Hepatotoxicity and carcinogenicity in female Sprague-Dawley rats treated with
2 2,3,7,8-tetrachlorordibenzo-p-dioxin (TCDD): a Pathology Working Group reevaluation. Regul Toxicol Pharmacol
3 15:245-252.
4 Kociba, RJ; Keyes, DG; Beyer, JE; et al. (1978) Results of a two-year chronic toxicity and oncogenicity study of
5 2,3,7,8-tetrachlorodibenzo-p-dioxin in rats. Toxicol Appl Pharmacol 46(2):279-303.
6 NTP (National Toxicology Program). (1982) Bioassay of 2,3,7,8-tetrachlorodibenzo-p-dioxin for possible
7 carcinogenicity (gavage study). Tech. Rept. Ser. No. 201. U.S. Department of Health and Human Services, Public
8 Health Service, Research Triangle Park, NC.
9 NTP (National Toxicology Program). (2006) Studies of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) in female
10 Harlan Sprague-Dawley rats (gavage studies) Tech. Rep. Ser. No. 521. U.S. Department of Health and Human
11 Services, Public Health Service, Research Triangle Park, NC.
12 Toth, KJ; Sugar, S; Somfai-Relle S; et al. (1978) Carcinogenic bioassay of the herbicide 2,4,5-trichlorphenoxy
13 ethanol (TCPE) with Swiss mice. Prog Biochem Pharmacol 14:82-93.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 F-149 DRAFT—DO NOT CITE OR QUOTE
-------
-------
DRAFT
DO NOT CITE OR QUOTE
January 2010
Agency/Interagency Review Draft
APPENDIX G
Endpoints Excluded From Reference Dose
Derivation Based on Toxicological Relevance
NOTICE
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
1 APPENDIX G. ENDPOINTS EXCLUDED FROM REFERENCE DOSE DERIVATION
2 BASED ON TOXICOLOGICAL RELEVANCE
3
4
5 The National Academy of Sciences (NAS) committee commented on the low dose model
6 predictions and the need to discuss the biological significance of the noncancer health effects
7 modeled in the 2003 Reassessment. In selecting point of departure (POD) candidates from the
8 animal bioassays for derivation of the reference dose (RfD), U.S. Environmental Protection
9 Agency (EPA) had to consider the toxicological relevance of the identified endpoint(s) from any
10 given study. Often endpoints/effects may be sensitive, but lack general toxicological
11 significance due to not being clearly adverse (defined in the Integrated Risk Information System
12 (IRIS) glossary as a biochemical change, functional impairment, or pathologic lesion that affects
13 the performance of the whole organism, or reduces an organism's ability to respond to an
14 additional environmental challenge), being an adaptive response, or not being clearly linked to
15 downstream functional or pathological alterations. It is standard EPA RfD derivation policy not
16 to base a reference value on endpoints that are not adverse or not obvious precursors to an
17 adverse effect. For select studies, a rationale for lack of toxicological relevance of particular
18 endpoints reported is listed here. These endpoints were not considered for derivation of the RfD.
19 Kitchin and Woods (1979) administered female Sprague-Dawley rats a single gavage
20 dose of 2,3,7,8-tetrachl orodibenzo-/>dioxin (TCDD) and measured cytochrome P450 levels and
21 benzo(a)pyrene hydroxylase (BPH) activity as a marker of hepatic microsomal cytochrome
22 P448-mediated enzyme activity. They found a statistically significant increase in BPH at doses
23 >2 ng/kg and a significant increase in cytochrome P450 levels at doses >600 ng/kg. Aryl
24 hydrocarbon hydrolase and EROD were both significantly increased 3 months after exposure;
25 however the elevation did not maintain statistical significance at 6 months. No other indicators
26 of hepatic effects were analyzed. CYP induction alone is not considered a significant
27 toxicologically adverse effect given that CYPs are induced as a means of hepatic processing of
28 xenobiotic agents. Additionally, the role of CYP induction in hepatotoxicity and carcinogenicity
29 of TCDD is unknown, and CYP induction is not considered a relevant POD without obvious
30 pathological significance.
31 In multiple studies by Hassoun et al. (1998, 2000, 2002, 2003), various indicators of
32 oxidative stress were measured in hepatic and brain tissue of female B6C3F1 mice and Sprague-
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 G-l DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Dawley rats following 13 or 30 weeks of TCDD gavage dosing (5 days a week). Biomarkers for
oxidative stress included production superoxide anion, lipid peroxidation, and DNA single-strand
breaks. The authors report a statistically significant effect on several oxidative stress markers as
a result of TCDD exposure, the lowest dose producing an effect being 0.32 ng/kg-day (Hassoun
et al., 1998). In this study, all oxidative stress markers were significantly effected, but no other
indicators of brain pathology were assessed. Thus, it is impracticable to link the markers of
oxidative stress to a toxicological outcome in the brain, and this study and its endpoints are not
considered relevant POD candidates.
Burleson et al. (1996) analyzed the effect of a TCDD on viral host resistance following a
single gavage dose of TCDD by measuring mortality mediated by influenza virus challenge in
B6C3F1 female mice. The study authors found that TCDD at >10 ng/kg-day increased
influenza-induced mortality. The experimental design calls for a 30% mortality in untreated
animals (15% was achieved); mortality, itself, is not a direct result of TCDD exposure. None of
the other immunologically-relevant measures were affected by TCDD treatment in this study,
and no other effects were reported. The interpretation of these results with respect to humans is
problematic. Furthermore, the findings were not reproduced by Nohara et al. (2002) using the
same experimental design (see Section 2.4.2). Therefore, this endpoint is not considered relevant
as a POD candidate.
To examine the central nervous system response to TCDD, Kuchiiwa et al (2002)
analyzed the effect of in utero and lactational TCDD exposure on the serotonergic system in the
brainstem of male ddY mice. Female mice were administered TCDD by oral gavage once a
week for 8 weeks prior to pregnancy and, using an immunocytochemical detection method, the
raphe nuclei in the brainstem of male offspring was monitored for serotogergic neurons. TCDD
at 0.7 ng/kg-day caused a 25—50% reduction in the immunostaining of serotonin, however there
were no differences in external morphology, birth or postnatal body weights between
TCDD-exposed and control offspring. The authors suggest that these findings may indicate that
TCDD acts as a neuroteratogen by mediating long-term alterations in neuronal serotonin
synthesis and serotonergic function. However, no other relevant neurotoxicity endpoints were
examined or reported. Thus, reduced serotonin is not an adverse endpoint of toxicological
significance in and of itself, and this study is deemed unsuitable as a POD candidate.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 G-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Mally and Chipman (2002) evaluated the effect of TCDD on gap junctions,
hypothesizing that as a nongenotoxic carcinogen, TCDD may induce tumor formation by
disturbing tissue homeostasis. Female F344 rats were dosed with TCDD by oral gavage for
either 3 consecutive days or 2 days a week for 28 days. Gap junction connexin (Cx) plaque
expression and hepatocyte proliferation was measured. The study authors report a decrease in
Cx32 plaque number and area in the liver of rats exposed to 0.7 ng/kg-day and higher, however
they did not find an associated increase in hepatocyte proliferation. No clinical signs of toxicity
were observed, and histological examination of the liver revealed no abnormalities. In the
absence of additional indicators of hepatotoxicity, a decrease in Cx32 plaque formation is not
clearly linked to TCDD-mediated hepatotoxicity or hepatocarcinogenicty, nor is it considered an
adverse effect. This endpoint is not considered a toxicologically relevant POD.
Vanden Heuvel et al. (1994) analyzed changes in hepatic mRNA following a single
administration of TCDD to female Sprague-Dawley rats by oral gavage. Four days after
treatment, animals were sacrificed and livers were excised. Using reverse transcriptase-
polymerase chain reaction (RT-PCR) on hepatic RNA, they compared levels of "dioxin
responsive" mRNA's (CYP1A1, UDP-glucuronosyltransferase I, plasminogen activator inhibitor
2, and transforming growth factor a) at various doses of TCDD and at control (baseline) levels.
They determined that CYP1 Al elicited the most sensitive response to TCDD, with a statistically
significant increase (3-fold) in mRNA from rat livers exposed to 1 ng/kg-day TCDD. Induction
of CYP1 Al expression is not considered an adverse effect, as the role of CYP1 Al in
TCDD-mediated carcinogenicity is unsettled. Therefore, in the absence of other indicators of
hepatoxicity, increases in liver CYP1 Al cannot be considered toxicologically relevant for a POD
candidate.
Devito et al. (1994) assessed the activity of CYP1A1 and CYP1A2, the amount of
phosphorylation of phosphotyrosyl proteins (pp32, pp34, and pp38), and the levels of estrogen
receptor in the liver, uterus, lung and skin tissue of female B6C3F1 mice administered TCDD for
5 days a week for 13 weeks. The authors hypothesized that these measurements may be
sensitive biomarkers for exposure to TCDD. Body weights were also recorded weekly.
Induction of CY1A1 and CYP1A2, as well as increased phosphorylated forms of pp32, pp34,
and pp38 were sensitive indicators of TCDD exposure, with statistically significant changes seen
at 1.07 ng/kg-day. EROD activity in the ling, skin, and liver was also observed with significant
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 G-3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
increases at this dose. However, the authors did not find a change in rat body or terminal organ
weights, nor did they note any pathology in the animals at this dose level. The role of CYPs and
phosphorylated pp32, pp34, and pp38 in TCDD-mediated toxicity is unknown, and changes in
the activity or function of these proteins are not considered adverse. Therefore, these endpoints
are not considered suitable as PODs.
Because TCDD had been detected in the soil of contaminated locations, determining the
bioavailability of TCDD from ingested soil may be important to the calculation of safe exposure
levels. Lucier et al. (1986) fed adult female Sprague-Dawley rats TCDD contaminated soil or
gave them TCDD in corn oil at various doses and compared the effects of TCDD on biochemical
parameters from liver tissue. They found that equivalent doses of TCDD in corn oil and soil
produced similar increases in hepatic aryl hydrocarbon hydroxylase activity (AHH) and UDP
glucuronyltransferase activity. They determined that AHH was statistically induced 1.8-fold at
15 ng/kg in corn oil and 40 ng/kg in soil. Cytochrome P450 was significantly increased at higher
doses. No clinical signs of acute toxicity or changes in body weight were observed. The
association between AHH activity and TCDD-mediated hepatotoxicity is unknown and no
adverse endpoints were measured. Thus, this endpoint is not suitable as a POD candidate.
Sugita-Konishi et al. (2003) investigated the change in host resistance of mice offspring
lactationally exposed to TCDD. Pregnant C57BL/6NCji mice were administered TCDD via
drinking water from parturition to weaning of the offspring (17 days). One group of offspring
was then infected with Listeria monocytogenes and blood and spleen samples were collected
various time points post infection. Uninfected, TCDD exposed offspring were weighed and their
spleens and thymuses removed for assay of cellular content and protein expression. TCDD
exposure caused a statistically-significant decrease in relative spleen weight and a statistically-
significant increase in thymic CD4+ cells in the high-dose group (11.3 ng/kg-day). Offspring
infected with Listeria following TCDD exposure exhibited a statistically significant increase in
serum tumor necrosis factor alpha (TNF-a) 2 days after infection in both sexes in the low-
(1.14 ng/kg-day) and high-dose groups. The authors conclude that exposure to TCDD disrupted
the host resistance of the offspring at the lowest dose tested, despite the primary immune
parameters being unaffected. Without an obvious association between TCDD and immune
function, however, this endpoint is not suitable for identification of a LOAEL. Thus, the
LOAEL for this study is 11.3 ng/kg-day, and the NOAEL is 1.14 ng/kg-day.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 G-4 DRAFT—DO NOT CITE OR QUOTE
-------
1 Sewall et al. (1993) investigated alterations in the epidermal growth factor receptor
2 (EGFR) pathway in a two-stage initiation promotion model of TCDD hepatic cancer. EGFR
3 signaling has been implicated in the altered cell growth induction by tumor promoters. Female
4 Sprague-Dawley rats were administered TCDD biweekly by oral gavage for 30 weeks following
5 initiation by a single dose of diethylnitrosamine (DEN). A group also received TCDD without
6 prior DEN initiation. Livers were harvested and fixed from sacrificed animals and sections
7 tested for EGFR binding, autophosphorylation, immunolocalization, and hepatic cell
8 proliferation. The authors report a significant dose-dependent decrease in plasma membrane
9 EGFR maximum binding capacity in TCDD-exposed rats beginning at 3.5 ng/kg-day. However,
10 at this same dose, the authors note a statistically significant decrease in cell proliferation (as
11 measured by DNA replication labeling), with increases in proliferation only occurring at higher
12 doses (125 ng/kg-day). No other indicators of hepatic toxicity or tumorigenicity were assessed.
13 The role of EGFR in TCDD-mediated hepatotoxicity and hepatocarcinogenicity is unknown, and
14 as such, this endpoint cannot be unequivocally linked to TCDD-induced hepatic effects nor
15 labeled as adverse. Thus, it is not suitable as a POD candidate.
16
17 G.l. REFERENCES
18 Burleson, GR; Lebrec, H; Yang, YG; et al. (1996) Effect of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) on
19 influenza virus host resistance in mice. Fund Appl Toxicol 29:40-47.
20 Devito, MJ; Ma, X; Babish, JG; et al. (1994) Dose-response relationships in mice following subchronic exposure to
21 2,3,7,8-tetrachlorodibenzo-p-dioxin: CYP1A1, CYP1A2, estrogen receptor, and protein tyrosine phosphoylation.
22 Toxicol Appl Pharmacol 124:82-90.
23 Hassoun, EA; Wilt, SC; DeVito, MJ; et al. (1998) Induction of oxidative stress in brain tissues of mice after
24 subchronic exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol Sci 42:23-27.
25 Hassoun, EA; Li, F; Abushaban, A; et al. (2000) The relative abilities of TCDD and its congeners to induce
26 oxidative stress in the hepatic and brain tissues of rats after subchronic exposure. Toxicology 145:103-113.
27 Hassoun, EA; Wang, H; Abushaban, A. (2002) Induction of oxidative stress following chronic exposure to TCDD,
28 2,3,4,7,8-pentachlorodibenzofuran, and 2,3 ',4,4',5-pentachlorobiphenyl. J Toxicol Environ Health A 65:825-842.
29 Hassoun, EA; Al-Ghafri, M; Abushaban, A. (2003) The role of antioxidant enzymes in TCDD-induced oxidative
30 stress in various brain regions of rats after subchronic exposure. Free Rad Biol Medicine 35(9): 1028-1036.
31 Kitchin, KT; Woods, JS. (1979) 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) effects on hepatic microsomal
32 cytochrome P-448-mediated enzyme activities. Toxicol Appl Pharmacol 47:537-546.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 G-5 DRAFT—DO NOT CITE OR QUOTE
-------
1 Kuchiiwa, S; Cheng, S-B; Nagatomo, I; et al. (2002) In utero and lactational exposure to 2,3,7,8-tetrachlorodibenso-
2 p-dioxin decreases serotonin-immunoreactive neurons in raphe nuclei of male mouse offspring. Neurosci Lett
3 317:73-76.
4 Lucier, GW; Rumbaugh, RC; McCoy, Z; et al. (1986) Ingestion of soil contaminated with 2,3,7,8-tetrachloro-
5 dibenzo-p-dioxin (TCDD) alters hepatic enzyme activities in rats. Fund Appl Toxicol 6:364-371.
6 Mally, A; Chipman, JK. (2002) Non-genotoxic carcinogens: early effects on gap junctions, cell proliferation and
7 apoptosis in the rat. Toxicology 180:233-248.
8 Nohara, K; Izumi, H; Tamura, S; et al. (2002) Effect of low-dose 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) on
9 influenza A virus-induced mortality in mice. Toxicology 170:131-138.
10 Sewall, CH; Lucier, GW; Tritscher, AM; et al. (1993) TCDD-mediated changes in hepatic epidermal growth factor
11 receptor may be a critical event in the hepatocarcinogenic action of TCDD. Carcinogenesis 14:1885-1893.
12 Sugita-Konishi, Y; Kobayashi, K; Naito, H; et al. (2003) Effect of lactational exposure to 2,3,7,8-
13 tetrachlorodibenzo-p-dioxin on the susceptibility to Listeria infection. Biosci Biotechnol Biochem 67(l):89-93.
14 U.S. EPA. (2003) Exposure and human health reassessment of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and
15 related compounds [NAS review draft]. Volumes 1-3. National Center for Environmental Assessment, Washington,
16 DC; EPA/600/P-00/001 Cb. Available at: http://www.epa.gov/nceawwwl/pdfs/dioxin/nas-review/.
17 Vanden Heuvel, JP; Clark, GC; Tritscher, A; et al. (1994) Accumulation of polychlorinated dibenzo-p-dioxins and
18 dibenzofurans in liver of control laboratory rats. Fundam Appl Toxicol 23:465-469.
19
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 G-6 DRAFT—DO NOT CITE OR QUOTE
-------
-------
DRAFT
DO NOT CITE OR QUOTE
January 2010
Agency/Interagency Review Draft
APPENDIX H
Cancer Precursor Benchmark Dose Modeling
NOTICE
THIS DOCUMENT IS AN AGENCY/INTERAGENCY REVIEW DRAFT. It has not been
formally released by the U.S. Environmental Protection Agency and should not at this stage be
construed to represent Agency policy. It is being circulated for comment on its technical
accuracy and policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, OH
-------
CONTENTS—APPENDIX H: Cancer Precursor Benchmark Dose Modeling
APPENDIX H. CANCER PRECURSOR BENCHMARK DOSE MODELING 1
H. 1. BMDS INPUT TABLES 1
H. 1.1. Hassoun et al. (2000) 1
II.1.2. Kitchinetal. (1979) 1
H. 1.3. National Toxicology Program. (2006), 31 Week Exposure 2
H. 1.4. National Toxicology Program (2006), 53 Week Exposure 2
H. 1.5. National Toxicology Program (2006), 2 Year Exposure 2
H.1.6. Van Birgelen et al. (1995a, b) 3
H.1.7. Vanden Heuvel et al. (1994) 3
II.2. ALTERNATE DOSE: BLOOD SERUM BMDS RESULTS 3
H.2.1. Hassoun et al. (2000): CytC liver 3
H.2.1.1. Summary Table of BMDS Modeling Results 3
H.2.1.2. Figure for Selected Model: Exponential (M5), Constant
Variance, Power Restricted >1 5
H.2.1.3. Output File for Selected Model: Exponential (M5),
Constant Variance, Power Restricted >1 5
H.2.1.4. Figure for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted 8
H.2.1.5. Output file for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted 8
H.2.1.6. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted 11
H.2.1.7. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted 11
H.2.1.8. Figure for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 14
H.2.1.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 14
11.2.2. Hassoun et al. (2000): DNA SSB 17
H.2.2.1. Summary Table of BMDS Modeling Results 17
H.2.2.2. Figure for Selected Model: Exponential (M4), Constant
Variance, Power Restricted >1 18
H.2.2.3. Output File For Selected Model: Exponential (M4),
Constant Variance, Power Restricted >1 18
H.2.2.4. Figure for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted 21
H.2.2.5. Output File for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted 21
H.2.2.6. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted 24
H.2.2.7. Output File For Unrestricted Model: Hill, Constant
Variance, n Unrestricted 24
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-ii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
H.2.2.8. Figure for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 27
H.2.2.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 27
H.2.3. Hassoun et al. (2000): TBARs Liver 30
H.2.3.1. Summary Table ofBMDS Modeling Results 30
H.2.3.2. Figure for Selected Model: Exponential (M5), Constant
Variance, Power Unrestricted 31
H.2.3.3. Output File for Selected Model: Exponential (M5),
Constant Variance, Power Unrestricted 31
H.2.3.4. Figure for Unrestricted Model: Hill, Constant Variance, n
Restricted >1 34
H.2.3.5. Output File for Unrestricted Model: Hill, Constant
Variance, n Restricted >1 34
H.2.3.6. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted 37
H.2.3.7. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted 37
H.2.3.8. Figure for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 40
H.2.3.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 40
H.2.4. Kitchin et al. (1979): BaP Hydrolase Activity 43
H.2.4.1. Summary Table ofBMDS Modeling Results 43
H.2.4.2. Figure for Selected Model: Exponential (M5),
Nonconstant Variance, Power Restricted >1 44
H.2.4.3. Output File for Selected Model: Exponential (M5),
Nonconstant Variance, Power Restricted >1 44
H.2.4.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 47
H.2.4.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 47
H.2.4.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 50
H.2.4.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 51
H.2.4.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 53
H.2.4.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 53
H.2.5. National Toxicology Program. (2006): EROD Liver Week 53 57
H.2.5.1. Summary Table ofBMDS Modeling Results 57
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-iii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
H.2.5.2. Figure for Selected Model: Hill, Nonconstant Variance, n
Restricted >1 58
H.2.5.3. Output file for Selected Model: Hill, Nonconstant
Variance, n Restricted >1 58
H.2.5.4. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 61
H.2.5.5. Output file for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 61
H.2.6. National Toxicology Program. (2006): Lung EROD Week 31 64
H.2.6.1. Summary Table of BMDS Modeling Results 64
H.2.6.2. Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 65
H.2.6.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 65
H.2.6.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 68
H.2.6.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 68
H.2.6.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 71
H.2.6.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 71
H.2.6.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 74
H.2.6.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 74
H.2.7. National Toxicology Program. (2006): Lung EROD Week 53 77
H.2.7.1. Summary Table of BMDS Modeling Results 77
H.2.7.2. Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 78
H.2.7.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 78
H.2.7.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 81
H.2.7.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 81
H.2.7.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 84
H.2.7.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 84
H.2.7.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 87
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-iv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
H.2.7.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 87
H.2.8. National Toxicology Program. (2006): Tbll 1 Index Week 31 90
H.2.8.1. Summary Table of BMDS Modeling Results 90
H.2.8.2. Figure for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 91
H.2.8.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 91
11.2.9. Van Birgelen et al. (1995b): T4 UGT 94
H.2.9.1. Summary Table of BMDS Modeling Results 94
H.2.9.2. Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 95
H.2.9.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 95
H.2.9.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 98
H.2.9.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 98
H.2.9.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 101
H.2.9.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 101
H.2.9.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 105
H.2.9.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 105
11.2.10. Van Birgelen et al. (1995b): UGT 1A1 108
H.2.10.1. Summary Table of BMDS Modeling Results 108
H.2.10.2.Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 109
H.2.10.3.Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 109
H.2.10.4.Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 112
H.2.10.5.Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 112
H.2.10.6.Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 115
H.2.10.7.Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 115
H.2.10.8.Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 118
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-v DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
H.2.10.9.Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 118
H.2.11. Vanden Heuvel et al. (1994): Hepatic CYP1A1 mRNA Expression... 121
H.2.11.1. Summary Table ofBMDS Modeling Results 121
H.2.11.2.Figure for Selected Model: Exponential (M5),
Nonconstant Variance, Power Restricted >1 122
H.2.11.3.Output File for Selected Model: Exponential (M5),
Nonconstant Variance, Power Restricted >1 122
11.3. ADMINISTERED DOSE BMDS RESULTS 125
H.3.1. Hassoun et al. (2000): CytC Liver 125
H.3.1.1. Summary Table ofBMDS Modeling Results 125
H.3.1.2. Figure for Selected Model: Exponential (M4), Constant
Variance, Power Restricted >1 126
H.3.1.3. Output File for Selected Model: Exponential (M4),
Constant Variance, Power Restricted >1 126
H.3.1.4. Figure for Unrestricted Model: Exponential (M5), Constant
Variance, Power Unrestricted 129
H.3.1.5. Output File for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted 129
H.3.1.6. Figure for Unrestricted Model: Hill, Constant Variance, n
Unrestricted 132
H.3.1.7. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted 132
H.3.1.8. Figure for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 135
H.3.1.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 135
11.3.2. Hassoun et al. (2000): DNA SSB 138
H.3.2.1. Summary Table ofBMDS Modeling Results 138
H.3.2.2. Figure for Selected Model: Exponential (M5), Constant
Variance, Power Unrestricted 139
H.3.2.3. Output File for Selected Model: Exponential (M5),
Constant Variance, Power Unrestricted 139
H.3.2.4. Figure for Unrestricted Model: Hill, Constant Variance, n
Restricted >1, Bound Hit 142
H.3.2.5. Output File for Unrestricted Model: Hill, Constant
Variance, n Restricted >1, Bound Hit 142
H.3.2.6. Figure for Unrestricted Model: Hill, Constant Variance,
n Unrestricted 145
H.3.2.7. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted 145
H.3.2.8. Figure for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 148
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-vi DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
H.3.2.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 148
H.3.3. Hassoun et al. (2000): TBARs Liver 151
H.3.3.1. Summary Table of BMDS Modeling Results 151
H.3.3.2. Figure for Selected Model: Exponential (M4), Constant
Variance, Power Restricted >1 152
H.3.3.3. Output File for Selected Model: Exponential (M4),
Constant Variance, Power Restricted >1 152
H.3.3.4. Figure for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted 155
H.3.3.5. Output File for Unrestricted Model: Exponential (M5),
Constant Variance, Power Unrestricted 155
H.3.3.6. Figure for Unrestricted Model: Hill, Constant Variance,
n Unrestricted 158
H.3.3.7. Output File for Unrestricted Model: Hill, Constant
Variance, n Unrestricted 158
H.3.3.8. Figure for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 161
H.3.3.9. Output File for Unrestricted Model: Power, Constant
Variance, Power Unrestricted 161
H.3.4. Kitchin et al. (1979): BaP Hydrolase Activity 164
H.3.4.1. Summary Table of BMDS Modeling Results 164
H.3.4.2. Figure for Selected Model: Exponential (M5),
Nonconstant Variance, Power Restricted >1 165
H.3.4.3. Output File for Selected Model: Exponential (M5),
Nonconstant Variance, Power Restricted >1 165
H.3.4.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 168
H.3.4.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 168
H.3.4.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 171
H.3.4.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 172
H.3.4.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 174
H.3.4.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 175
H.3.5. National Toxicology Program. (2006): EROD Liver Week 53 177
H.3.5.1. Summary Table of BMDS Modeling Results 177
H.3.5.2. Figure for Selected Model: Hill, Nonconstant Variance,
n Restricted >1 178
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-vii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
H.3.5.3. Output File for Selected Model: Hill, Nonconstant
Variance, n Restricted >1 179
H.3.5.4. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 181
H.3.5.5. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 181
H.3.6. National Toxicology Program. (2006): Lung EROD Week 31 184
H.3.6.1. Summary Table ofBMDS Modeling Results 184
H.3.6.2. Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 185
H.3.6.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 185
H.3.6.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 188
H.3.6.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 188
H.3.6.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 191
H.3.6.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 191
H.3.6.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 194
H.3.6.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 194
H.3.7. National Toxicology Program. (2006): Lung EROD Week 53 197
H.3.7.1. Summary Table ofBMDS Modeling Results 197
H.3.7.2. Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 198
H.3.7.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 198
H.3.7.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 201
H.3.7.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 201
H.3.7.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 204
H.3.7.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 204
H.3.7.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 207
H.3.7.9. Output File for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 207
H.3.8. National Toxicology Program. (2006): Tblll Index Week 31 210
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-viii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
H.3.8.1. Summary Table ofBMDS Modeling Results 210
H.3.8.2. Figure for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 211
H.3.8.3. Output File for Selected Model: Exponential (M2),
Nonconstant Variance, Power Restricted >1 211
11.3.9. Van Birgelenetal. (1995b): T4 UGT 214
H.3.9.1. Summary Table ofBMDS Modeling Results 214
H.3.9.2. Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 215
H.3.9.3. Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 215
H.3.9.4. Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 218
H.3.9.5. Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 218
H.3.9.6. Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 221
H.3.9.7. Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 221
H.3.9.8. Figure for Unrestricted Model: Power, Nonconstant
Variance, Power Unrestricted 224
H.3.9.9. Output File for Unrestricted Model: Power,
Nonconstant Variance, Power Unrestricted 224
11.3.10. Van Birgelen et al. (1995b): UGT 1A1 227
H.3.10.1. Summary Table ofBMDS Modeling Results 227
H.3.10.2.Figure for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 228
H.3.10.3.Output File for Selected Model: Exponential (M4),
Nonconstant Variance, Power Restricted >1 228
H.3.10.4.Figure for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 231
H.3.10.5.Output File for Unrestricted Model: Exponential (M5),
Nonconstant Variance, Power Unrestricted 231
H.3.10.6.Figure for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 234
H.3.10.7.Output File for Unrestricted Model: Hill, Nonconstant
Variance, n Unrestricted 234
H.3.10.8.Figure for Unrestricted Model: Power, Nonconstant Variance,
Power Unrestricted 237
H.3.10.9.Output File for Unrestricted Model: Power,
Nonconstant Variance, Power Unrestricted 237
H.3.11. Vanden Heuvel et al. (1994): Hepatic CYP1A1 mRNA
Expression 240
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-ix DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
H.3.11.1.Summary Table ofBMDS Modeling Results 240
H.3.11.2.Figure for Selected Model: Exponential (M5),
Nonconstant Variance, Power Restricted >1 241
H.3.11.3.Output File for Selected Model: Exponential (M5),
Nonconstant Variance, Power Restricted >1 241
H.4. REFERENCES 244
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-x DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX H. CANCER PRECURSOR BENCHMARK DOSE MODELING
2
3
4 H.l. BMDS INPUT TABLES
5 H.l.l. Hassoun et al. (2000)
Endpoint
Administered Dose (ng/kg-day)
0
3
10
22
46
100
Internal Dose (ng/kg blood) a
0
1,068
2,542
4,489
7,718
13,960
n = 6
n = 6
n = 6
n = 6
n = 6
n = 6
CytC liver
0.15 ±0.07
0.18 ± 0.05b
0.19 ±0.06
0.27 ± 0.06c
0.39 ± 0.06c
0.44 ± 0.11c
DNA SSB
7.41 ± 1.54
10.78 ± 1.25 b-c
13.6 ± 1.69c
15.3 ± 1.71c
20.4 ± 2.25c
23.5 ± 1.37c
TBARs liver
1.47 ±0.29
1.55 ± 0.54b
2.15 ± 0.36c
2.28 ± 0.25c
2.62 ± 0.52c
2.29 ± 0.49c
aFrom the Emond PRPK model described in 3.3.
bLOEL for selected endpoint.
Statistically significant as compared to control (p < 0.05).
6
7
8 H.1.2. Kitchin et al. (1979)
Endpoint
Administered Dose (ng/kg-day)
0
0.2
0.667
1.33
6.67
20
Internal Dose (ng/kg blood) a
0
70
232
463
2,318
6,949
n = 9
n = 4
n = 4
n = 4
n = 4
n = 4
BaP hydrolase activity
(continued on next line)
4.9 ±1.11
4.9 ± 1.18 b
6.7 ± 1.4 c'd
7.2 ± 1.8d
8.3 ± 0.26e
14 ± 5e
Endpoint
Administered Dose (ng/kg-day)
66.7
200
667
1,670
6,670
Internal Dose (ng/kg blood) a
23,185
69,657
232,550
581,930
2,332,100
n = 4
n = 4
n = 4
n = 4
n = 4
BaP hydrolase activity
(continued)
59 ± 6.8e
96 ± 46e
155 ± 16.4e
182±26e
189±26e
aFrom the Emond PRPK model described in 3.3.
''NOEL for selected endpoint.
°LOEL for selected endpoint.
Statistically significant as compared to control (p < 0.05).
"Statistically significant as compared to control (p < 0.001).
9
10
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l DRAFT—DO NOT CITE OR QUOTE
-------
1 H.1.3. National Toxicology Program. (2006), 31 Week Exposure
Endpoint
Administered Dose (ng/kg-day)
0
2.14
7.14
15.7
32.9
71.4
Internal Dose (ng/kg blood) a
0
1,284
2,932
5,075
8,629
15,503
n = 9
n = 10
n = 10
n = 10
n = 10
n = 10
Tbll 1 Index ,week 31
0.33 ± 0.19
0.85 ± 0.65 b
0.96 ± 0.74b
0.79 ± 0.46 b
1.33 ± 1.12 b
3.85 ± 3.08b
Lung EROD, week 31
2.07 ±0.97
25.34 ± 2.55c
30.39 ± 5.83c
50.19 ± 8.68c
49.07 ± 13.91c
48.42 ± 8.93 c
aFrom the Emond PRPK model described in 3.3.
Statistically significant as compared to control (p < 0.05).
0Statistically significant as compared to control (p < 0.01).
2
3
4 H.1.4. National Toxicology Program (2006), 53 Week Exposure
Endpoint
Administered Dose (ng/kg-day)
0
2.14
7.14
15.7
32.9
71.4
Internal Dose (ng/kg blood) a
0
1,354
3,056
5,259
8,918
16,001
n = 8
n = 8
n = 8
n = 8
n = 8
n = 8
Liver EROD, week 53
3.61 ±0.49
7.27 ± 0.56b
14.76 ± 1.61 b
17.28 ± 1.59b
20.58 ± 3.05 b
21.21 ±3.82b
Lung EROD, week 53
3.01 ± 1.58
27.15 ±5.27b
42.85 ± 11.15 b
36.57 ± 12.99b
43.75 ± 18.55 b
43.71 ±6.32b
"From the Emond PRPK model described in 3.3.
Statistically significant as compared to control (p < 0.01).
5
6
7 H.1.5. National Toxicology Program (2006), 2 Year Exposure
Administered Dose (ng/kg-day)
0
2.14
7.14
15.7
32.9
71.4
Internal Dose (ng/kg blood) a
0
1,408
3,137
5,393
9,129
16,361
Endpoint
n = 53
n = 54
n = 53
n = 53
n = 53
n = 53
Toxic Hepatopathy
0/53 (0%)
2/54 (0%)
8/53 (20%)b
30/53 (60%)b
45/53 (80%)b
53/53 (100%)b
aFrom the Emond PRPK model described in 3.3.
Statistically significant as compared to control (p < 0.01).
8
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-2 DRAFT—DO NOT CITE OR QUOTE
-------
1 H.1.6. Van Birgelen et al. (1995a, b)
Endpoint
Administered Dose (ng/kg-day)
0
14
26
47
320
1,024
Internal Dose (ng/kg blood) a
0
3,969
6,479
9,968
47,606
137,820
n = 8
n = 8
n = 8
n = 8
n = 8
n = 8
T4UGT
0.33 ±0.2
0.6 ±0.42
0.64 ± 0.45 b
0.87 ± 0.91 b
2.08 ± 1.33 b
2.59 ± 0.88b
UGT 1A1
101 ± 15.59
194 ± 36.37b
Not reported.
304 ± 17.32 b
452 ± 48.5 b
296 ± 148.96 b
aFrom the Emond PRPK model described in 3.3.
Statistically significant as compared to control (p < 0.05).
2
3
4 H.1.7. Vanden Heuvel et al. (1994)
Administered Dose (ng/kg-day)
0
0.1
1
10
100
1,000
10,000
Endpoint
Internal Dose (ng/kg blood) a
0
4
36
302
2,149
14,301
114,690
n = 13
n = 5
n= 12
n = 7
n = 7
n = 11
n = 5
Hepatic CYP1A1
mRNA Expression
5.4 ±3.61
7.2 ±5.59
14.8 ± 14.9 b
12.8 ±4.5b
536 ±320.14 b
18000 ±
15223.31 b
36700 ±
22137.07b
aFrom the Emond PRPK model described in 3.3.
Statistically significant as compared to control (p < 0.05).
5
6
7 H.2. ALTERNATE DOSE: BLOOD SERUM BMDS RESULTS
8 H.2.1. Hassoun et al. (2000): CytC liver
9 H.2.1.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
0.39
10.22
0.04
-145.92
3.5E+03
2.5E+03
nonconstant variance,
power restricted >1
exponential (M3)
4
0.39
10.22
0.04
-145.92
3.5E+03
2.5E+03
nonconstant variance,
power restricted >1
exponential (M4)
3
0.39
3.38
0.34
-150.76
1.6E+03
9.7E+02
nonconstant variance,
power restricted >1
exponential (M5)
2
0.39
0.56
0.76
-151.58
3.0E+03
1.4E+03
nonconstant variance,
power restricted >1
exponential (M5)
2
0.39
0.56
0.76
-151.58
3.0E+03
1.4E+03
nonconstant variance,
power unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-3 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/>-Value3
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
Hill
2
0.39
0.66
0.72
-151.47
3.1E+03
error
nonconstant variance,
n restricted >1
Hill
2
0.39
0.66
0.72
-151.47
3.1E+03
error
nonconstant variance,
n unrestricted
linear
4
0.39
4.68
0.32
-151.45
2.1E+03
1.4E+03
nonconstant variance
polynomial
4
0.39
4.68
0.32
-151.45
2.1E+03
1.4E+03
nonconstant variance
power
4
0.39
4.68
0.32
-151.45
2.1E+03
1.4E+03
nonconstant variance,
power restricted >1,
bound hit
power
3
0.39
4.26
0.23
-149.87
1.6E+03
6.8E+02
nonconstant variance,
power unrestricted
exponential (M2)
4
0.39
12.17
0.02
-143.33
5.1E+03
4.3E+03
constant variance,
power restricted >1
exponential (M3)
4
0.39
12.17
0.02
-143.33
5.1E+03
4.3E+03
constant variance,
power restricted >1
exponential (M4)
3
0.39
3.37
0.34
-150.14
1.9E+03
1.2E+03
constant variance,
power restricted >1
exponential (M5)c
2
0.39
0.48
0.79
-151.03
3.3E+03
1.7E+03
constant variance,
power restricted >1
exponential (M5)d
2
0.39
0.48
0.79
-151.03
3.3E+03
1.7E+03
constant variance,
power unrestricted
Hill
2
0.39
0.59
0.74
-150.91
3.4E+03
1.8E+03
constant variance, n
restricted >1
Hilld
2
0.39
0.59
0.74
-150.91
3.4E+03
1.8E+03
constant variance, n
unrestricted
linear
4
0.39
6.42
0.17
-149.09
3.1E+03
2.4E+03
constant variance
polynomial
4
0.39
6.42
0.17
-149.09
3.1E+03
2.4E+03
constant variance
power
4
0.39
6.42
0.17
-149.09
3.1E+03
2.4E+03
constant variance,
power restricted >1,
bound hit
power d
3
0.39
4.84
0.18
-148.66
1.8E+03
7.4E+02
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAlternate model also presented in this appendix
1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-4 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.2.1.2. Figure for Selected Model: Exponential (M5), Constant Variance, Power Restricted
>1
Exponential_beta Model 5 with 0.95 Confidence Level
0.6
0.5
0.4
0.3
0.2
0.1
Exponential
BMDL
2000
4000
6000 8000
dose
10000
12000
14000
12:51 11/23 2009
H.2.1.3. Output File for Selected Model: Exponential (M5), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Exp_CV_BMRl_CytC_Liver.(d)
Gnuplot Plotting File:
Mon Nov 23 12:51:43 2009
TBARs, liver only (Table 2)
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 5
lnalpha
rho(S)
-5.48625
0
0.1387
4.08 913e-005
6.40231
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 5
-5.47298
0
0.156024
0. 00016181
2.85354
2.14237
Table of Stats From Input Data
Dose
Obs Mean
Obs Std Dev
0
6
0.
146
0
06614
1068
6
0.
177
0
05389
2542
6
0.
191
0
05634
4489
6
0.
271
0
05634
7718
6
0.
388
0
06369
1. 396e + 004
6
0.444
0.1102
Estimatec
Values
of
Interest
Est Mean
Est Std
Scaled Residual
0
0.156
0
0648
-0.3789
1068
0.1627
0
0648
0.5416
2542
0.1961
0
0648
-0.1919
4489
0.2705
0
0648
0. 01767
7718
0.3874
0
0648
0.02225
3 9 6e + 0 0 4
0.4443
0
0648
-0.0107
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 80.75258 7 -147.5052
A2 83.37355 12 -142.7471
A3 80.75258 7 -147.5052
R 55.82002 2 -107.64
5 80.51365 5 -151.0273
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
55.11
5.242
5.242
0.4779
p-value
10
5
5
2
< 0.0001
0.3871
0.3871
0.7875
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 3257.85
BMDL = 17 0 9.28
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.2.1.4. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
0.6
Exponential
0.5
0.4
0.3
0.2
0.1
BMDL
BMD
0
2000
4000
6000
8000
10000
12000
14000
dose
12:51 11/23 2009
H.2.1.5. Output file for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Exp_CV_Unrest_BMRl_CytC_Liver.(d)
Gnuplot Plotting File:
Mon Nov 23 12:51:49 2009
TBARs, liver only (Table 2)
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable
Model 5
lnalpha
rho(S)
-5.48625
0
0.1387
4.08 913e-005
6.40231
1
(S)
Specified
Parameter Estimates
Variable
lnalpha
rho
Model 5
-5.47298
0
0.156024
0. 00016181
2.85354
2.14237
Table of Stats From Input Data
Dose
Obs Mean
Obs Std Dev
0
6
0
146
0
06614
1068
6
0
177
0
05389
2542
6
0
191
0
05634
4489
6
0
271
0
05634
7718
6
0
388
0
06369
1.396e+004
0.444
0.1102
Estimated
Values of
Interest
Dose
Est Mean
Est
Std
Scaled Residual
0
0.156
0.
0648
-0.3789
1068
0.1627
0.
0648
0.5416
2542
0.1961
0.
0648
-0.1919
4489
0.2705
0.
0648
0. 01767
7718
0.3874
0.
0648
0.02225
3 9 6e + 0 0 4
0.4443
0.
0648
-0.0107
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-9 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Var{e(ij)} = Sigma(i)/X2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 80.75258 7 -147.5052
A2 83.37355 12 -142.7471
A3 80.75258 7 -147.5052
R 55.82002 2 -107.64
5 80.51365 5 -151.0273
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test
1:
Test
2 :
Test
3:
Test -2*log(Likelihood Ratio)
Test 1 55.11
Test 2 5.242
Test 3 5.242
Test 7a 0.477 9
p-value
10
5
5
2
< 0.0001
0.3871
0.3871
0.7875
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 3257.85
BMDL = 17 0 9.28
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
H.2.1.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
0.6 F
0.5
0.4
0.3
0.2
0.1
Hill
BMDL
BMD
2000
4000
6000 8000
dose
10000
12000
14000
12:51 11/23 2009
H.2.1.7. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Hill_CV_Unrest_BMRl_CytC_Liver.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\Hill_CV_Unrest_BMRl_CytC_Liver.plt
Mon Nov 23 12:51:51 2009
TBARs, liver only (Table 2)
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
0.004972
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
intercept
0 Specified
0.146
0.298
18
10285.5
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
-2 . 4e-008
4 . 5e-008
-5.6e-008
2 . 4e-008
intercept
-2 . 4e-008
1
-0. 61
0.52
0. 098
v
4.5e-008
-0. 61
1
-0. 83
0.6
n
-5.6e-008
0.52
-0. 83
1
-0.48
k
2 . 4e-008
0. 098
0.6
-0.48
1
Parameter Estimates
Variable
alpha
intercept
Estimate
0.00421237
0.159647
0.303055
2.91267
5344.44
Std. Err.
0.000992864
0.0202575
0.0583218
1.3832
923.736
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
0.00226639
0.119943
0.188746
0.201656
3533.95
0. 00615834
0.199351
0. 417363
5.62369
7154.93
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
6
0
146
0.16
0.0661
0
0649
-0.515
1068
6
0
177
0.162
0.0539
0
0649
0.551
2542
6
0
191
0.191
0.0563
0
0649
0.00541
4489
6
0
271
0.273
0.0563
0
0649
-0.0938
7718
6
0
388
0.385
0.0637
0
0649
0.101
1.396e+004
0.444
0.445
0.11
0.0649
-0.0481
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 80.752584 7 -147.505168
A2 83.373547 12 -142.747094
A3 80.752584 7 -147.505168
fitted 80.455153 5 -150.910305
R 55.820023 2 -107.640047
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 55.107 10 <.0001
Test 2 5.24193 5 0.3871
Test 3 5.24193 5 0.3871
Test 4 0.594862 2 0.7427
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 3420.29
BMDL = 1757.33
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
H.2.1.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
0.5
0.4
0.3
0.2
0.1
Power
BMDL
2000
4000
6000 8000
dose
10000
12000
14000
12:51 11/23 2009
H.2.1.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Pwr_CV_Unrest_BMRl_CytC_Liver.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\Pwr_CV_Unrest_BMRl_CytC_Liver.plt
Mon Nov 23 12:51:51 2009
TBARs, liver only (Table 2)
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-00£
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
0.004972
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho = 0
control = 0.146
slope = 2.56594e-005
power = 0.980719
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
control
slope
power
alpha
1
4.5e-009
-1. 5e-009
1.3e-009
control
4.5e-009
1
-0.71
0. 68
slope
-1.5e-009
-0.71
1
-1
power
1.3e-009
0. 68
-1
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
0. 00474021
0.13485
0. 000217707
0.766684
Std. Err.
0. 00111728
0.0248098
0.000332694
0.157896
Lower Conf. Limit
0. 00255039
0.0862235
-0.000434363
0. 457213
Upper Conf. Limit
0.00693004
0.183476
0. 000869776
1. 07615
Table of Data and Estimated Values of Interest
Dose N
Obs Mean
Est Mean
Obs
Std
Dev
Est
Std
Dev
Scaled Res
0 6
0
146
0
135
0.
0661
0.
0688
0
.397
1068 6
0
177
0
181
0.
0539
0.
0688
-0
. 126
2542 6
0
191
0
224
0.
0563
0.
0688
-
1.16
4489 6
0
271
0
272
0.
0563
0.
0688
-0.
0436
7718 6
0
388
0
343
0.
0637
0.
0688
1. 6
.3 9 6e + 0 0 4
6
0.444
0. 463
0.11
0
0688
-0.
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Model
Likelihoods of Interest
Log(likelihood) # Param's
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
A1
80.
.752584
7
-147 .
. 505168
A2
83.
. 373547
12
-142 .
.747094
A3
80.
.752584
7
-147 .
. 505168
fitted
78 .
. 330124
4
-148 .
. 660249
R
55.
. 820023
2
-107 .
. 640047
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 55.107 10 <.0001
Test 2 5.24193 5 0.3871
Test 3 5.24193 5 0.3871
Test 4 4.84492 3 0.1835
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1823.19
BMDL = 743.833
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-16 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.2. Hassoun et al. (2000): DNA SSB
H.2.2.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
p-Value a
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
0.75
39.08
<0.0001
112.91
4.1E+03
2.5E+03
nonconstant variance,
power restricted >1
exponential (M3)
4
0.75
39.08
<0.0001
112.91
4.1E+03
2.5E+03
nonconstant variance,
power restricted >1
exponential (M4)
3
0.75
4.29
0.23
80.12
5.6E+02
3.6E+02
nonconstant variance,
power restricted >1
exponential (M5)
3
0.75
4.29
0.23
80.12
5.6E+02
3.6E+02
nonconstant variance,
power restricted >1
exponential (M5)
3
0.75
4.29
0.23
80.12
5.6E+02
3.6E+02
nonconstant variance,
power unrestricted
Hill
3
0.75
4.16
0.25
79.99
5.0E+02
3.9E+02
nonconstant variance, n
restricted >1, bound hit
Hill
2
0.75
3.82
0.15
81.65
3.7E+02
1.3E+02
nonconstant variance, n
unrestricted
linear
4
0.75
25.42
<.0001
99.26
1.6E+03
9.6E+02
nonconstant variance
polynomial
4
0.75
25.42
<.0001
99.26
1.6E+03
9.6E+02
nonconstant variance
power
4
0.75
25.42
<.0001
99.26
1.6E+03
9.6E+02
nonconstant variance,
power restricted >1,
bound hit
power
3
0.75
5.14
0.16
80.98
1.9E+02
7.4E+01
nonconstant variance,
power unrestricted
exponential (M2)
4
0.75
38.85
<0.0001
111.13
3.6E+03
3.0E+03
constant variance,
power restricted >1
exponential (M3)
4
0.75
38.85
<0.0001
111.13
3.6E+03
3.0E+03
constant variance,
power restricted >1
exponential (M4)c
3
0.75
4.30
0.23
78.59
6.6E+02
5.0E+02
constant variance,
power restricted >1
exponential (M5)
3
0.75
4.30
0.23
78.59
6.6E+02
5.0E+02
constant variance,
power restricted >1
exponential (M5)d
3
0.75
4.30
0.23
78.59
6.6E+02
5.0E+02
constant variance,
power unrestricted
Hill
3
0.75
4.31
0.23
78.59
6.0E+02
4.4E+02
constant variance, n
restricted >1, bound hit
Hilld
2
0.75
4.09
0.13
80.38
4.8E+02
1.5E+02
constant variance, n
unrestricted
linear
4
0.75
25.33
<.0001
97.62
2.0E+03
1.6E+03
constant variance
polynomial
4
0.75
25.33
<.0001
97.62
2.0E+03
1.6E+03
constant variance
power
4
0.75
25.33
<.0001
97.62
2.0E+03
1.6E+03
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
p-Value a
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power d
3
0.75
5.61
0.13
79.89
2.5E+02
1.1E+02
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.2.2.2. Figure for Selected Model: Exponential (M4), Constant Variance, Power Restricted
>1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
25
20
15
10
5
BMDL
BMD
0
2000
4000
6000
8000
10000
12000
14000
dose
12:50 11/23 2009
H.2.2.3. Output File For Selected Model: Exponential (M4), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Exp_CV_BMRl_DNA_SSB.(d)
Gnuplot Plotting File:
Mon Nov 23 12:50:08 2009
DNA single-strand breaks, liver only (Table 3)
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
Specified
lnalpha 0.841244
rho(S) 0
a 7.0395
b 0.000187891
c 3.50522
d 1
Parameter Estimates
Variable Model 4
lnalpha 0.960792
rho 0
a 7.75279
b 0.000136903
c 3.39666
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 6 7.41 1.543
1068 6 10.78 1.249
2542 6 13.6 1.69
4489 6 15.3 1.715
7718 6 20.4 2.254
1.396e+004 6 23.5 1.372
Estimated Values of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dose
Est Mean
Est Std
Scaled Residual
0
1068
2542
4489
7718
1.396e+004
7 .753
10.28
13.21
16.28
19. 87
23.59
1. 617
1. 617
1 . 617
1 . 617
1 . 617
1 . 617
-0.5194
0.7575
0.5853
-1.49
0.7958
-0.1293
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -33.14239 7 80.28478
A2 -31.81197 12 87.62394
A3 -33.14239 7 80.28478
R -80.44209 2 164.8842
4 -35.29426 4 78.58852
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
97 .26
2 . 661
2 . 661
4 . 304
p-value
10
5
5
3
< 0.0001
0.7521
0.7521
0.2305
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-20 DRAFT—DO NOT CITE OR QUOTE
-------
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 664.925
BMDL = 504.974
H.2.2.4. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
25
20
15
10
5
BMDL
BMD
0
2000
4000
6000
8000
10000
12000
14000
dose
12:50 11/23 2009
H.2.2.5. Output File for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Exp_CV_Unrest_BMRl_DNA_SSB.(d)
Gnuplot Plotting File:
Mon Nov 23 12:50:16 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-21 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
DNA single-strand breaks, liver only (Table 3)
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
Specified
lnalpha 0.841244
rho(S) 0
a 7.0395
b 0.000187891
c 3.50522
d 1
Parameter Estimates
Variable Model 5
lnalpha 0.960792
rho 0
a 7.75279
b 0.000136903
c 3.39666
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 6 7.41 1.543
1068 6 10.78 1.249
2542 6 13.6 1.69
4489 6 15.3 1.715
7718 6 20.4 2.254
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-22 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
1.3 9 6e + 0 0 4 6
23.5
1. 372
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0
7 .753
1. 617
-0.5194
1068
10.28
1. 617
0.7575
2542
13.21
1 . 617
0.5853
4489
16.28
1 . 617
-1.49
7718
19. 87
1 . 617
0.7958
3 9 6e + 0 0 4
23.59
1 . 617
-0.1293
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -33.14239 7 80.28478
A2 -31.81197 12 87.62394
A3 -33.14239 7 80.28478
R -80.44209 2 164.8842
5 -35.29426 4 78.58852
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
97 .26
2 . 661
2 . 661
4 . 304
D. F.
10
5
5
3
p-value
< 0.0001
0.7521
0.7521
0.2305
The p-value for Test 1 is less than .05. There appears to be
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-23 DRAFT—DO NOT CITE OR QUOTE
-------
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 664.925
BMDL = 504.974
H.2.2.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
Hill
25
20
15
10
5
ISMDL
BMD
0
2000
4000
6000
8000
10000
12000
14000
dose
12:50 11/23 2009
H.2.2.7. Output File For Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Hill_CV_Unrest_BMRl_DNA_SSB.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\Hill_CV_Unrest_BMRl_DNA_SSB.plt
Mon Nov 23 12:50:18 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
DNA single-strand breaks, liver only (Table 3)
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
rho
intercept
2 .7831
0
7 .41
16.09
0.235041
10849.8
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
-2 . 8e-008
7 . 4e-008
-6.3e-008
7 . 2e-008
intercept
-2 . 8e-008
1
-0.34
0.47
-0.28
v
7 . 4e-008
-0.34
1
-0. 95
1
n
. 3e-008
0.47
-0. 95
1
-0. 95
k
7 . 2e-008
-0.28
1
-0. 95
1
Parameter Estimates
Variable
alpha
intercept
Estimate
2 .59837
7 .4823
32 . 65
0.876148
14136.3
Std. Err.
0.612441
0.666037
17.9338
0.260495
16730.5
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
1.398
6.17689
-2.49963
0.365588
-18654.9
3.79873
8 .78771
67.7 997
1.38671
46927 . 4
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
7 .41
7.48
1. 54
1. 61
-0.11
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
1068 6 10.8 10.6
2542 6 13.6 13.4
4489 6 15.3 16.2
7718 6 20.4 19.6
1.396e+004 6 23.5 23.7
1.25 1.61 0.336
1.69 1.61 0.27
1.71 1.61 -1.41
2.25 1.61 1.25
1.37 1.61 -0.331
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'"2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -33.142389 7 80.284779
A2 -31.811970 12 87.623940
A3 -33.142389 7 80.284779
fitted -35.187895 5 80.375790
R -80.442086 2 164.884172
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
97.2602
2.66084
2.66084
4.09101
10
5
5
2
<.0001
0.7521
0.7521
0.1293
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-26 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 483.289
BMDL = 153.67 8
H.2.2.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
¦ ¦
Power
25
20
15
10
E
12:50 11/23 2009
2000
4000
6000 8000
dose
10000 12000
14000
H.2.2.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Pwr_CV_Unrest_BMRl_DNA_SSB.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\Pwr_CV_Unrest_BMRl_DNA_SSB.plt
Mon Nov 23 12:50:18 2009
DNA single-strand breaks, liver only (Table 3)
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 2.7831
rho = 0 Specified
control = 7.41
slope = 0.0433022
power = 0.620052
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha control slope power
alpha 1 2.5e-009 -5.4e-009 5.7e-009
control 2.5e-009 1 -0.71 0.66
slope -5.4e-009 -0.71 1 -1
power 5.7e-009 0.66 -1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
2 .71023
7 .26415
0.0685886
0.575949
Std. Err.
0.638807
0.64416
0.0392449
0.0589672
Lower Conf. Limit
1.45819
6.00162
-0.00833
0.460375
Upper Conf. Limit
3.96226
8 . 52668
0.145507
0.691523
Table of Data and Estimated Values of Interest
2 N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
6
7 .41
7 .26
i—1
CJi
1. 65
0.217
1068
6
i—1
o
CD
11.1
1. 25
1. 65
-0.433
2542
6
13. 6
13.5
1.69
1. 65
0. 094
4489
6
15.3
16
1.71
1. 65
-0.993
7718
6
20.4
19.2
2 . 25
1. 65
i—1
CO
CJI
1.396e+004 6 23.5 24 1.37 1.65 -0.735
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -33.142389 7 80.284779
A2 -31.811970 12 87.623940
A3 -33.142389 7 80.284779
fitted -35.946581 4 79.893162
R -80.442086 2 164.884172
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
97 .2602
2.66084
2.66084
5.60838
10
5
5
3
<.0001
0.7521
0.7521
0.1323
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD =24 9.162
BMDL = 111.676
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-29 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.3. Hassoun et al. (2000): TBARs Liver
H.2.3.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
X 2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
0.33
17.56
0.00
-7.04
7.4E+03
3.9E+03
nonconstant variance,
power restricted >1
exponential (M3)
4
0.33
17.56
0.00
-7.04
7.4E+03
3.9E+03
nonconstant variance,
power restricted >1
exponential (M4)
3
0.33
4.35
0.23
-18.26
1.0E+03
5.1E+02
nonconstant variance,
power restricted >1
exponential (M5)
2
0.33
2.78
0.25
-17.82
1.6E+03
6.5E+02
nonconstant variance,
power restricted >1
exponential (M5)
2
0.33
2.78
0.25
-17.82
1.6E+03
6.5E+02
nonconstant variance,
power unrestricted
Hill
2
0.33
2.52
0.28
-18.09
1.7E+03
8.1E+02
nonconstant variance,
n restricted >1
Hill
2
0.33
2.52
0.28
-18.09
1.7E+03
8.1E+02
nonconstant variance,
n unrestricted
linear
4
0.33
15.72
0.00
OO
OO
OO
I
5.1E+03
2.4E+03
nonconstant variance
polynomial
4
0.33
15.72
0.00
OO
OO
OO
1
5.1E+03
2.4E+03
nonconstant variance
power
4
0.33
15.72
0.00
OO
OO
OO
1
5.1E+03
2.4E+03
nonconstant variance,
power restricted >1,
bound hit
power
3
0.33
8.40
0.04
-14.21
5.3E+02
8.3E+00
nonconstant variance,
power unrestricted
exponential (M2)
4
0.33
18.02
0.00
-8.52
9.6E+03
6.7E+03
constant variance,
power restricted >1
exponential (M3)
4
0.33
18.02
0.00
-8.52
9.6E+03
6.7E+03
constant variance,
power restricted >1
exponential (M4)
3
0.33
4.79
0.19
-19.75
1.2E+03
6.3E+02
constant variance,
power restricted >1
exponential (M5)
2
0.33
2.86
0.24
-19.68
1.9E+03
8.4E+02
constant variance,
power restricted >1
exponential (M5)d
2
0.33
2.86
0.24
-19.68
1.9E+03
8.4E+02
constant variance,
power unrestricted
Hillc
2
0.33
2.60
0.27
-19.93
1.8E+03
9.6E+02
constant variance, n
restricted >1
Hilld
2
0.33
2.60
0.27
-19.93
1.8E+03
9.6E+02
constant variance, n
unrestricted
linear
4
0.33
16.75
0.00
-9.79
8.0E+03
5.3E+03
constant variance
polynomial
4
0.33
16.75
0.00
-9.79
8.0E+03
5.3E+03
constant variance
power
4
0.33
16.75
0.00
-9.79
8.0E+03
5.3E+03
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-30 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
X 2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power d
3
0.33
9.75
0.02
-14.79
1.0E+03
5.7E+01
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.2.3.2. Figure for Selected Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
3
2.5
2
.5
1
BMDL
BMD
0
2000
4000
6000
8000
10000
12000
14000
dose
12:51 11/23 2009
H.2.3.3. Output File for Selected Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Exp_CV_Unrest_BMRl_TBARs_Liver.(d)
Gnuplot Plotting File:
Mon Nov 23 12:51:02 2009
TBARs, liver only (Table 2)
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-31 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
Specified
lnalpha -1.90388
rho(S) 0
a 1.39555
b 0.000142164
c 1.97051
d 1
Parameter Estimates
Variable Model 5
lnalpha -1.82448
rho 0
a 1.46526
b 0.000431089
c 1.63651
d 2.96871
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 6 1.469 0.2915
1068 6 1.549 0.5389
2542 6 2.15 0.3625
4489 6 2.28 0.2474
7718 6 2.619 0.5168
1.396e+004 6 2.292 0.4874
Estimated Values of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-32 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dose
Est Mean
Est Std
Scaled Residual
0
1068
2542
4489
7718
1.396e+004
1. 465
1. 554
2 .147
2 .397
2 .398
2 .398
0.4016
0.4016
0.4016
0.4016
0.4016
0.4016
0.0228
-0.03039
0.01965
-0.7145
1.348
-0.646
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 16.26977 7 -18.53954
A2 19.12783 12 -14.25565
A3 16.26977 7 -18.53954
R 2.44294 2 -0.8858799
5 14.84065 5 -19.6813
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
33.37
5.716
5.716
2 . 858
p-value
10
5
5
2
0. 000236
0.3348
0.3348
0.2395
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-33 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1911.82
BMDL = 8 4 0.4 87
H.2.3.4. Figure for Unrestricted Model: Hill, Constant Variance, n Restricted >1
Hill Model with 0.95 Confidence Level
Hill
3
2.5
2
.5
1
BMDL
BMD
0
2000
4000
6000
8000
10000
12000
14000
dose
12:50 11/23 2009
H.2.3.5. Output File for Unrestricted Model: Hill, Constant Variance, n Restricted >1
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\E.MDS21\Nov23\Elood\Hill_CV_E.MRl_TEARs_Liver. (d)
Gnuplot Plotting File : C : \USEPA\E.MDS21\Nov23\Elood\Hill_CV_E.MRl_TEARs_Liver . pit
Mon Nov 2 3 12:50:58 2 00 9
TEARs, liver only (Table 2)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-34 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
rho
intercept
0.178788
0
1.469
1.15
1. 27851
2801.9
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
1. le-007
-1. 4e-007
1. 2e-007
-7.6e-009
intercept
1. le-007
1
-0. 82
0.48
0.52
v
-1. 4e-007
-0. 82
1
-0. 61
-0.22
n
1. 2e-007
0.48
-0. 61
1
0.29
k
-7 . 6e-009
0.52
-0.22
0.29
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
intercept
Estimate
0.16017
1.46138
0.963032
3.44649
2002 . 3
Std. Err.
0.0377524
0.152797
0.20228
2.43475
562.074
Lower Conf. Limit
0.0861764
1.1619
0.56657
-1. 32553
900.655
Upper Conf. Limit
0.234163
1.76086
1.35949
8.21851
3103.95
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
6
1.47
1.46
0.291
0
4
0.0466
1068
6
1. 55
1. 56
0.539
0
4
-0.0697
2542
6
2 .15
2 .13
0.363
0
4
0.12
4489
6
2 .28
2 . 37
0.247
0
4
o
CJi
7718
6
2 . 62
2 .42
0.517
0
4
1. 25
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-35 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
1.396e+004 6 2.29 2.42 0.487 0.4 -0.803
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 16.269770 7 -18.539539
A2 19.127827 12 -14.255654
A3 16.269770 7 -18.539539
fitted 14.967385 5 -19.934770
R 2.442940 2 -0.885880
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels:
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
33.3698
5.71611
5.71611
2.60477
10
5
5
2
0.000236
0.3348
0.3348
0.2719
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-36 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Confidence level =
BMD =
BMDL =
0. 95
1813.69
957.252
H.2.3.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
2.5
1.5
Hill
BMDL
BMD
2000
4000
6000 8000
dose
10000
12000
14000
12:51 11/23 2009
H.2.3.7. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Hill_CV_Unrest_BMRl_TBARs_Liver.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\Hill_CV_Unrest_BMRl_TBARs_Liver.plt
Mon Nov 23 12:51:04 2009
TBARs, liver only (Table 2)
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-37 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
rho
intercept
0.178788
0
1.469
1.15
1. 27851
2801.9
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
1. le-007
-1. 4e-007
1. 2e-007
-7 . 6e-009
intercept
1. le-007
1
-0. 82
0.48
0.52
v
-1. 4e-007
-0. 82
1
-0. 61
-0.22
n
1. 2e-007
0.48
-0. 61
1
0.29
k
-7 . 6e-009
0.52
-0.22
0.29
1
Parameter Estimates
Variable
alpha
intercept
Estimate
0.16017
1.46138
0.963032
3.44649
2002 . 3
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
0.0377524 0.0861764 0.234163
0.152797 1.1619 1.76086
0.20228 0.56657 1.35949
2.43475 -1.32553 8.21851
562.074 900.655 3103.95
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
6
1.47
1.46
0.291
0
4
0.0466
1068
6
1. 55
1. 56
0.539
0
4
-0.0697
2542
6
2 .15
2 .13
0.363
0
4
0.12
4489
6
2 .28
2 . 37
0.247
0
4
-0.54
7718
6
2 . 62
2 .42
0.517
0
4
1. 25
1. 396e + 004
2.29
2 .42
0. 487
0.4
-0.803
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-38 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model
A1
A2
A3
fitted
R
Log(likelihood)
16.269770
19.127827
16.269770
14.967385
2.442940
Param'
7
12
7
5
2
AIC
-18.539539
-14 . 255654
-18.539539
-19.934770
-0.885880
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
33.3698
5.71611
5.71611
2.60477
10
5
5
2
0.000236
0.3348
0.3348
0.2719
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1813.69
BMDL = 957.252
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-39 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
H.2.3.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
3
2.5
2
1.5
1
BMD
0
2000
4000
6000
8000
10000
12000
14000
dose
12:51 11/23 2009
H.2.3.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Blood\Pwr_CV_Unrest_BMRl_TBARs_Liver.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Blood\Pwr_CV_Unrest_BMRl_TBARs_Liver.plt
Mon Nov 23 12:51:05 2009
TBARs, liver only (Table 2)
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 0.178788
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-40 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho = 0
control = 1.469
slope = 0.000328724
power = 0.885141
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
control
slope
power
alpha
1
3.9e-011
-2.9e-010
3.6e-010
control
3.9e-011
1
-0.59
0.47
slope
-2.9e-010
-0.59
1
-0. 99
power
3.6e-010
0.47
-0. 99
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
0.195332
1.42145
0. 0382805
0.353387
Std. Err.
0. 0460403
0.17171
0. 0492936
0.132966
Lower Conf. Limit
0.105095
1.0849
-0.0583331
0.092777 9
Upper Conf. Limit
0.28557
1.75799
0.134894
0.613996
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 i
1068 e
2542 (
4489 e
7718 £
1.396e+004
1.47
1. 55
2 .15
2 .28
2 . 62
42
87
2 . 03
2 .17
2 . 33
0.291
0.539
0.363
0.247
0.517
0.442
0.442
0.442
0.442
0.442
0.264
-1.79
0. 649
0. 616
1. 62
2.29
2 . 54
0. 487
0.442
-1. 36
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model
Log(likelihood)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-41 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
A1
16.269770
7
-18.539539
A2
19.127827
12
-14 . 255654
A3
16.269770
7
-18.539539
fitted
11.394946
4
-14.789892
R
2.442940
2
-0.885880
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 33.3698 10 0.000236
Test 2 5.71611 5 0.3348
Test 3 5.71611 5 0.3348
Test 4 9.74965 3 0.02082
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1014.75
BMDL = 5 6.7719
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-42 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.4. Kitchin et al. (1979): BaP Hydrolase Activity
H.2.4.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x 2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
9
<0.0001
247.10
<0.0001
452.74
9.3E+05
4.3E+05
nonconstant variance,
power restricted >1
exponential (M3)
9
<0.0001
247.10
<0.0001
452.74
9.3E+05
4.3E+05
nonconstant variance,
power restricted >1
exponential (M4)
8
<0.0001
18.95
0.02
226.59
6.3E+02
4.8E+02
nonconstant variance,
power restricted >1
exponential (M5)c
7
<0.0001
16.76
0.02
226.41
1.2E+03
5.6E+02
nonconstant variance,
power restricted >1
exponential (M5)d
7
<0.0001
16.76
0.02
226.41
1.2E+03
5.6E+02
nonconstant variance,
power unrestricted
Hill
7
<.0001
296.88
<.0001
506.53
error
error
nonconstant variance, n
restricted >1
Hilld
7
<.0001
296.88
<.0001
506.53
error
error
nonconstant variance, n
unrestricted
linear
9
<.0001
94.23
<.0001
299.87
9.6E+02
6.9E+02
nonconstant variance
polynomial
9
<.0001
-197.64
<.0001
8.00
error
error
nonconstant variance
power
9
<.0001
94.23
<.0001
299.87
9.6E+02
6.9E+02
nonconstant variance,
power restricted >1,
bound hit
power d
8
<.0001
63.63
<.0001
271.27
1.0E+02
4.4E+01
nonconstant variance,
power unrestricted
exponential (M2)
9
<0.0001
129.40
<0.0001
451.64
1.3E+06
1.1E+06
constant variance,
power restricted >1
exponential (M3)
9
<0.0001
129.40
<0.0001
451.64
1.3E+06
1.1E+06
constant variance,
power restricted >1
exponential (M4)
8
<0.0001
6.96
0.54
331.23
9.5E+03
7.3E+03
constant variance,
power restricted >1
exponential (M5)
8
<0.0001
6.96
0.54
331.23
9.5E+03
7.3E+03
constant variance,
power restricted >1
exponential (M5)
8
<0.0001
6.96
0.54
331.23
9.5E+03
7.3E+03
constant variance,
power unrestricted
Hill
7
<.0001
35.69
<.0001
361.95
6.3E+04
2.6E+03
constant variance, n
restricted >1
Hill
7
<.0001
35.69
<.0001
361.95
6.3E+04
3.0E+02
constant variance, n
unrestricted
linear
9
<.0001
120.38
<.0001
442.64
6.6E+05
5.1E+05
constant variance
polynomial
9
<.0001
120.38
<.0001
442.64
6.6E+05
5.1E+05
constant variance
power
9
<.0001
120.38
<.0001
442.64
6.6E+05
5.1E+05
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-43 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
x 2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
8
<.0001
51.09
<.0001
375.35
4.1E+02
8.6E+01
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.2.4.2. Figure for Selected Model: Exponential (M5), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 5 with 0.95 Confidence Level
250
200
150
100
Exponential
S S/IDLBMD
500000
1e+006
1,5e+006
2e+006
dose
13:25 11/20 2009
H.2.4.3. Output File for Selected Model: Exponential (M5), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_BMRl_BaP_hydro_act.(d)
Gnuplot Plotting File:
Fri Nov 20 13:25:02 2009
Kitchin 1979, Tbl3, BaP hydrolase activity
The form of the response function by Model:
Model 2: Y[dose] = a * exp{sign * b * dose}
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-44 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 3: Y[dose] = a * exp{sign * (b * dosej^d}
Model 4: Y[dose] = a * [c-(c-l) * exp{-b * dose}]
Model 5: Y[dose] = a * [c-(c-l) * exp{-(b * doseJ'M}]
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 11
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -3.27793
rho 1.92227
a 4.655
b 1.52141e-006
c 42.6316
d 1
Parameter Estimates
Variable Model 5
lnalpha
rho
-2.64351
1.93772
5.43367
1.65224e-005
31.204
1.21424
Table of Stats From Input Data
Dose
Obs Mean
Obs Std Dev
0
9
4 . 9
1.11
69.5
4
4 . 9
1.18
232
4
6.7
1. 4
463
4
7 . 2
1. 8
2318
4
8 . 3
0.26
6 9 4 9
4
14
5
2.319e+004
6.966e+004
2.326e+005
5.819e+005
2.332e+006
59
96
155
182
189
46
16.4
26
26
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-45 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dose
Estimated Values of Interest
Est Mean Est Std Scaled Residual
0
5. 434
1.
375
-1.165
69.5
5.478
1.
385
-0.8342
232
5. 625
1.
421
1. 513
463
5. 875
1.
483
1.787
2318
8 . 529
2 .
127
-0.2151
6 9 4 9
16. 87
4 .
119
-1.392
319e+004
49.41
11
. 67
1.644
966e+004
119. 4
27
.44
-1.708
326e+005
168 . 6
38
. 32
-0.7087
819e+005
169.6
38
. 53
0.6461
332e+006
169.6
38
. 53
1. 009
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -158.1306 12 340.2613
A2 -84.80028 22 213.6006
A3 -98.82189 13 223.6438
R -234.6252 2 473.2504
5 -107.2031 6 226.4062
Additive constant for all log-likelihoods = -45.03. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
299.6
146.7
28 . 04
16.76
20
10
9
7
p-value
< 0.0001
< 0.0001
0.0009381
0.01899
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-46 DRAFT—DO NOT CITE OR QUOTE
-------
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1182.79
BMDL = 556.132
H.2.4.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
250
200
150
100
Exponential
S S/IDLBMD
500000
1e+006 1.5e+006
dose
2e+006
13:25 11/20 2009
H.2.4.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-47 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_Unrest_BMRl_BaP_hydro_act.(d)
Gnuplot Plotting File:
Fri Nov 20 13:25:17 2009
Kitchin 1979, Tbl3, BaP hydrolase activity
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'""d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * doseJ'M}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 11
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -3.27793
rho 1.92227
a 4.655
b 1.52141e-006
c 42.6316
d 1
Parameter Estimates
Variable Model 5
lnalpha -2.64351
rho 1.93772
a 5.43367
b 1. 65224e-005
c 31.204
d 1.21424
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 9 4.9 1.11
69.5 4 4.9 1.18
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
232
463
2318
6 9 4 9
2.319e+004
6.966e+004
2.326e+005
5.819e+005
2.332e+006
6.7
7 . 2
8 . 3
14
59
96
155
182
189
1. 4
1. 8
0.26
5
46
16.4
26
26
Estimated Values of Interest
Est Mean Est Std Scaled Residual
0
5. 434
1.
375
-1.165
69.5
5.478
1.
385
-0.8342
232
5. 625
1.
421
1. 513
463
5. 875
1.
483
1.787
2318
8 . 529
2 .
127
-0.2151
6 9 4 9
16. 87
4 .
119
-1.392
319e+004
49.41
11
. 67
1.644
966e+004
119. 4
27
.44
-1.708
326e+005
168 . 6
38
. 32
-0.7087
819e+005
169.6
38
. 53
0.6461
332e+006
169.6
38
. 53
1. 009
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -158.1306 12 340.2613
A2 -84.80028 22 213.6006
A3 -98.82189 13 223.6438
R -234.6252 2 473.2504
5 -107.2031 6 226.4062
Additive constant for all log-likelihoods = -45.03. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-49 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
-2*log(Likelihood Ratio) D. F. p-value
Test 1 299.6 20 < 0.0001
Test 2 146.7 10 < 0.0001
Test 3 28.04 9 0.0009381
Test 7a 16.76 7 0.01899
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is less than .1. Model 5 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1182.79
BMDL = 556.132
H.2.4.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
250
200
150
Q)
w
c
o
Q.
CO
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
H.2.4.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Hill_Unrest_BMRl_BaP_hydro_act.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_BaP_hydro_act.plt
Fri Nov 20 13:25:18 2009
Kitchin 1979, Tbl3, BaP hydrolase activity
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 11
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 5.7 0855
rho = 0
intercept = 4.9
v = 184 .1
n = 18
k = 392820
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v
lalpha NA NA NA NA
rho NA NA NA NA
intercept NA NA 1 NA
v NA NA NA NA
n NA NA 0. 2 8 NA
k NA NA 0.1 NA
NA
NA
0.28
NA
1
-0. 98
NA
NA
0.1
NA
-0. 98
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha 10.1833 NA NA NA
rho 0.0839751 NA NA NA
intercept -2.28069e-006 NA NA NA
v 184.001 NA NA NA
n 17.9976 NA NA NA
k 1.41183e+007 NA NA NA
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-51 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
At least some variance estimates are negative.
THIS USUALLY MEANS THE MODEL HAS NOT CONVERGED!
Try again from another starting point.
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 9
4 . 9
-2 . 28e-006
1.
11
94 . 3
0.156
69.5 4
4 . 9
-2 . 28e-006
1.
18
94 . 3
0.104
232 4
6.7
-2 . 28e-006
1
. 4
94 . 3
0.142
463 4
7 . 2
-2 . 28e-006
1
. 8
94 . 3
0.153
2318 4
8 . 3
-2 . 28e-006
0.
26
94 . 3
0.176
6 9 4 9 4
14
-2 . 28e-006
5
94 . 3
0.297
2.319e+004
4
59 -2.28e-006
6.8
94 . 3
1. 25
6.966e+004
4
96 -2.28e-006
46
94 . 3
2 . 04
2.326e+005
4
155 -2.28e-006
16.4
94 . 3
3.29
5.819e+005
4
182 -2.28e-006
26
94 . 3
3.86
2.332e+006
4
189 -2.28e-006
26
94 . 3
4 . 01
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model
Log(likelihood)
# Param's
AIC
A1
-158.130647
12
340.261294
A2
-84 . 800279
22
213.600558
A3
-98.821893
13
223.643786
fitted
-247.263464
6
506.526929
R
-234.625213
2
473.250426
Explanation of Tests
Test 1:
Test 2
Test 3
Test 4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
299.65
146.661
28.0432
20
10
9
<.0001
<.0001
0.0009381
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-52 DRAFT—DO NOT CITE OR QUOTE
-------
Test 4
296.883
7
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1.#QNAN
BMDL computation failed.
H.2.4.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
250
Power
200
150
100
0
500000
1e+006
1,5e+006
2e+006
dose
13:25 11/20 2009
H.2.4.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-53 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_BaP_hydro_act.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Pwr_Unrest_BMRl_BaP_hydro_act.plt
Fri Nov 20 13:25:19 2009
Kitchin 1979, Tbl3, BaP hydrolase activity
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 11
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 5.7 0855
rho = 0
control = 4.9
slope = 0.0304965
power = 0.5 937 4 3
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho control slope power
lalpha 1 -0.9 -0.45 0.25 -0.23
rho -0.9 1 0.35 -0.18 0.12
control -0.45 0.35 1 -0.44 0.42
slope 0.25 -0.18 -0.44 1 -0.98
power -0.23 0.12 0.42 -0.98 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha -3.42042 0.570798 -4.53917 -2.30168
rho 2.42941 0.164247 2.10749 2.75133
control 4.52558 0.315791 3.90665 5.14452
slope 0.0642178 0.0318952 0.00170434 0.126731
power 0.619697 0.0482017 0.525223 0.71417
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 9 4.9 4.53 1.11 1.13 0.993
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-54 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
69.5 4
232 4
463 4
2318 4
6 9 4 9 4
2.319e+004
6.966e+004
2.326e+005
5.819e+005
2.332e+006
4 . 9
6.7
7 . 2
8 . 3
14
59
96
155
182
189
5.42
6.4
7 .41
12 . 3
20
37 .1
68 . 9
140
244
572
1.18
1. 4
1. 8
0.26
5
46
16.4
26
26
1.41
1.72
2 . 06
3. 83
6.86
-0.732
0.344
-0.2
-2 .
-1.
11
74
14 . 6
30. 9
73.4
144
404
3. 01
1.75
0.397
-0.868
-1.89
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma^2
Likelihoods of Interest
Model
Log(likelihood)
# Param's
AIC
A1
-158.130647
12
340.261294
A2
-84 . 800279
22
213.600558
A3
-98.821893
13
223.643786
fitted
-130.634662
5
271. 269325
R
-234.625213
2
473.250426
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
299.65
146.661
28.0432
63.6255
20
10
9
<.0001
<.0001
0.0009381
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-55 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 102.508
BMDL = 44 . 1703
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-56 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.5. National Toxicology Program. (2006): EROD Liver Week 53
H.2.5.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
<0.0001
113.40
<0.0001
203.18
7.6E+03
5.1E+03
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
113.40
<0.0001
203.18
7.6E+03
5.1E+03
nonconstant variance,
power restricted >1
exponential (M4)
3
<0.0001
20.95
0.00
112.76
1.2E+02
8.2E+01
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
6.69
0.04
100.50
3.1E+02
2.0E+02
nonconstant variance,
power restricted >1
Hillc
2
<0001
3.09
0.21
96.90
4.2E+02
3.0E+02
nonconstant variance,
n restricted >1
Hilld
2
<.0001
3.09
0.21
96.90
4.2E+02
3.0E+02
nonconstant variance, n
unrestricted
linear
4
<.0001
73.05
<.0001
162.86
1.5E+02
1.1E+02
nonconstant variance
polynomial
4
<.0001
-81.81
<.0001
8.00
6.0E-08
error
nonconstant variance
power
4
<.0001
73.05
<.0001
162.86
1.5E+02
1.1E+02
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
4
<0.0001
77.17
<0.0001
201.35
7.0E+03
5.9E+03
constant variance,
power restricted >1
exponential (M3)
4
<0.0001
77.17
<0.0001
201.35
7.0E+03
5.9E+03
constant variance,
power restricted >1
exponential (M4)
3
<0.0001
7.36
0.06
133.54
4.6E+02
3.6E+02
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
4.20
0.12
132.37
7.3E+02
4.5E+02
constant variance,
power restricted >1
Hill
2
<.0001
2.31
0.31
130.49
9.1E+02
6.1E+02
constant variance, n
restricted >1
Hill
2
<.0001
2.31
0.31
130.49
9.1E+02
6.1E+02
constant variance, n
unrestricted
linear
4
<.0001
64.69
<.0001
188.86
4.0E+03
3.2E+03
constant variance
polynomial
4
<.0001
64.69
<.0001
188.86
4.0E+03
3.2E+03
constant variance
power
4
<.0001
64.69
<.0001
188.86
4.0E+03
3.2E+03
constant variance,
power restricted >1,
bound hit
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAlternate model also presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-57 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
H.2.5.2. Figure for Selected Model: Hill, Nonconstant Variance, n Restricted >1
Hill Model with 0.95 Confidence Level
25
20
15
10
pMDLBMD
10000
12000
14000
16000
11:23 11/19 2009
H.2.5.3. Output file for Selected Model: Hill, Nonconstant Variance, n Restricted >1
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_BMRl_Tbll2_wk53_EROD_liv.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\AD\Blood\Hill_BMRl_Tbll2_wk53_EROD_liv.plt
Thu Nov 19 11:23:04 2009
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter restricted to be greater than 1
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial
lalpha =
rho =
Parameter Values
1.59547
0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-58 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
intercept
3. 614
17 .599
2 . 06282
3589.94
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
-0. 96
-0.16
0. 088
-0. 06
0. 042
rho
-0. 96
1
0.15
-0.12
0. 062
-0.045
intercept
-0.16
0.15
1
-0.18
0.13
0. 073
v
0. 088
-0.12
-0.18
1
-0.7
0. 82
n
-0. 06
0. 062
0.13
-0.7
1
-0.78
k
0. 042
-0.045
0. 073
0. 82
-0.78
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
-4 .86517
2 .26994
3. 62886
17 .8693
2 .12332
2573.21
Std. Err.
0.742003
0.287404
0.133846
0. 946035
0.238762
216.955
Lower Conf. Limit
-6.31947
1.70664
3.36652
16.0151
1.65535
2147.99
Upper Conf. Limit
-3.41087
2.83324
3.89119
19.7235
2 .59128
2998.43
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
1354
3056
5259
8918
1.6e+004
3. 61
7 . 27
14 . 8
17 . 3
20.6
21. 2
3. 63
7 . 27
14 . 2
18 . 3
20.3
21.1
0.486
0.557
1. 61
1.59
3. 05
3. 82
0.379
0. 834
1.78
2 . 38
2 . 68
2 .
-0.111
0. 013
0. 925
-1.19
0.29
0 . 077
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Mu(i) + e(ij ;
Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma^2
Likelihoods of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-59 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
Model Log(likelihood) # Param's AIC
A1 -59.086537 7 132.173073
A2 -37.515858 12 99.031716
A3 -40.906180 8 97.812359
fitted -42.452016 6 96.904031
R -116.710291 2 237.420582
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 158.389 10 <.0001
Test 2 43.1414 5 <.0001
Test 3 6.78064 4 0.1479
Test 4 3.09167 2 0.2131
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 423.477
BMDL = 304.577
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-60 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
H.2.5.4. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
Hill
25
20
15
10
5
BMDL BMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
11:23 11/19 2009
H.2.5.5. Output file for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_Tbll2_wk53_EROD_liv.(d)
Gnuplot Plotting File:
C:\USEPA\BMDS21\AD\Blood\Hill_Unrest_BMRl_Tbll2_wk53_EROD_liv.pit
Thu Nov 19 11:23:13 2009
0
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 1.59547
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-61 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho = 0
intercept = 3.614
v = 17.599
n = 2 . 06282
k = 3589.94
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -0.96 -0.16 0.088 -0.06 0.042
rho -0.96 1 0.15 -0.12 0.062 -0.045
intercept -0.16 0.15 1 -0.18 0.13 0.073
v 0.088 -0.12 -0.18 1 -0.7 0.82
n -0.06 0.062 0.13 -0.7 1 -0.78
k 0.042 -0.045 0.073 0.82 -0.78 1
Parameter Estimates
95.0?
Variable
Estimate
Std. Err.
Lower Conf. Limit
lalpha
-4 .86517
0.742003
-6.31947
rho
2 .26994
0.287404
1.70664
intercept
3. 62886
0.133846
3.36652
V
17 .8693
0. 946034
16.0151
n
2 .12332
0.238762
1.65535
k
2573.21
216.955
2147.99
Wald Confidence Interval
Upper Conf. Limit
-3.41087
2.83324
3.89119
19.7235
2 .59128
2998.43
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 3.61 3.63 0.486 0.379 -0.111
1354 8 7.27 7.27 0.557 0.834 0.013
3056 8 14.8 14.2 1.61 1.78 0.925
5259 8 17.3 18.3 1.59 2.38 -1.19
8918 8 20.6 20.3 3.05 2.68 0.29
1.6e+004 8 21.2 21.1 3.82 2.8 0.077
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-62 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -59.086537 7 132.173073
A2 -37.515858 12 99.031716
A3 -40.906180 8 97.812359
fitted -42.452016 6 96.904032
R -116.710291 2 237.420582
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 158.389 10 <.0001
Test 2 43.1414 5 <.0001
Test 3 6.78064 4 0.1479
Test 4 3.09167 2 0.2131
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 423.477
BMDL = 304.577
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-63 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.6. National Toxicology Program. (2006): Lung EROD Week 31
H.2.6.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
<0.0001
123.80
<0.0001
390.99
1.2E+04
8.1E+03
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
123.80
<0.0001
390.99
1.2E+04
8.1E+03
nonconstant variance,
power restricted >1
exponential (M4)c
3
<0.0001
13.01
0.00
282.20
4.1E+01
2.9E+01
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
13.01
0.00
282.20
4.1E+01
2.9E+01
nonconstant variance,
power restricted >1
exponential (M5)d
3
<0.0001
13.01
0.00
282.20
4.1E+01
2.9E+01
nonconstant variance,
power unrestricted
Hill
4
<.0001
59.49
<.0001
326.67
1.3E-11
1.3E-11
nonconstant variance, n
restricted >1, bound hit
Hilld
4
<.0001
59.49
<.0001
326.67
1.3E-11
1.3E-11
nonconstant variance, n
unrestricted
linear
4
<.0001
116.14
<.0001
383.32
4.1E+03
1.5E+03
nonconstant variance
polynomial
4
<.0001
123.25
<.0001
390.43
1.3E+04
7.5E+01
nonconstant variance
power
4
<.0001
116.14
<.0001
383.32
4.1E+03
1.5E+03
nonconstant variance,
power restricted >1,
bound hit
power d
3
<.0001
23.64
<.0001
292.82
4.5E-02
4.5E-02
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
80.66
<0.0001
390.82
9.0E+03
7.5E+03
constant variance, power
restricted >1
exponential (M3)
4
<0.0001
80.66
<0.0001
390.82
9.0E+03
7.5E+03
constant variance, power
restricted >1
exponential (M4)
3
<0.0001
12.08
0.01
324.24
4.6E+02
3.4E+02
constant variance, power
restricted >1
exponential (M5)
2
<0.0001
12.08
0.00
326.24
4.6E+02
3.4E+02
constant variance, power
restricted >1
exponential (M5)
2
<0.0001
12.08
0.00
326.24
4.6E+02
3.4E+02
constant variance, power
unrestricted
Hill
2
<.0001
14.17
0.00
328.33
5.1E+02
2.4E+02
constant variance, n
restricted >1
Hill
2
<.0001
14.17
0.00
328.33
5.1E+02
1.9E+02
constant variance, n
unrestricted
linear
4
<.0001
71.44
<.0001
381.60
5.6E+03
4.4E+03
constant variance
polynomial
4
<.0001
71.44
<.0001
381.60
5.6E+03
4.4E+03
constant variance
power
4
<.0001
71.44
<.0001
381.60
5.6E+03
4.4E+03
constant variance, power
restricted >1, bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-64 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
3
<.0001
23.14
<.0001
335.30
2.0E+01
2.0E+00
constant variance, power
unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.2.6.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
60
50
40
30
20
10
0
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
13:16 11/20 2009
H.2.6.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_BMRl_Lung_EROD_wk31.(d)
Gnuplot Plotting File:
Fri Nov 20 13:16:38 2009
Tbl 12, Week 31, Lung Microsomes EROD
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-65 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -1.42653
rho 1.46168
a 1.96745
b 0.000226755
c 26.7857
d 1
Parameter Estimates
Variable Model 4
lnalpha -1.47384
rho 1.57432
a 2.11972
b 0.000440068
c 23.6215
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 10 2.071 0.9708
1284 10 25.34 2.549
2932 10 30.39 5.831
5075 10 50.19 8.68
8629 10 49.07 13.91
1.55e + 004 10 48.42 8.933
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-66 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0
2 .
. 12
0.8646
-0.1782
1284
22 .
. 82
5.
612
1.423
2932
36.
.88
8 .
189
-2 . 506
5075
44 .
. 93
9.
567
1.738
8629
49
10
.24
0.02179
1.55e+004
50.
. 02
10
.41
-0.4854
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -152.0793 7 318.1586
A2 -123.367 12 270.734
A3 -129.5911 8 275.1823
R -206.5175 2 417.0349
4 -136.0978 5 282.1956
Additive constant for all log-likelihoods = -55.14. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
166.3
57 .42
12 .45
13. 01
10
5
4
3
p-value
< 0.0001
< 0.0001
0.01431
0.004608
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 6a is less than .1. Model 4 may not adeguately
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-67 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 41.3446
BMDL = 28.8946
H.2.6.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
60
50
40
30
20
10
0
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
13:16 11/20 2009
H.2.6.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_Unrest_BMRl_Lung_EROD_wk31.(d)
Gnuplot Plotting File:
Fri Nov 20 13:16:45 2009
Tbl 12, Week 31, Lung Microsomes EROD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-68 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'""d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)'""d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -1.42653
rho 1.46168
a 1.96745
b 0.000226755
c 26.7857
d 1
Parameter Estimates
Variable Model 5
lnalpha -1.47384
rho 1.57432
a 2.11972
b 0.000440068
c 23.6215
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 10 2.071 0.9708
1284 10 25.34 2.549
2932 10 30.39 5.831
5075 10 50.19 8.68
8629 10 49.07 13.91
1.55e + 004 10 48.42 8.933
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-69 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0 2.12 0.8646 -0.1782
1284 22.82 5.612 1.423
2932 36.88 8.189 -2.506
5075 44.93 9.567 1.738
8629 49 10.24 0.02179
1.55e+004 50.02 10.41 -0.4854
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(1)
Var{e(ij)} = Sigma"2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -152.0793 7 318.1586
A2 -123.367 12 270.734
A3 -129.5911 8 275.1823
R -206.5175 2 417.0349
5 -136.0978 5 282.1956
Additive constant for all log-likelihoods = -55.14. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
166.3
57 .42
12 .45
13. 01
10
5
4
3
p-value
< 0.0001
< 0.0001
0.01431
0.004608
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-70 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 41.3446
BMDL = 28.8946
H.2.6.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
60
50
40
30
20
10
B
Hill Model with 0.95 Confidence Level
Hill
I
i/IDLBMD
0 2000
13:16 11/20 2009
4000
6000
8000
dose
10000
12000
14000
16000
H.2.6.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Hill_Unrest_BMRl_Lung_EROD_wk31.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_Lung_EROD_wk31.plt
Fri Nov 20 13:16:47 2009
Tbl 12, Week 31, Lung Microsomes EROD
The form of the response function is:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-71 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.17467
rho = 0
intercept = 2.071
v = 48.119
n = 18
k = 4322.89
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -n -k
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha rho intercept v
lalpha 1 -0.95 -0.49 0.11
rho -0.95 1 0.45 -0.22
intercept -0.49 0.45 1 -0.15
v 0.11 -0.22 -0.15 1
Parameter Estimates
Variable
lalpha
rho
intercept
k
Estimate
-1.47774
1.8037
2 . 071
38.6102
18
1.5503e-011
Std. Err.
0.642044
0.187436
0.291246
1.9322
NA
NA
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-2 .73613
1.43633
1.50017
34 . 8232
-0.219359
2 .17107
2 . 64183
42 . 3972
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Dose
Table of
N
Data and Estimated Values
Obs Mean Est Mean
of Interest
Obs Std Dev
Est Std Dev
Scaled Res
0
10
2 . 07
2 . 07
0. 971
0. 921
-2 . 27e-007
1284
10
25.3
40.7
2 . 55
13.5
-3.59
2932
10
30. 4
40.7
5. 83
13.5
-2 .41
5075
10
50.2
40.7
CD
<£>
CD
13.5
2 . 23
8629
10
49.1
40.7
13. 9
13.5
1. 96
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-72 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
1.55e+004 10 48.4 40.7 8.93 13.5 1.81
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -152.079318 7 318.158637
A2 -123.366985 12 270.733969
A3 -129.591134 8 275.182269
fitted -159.335928 4 326.671856
R -206.517459 2 417.034919
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
166.301
57 .4247
12.4483
59.4896
10
5
4
4
<.0001
<.0001
0.01431
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-73 DRAFT—DO NOT CITE OR QUOTE
-------
BMD = 1.2 614 3e-011
BMDL = 1.2 614 3e-011
H.2.6.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
S/IDLBMD
10000
12000
14000
16000
13:16 11/20 2009
H.2.6.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_Lung_EROD_wk31.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Pwr_Unrest_BMRl_Lung_EROD_wk31.plt
Fri Nov 20 13:16:48 2009
Tbl 12, Week 31, Lung Microsomes EROD
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i)
exp(lalpha + log(mean(i)
rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-74 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.17467
rho = 0
control = 2.071
slope = 2
. 58483
power = 0.
313845
Asymptotic Correlation Matrix
of Parameter
Estimates
lalpha
rho
control
slope power
lalpha
1
-0. 94
-0.42
0.14 -0.14
rho
-0. 94
1
0.38
-0.17 0.15
control
-0.42
0.38
1
-0.11 0.094
slope
0.14
-0.17
-0.11
1 -1
power
-0.14
0.15
0. 094
-1 1
Parameter Estimates
95.0% Wald Confidence
Interval
Variable
Estimate
Std. Err.
Lower Conf. Limit Upper
Conf. Limit
lalpha
-1.53087
0.573913
-2.65572
-0.406018
rho
1. 64612
0.167167
1.31848
1.97376
control
2 .1082
0.270802
1. 57744
2.63896
slope
2 . 35281
0.945551
0. 499559
4 .20605
power
0.325737
0.0465627
0.234476
0. 416998
Table of Data and
Estimated Values
of Interest
Dose N
Obs Mean Est Mean
Obs Std Dev
Est Std Dev Scaled Res.
0 10
2 . 07
2 .11
0. 971
0.859 -0.137
1284 10
25.3
26.3
2 . 55
6.86 -0.453
2932 10
30. 4
33. 8
5. 83
8.43 -1.28
5075 10
50.2
40
8 . 68
9.69 3.33
8629 10
49.1
47 . 2
13. 9
11.1 0.545
1. 55e + 004
10
48.4
56.
. 93
12 . 9
-2 . 01
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Mu(i) + e(ij ;
Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-75 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -152.079318 7 318.158637
A2 -123.366985 12 270.733969
A3 -129.591134 8 275.182269
fitted -141.409222 5 292.818443
R -206.517459 2 417.034919
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 166.301 10 <.0001
Test 2 57.4247 5 <.0001
Test 3 12.4483 4 0.01431
Test 4 23.6362 3 <.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.0454117
BMDL = 0.0454112
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-76 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.7. National Toxicology Program. (2006): Lung EROD Week 53
H.2.7.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
<0.0001
64.02
<0.0001
314.33
1.8E+04
1.1E+04
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
64.02
<0.0001
314.33
1.8E+04
1.1E+04
nonconstant variance,
power restricted >1
exponential (M4)c
3
<0.0001
3.63
0.30
255.94
5.3E+01
3.3E+01
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
2.58
0.28
256.88
5.8E+02
3.6E+01
nonconstant variance,
power restricted >1
exponential (M5)d
2
<0.0001
2.58
0.28
256.88
5.8E+02
3.6E+01
nonconstant variance,
power unrestricted
Hill
3
<.0001
16.10
0.00
268.40
1.7E-05
1.7E-05
nonconstant variance, n
restricted >1
Hilld
3
<.0001
16.10
0.00
268.40
1.7E-05
1.7E-05
nonconstant variance, n
unrestricted
linear
4
<.0001
62.93
<.0001
313.23
1.5E+04
6.9E+03
nonconstant variance
polynomial
5
<.0001
81.88
<.0001
330.18
error
2.0E+03
nonconstant variance
power
4
<.0001
62.93
<.0001
313.23
1.5E+04
6.9E+03
nonconstant variance,
power restricted >1,
bound hit
power d
3
<.0001
8.76
0.03
261.07
1.1E-04
1.1E-04
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
39.91
<0.0001
316.45
1.1E+04
8.8E+03
constant variance, power
restricted >1
exponential (M3)
4
<0.0001
39.91
<0.0001
316.45
1.1E+04
8.8E+03
constant variance, power
restricted >1
exponential (M4)
3
<0.0001
3.69
0.30
282.22
4.0E+02
2.4E+02
constant variance, power
restricted >1
exponential (M5)
2
<0.0001
2.71
0.26
283.24
1.1E+03
2.7E+02
constant variance, power
restricted >1
exponential (M5)
2
<0.0001
2.71
0.26
283.24
1.1E+03
2.7E+02
constant variance, power
unrestricted
Hill
3
<.0001
2.71
0.44
281.24
1.2E+03
1.6E+02
constant variance, n
restricted >1, bound hit
Hill
3
<.0001
2.71
0.44
281.24
1.2E+03
3.8E+01
constant variance, n
unrestricted
linear
4
<.0001
36.71
<.0001
313.25
8.3E+03
5.9E+03
constant variance
polynomial
4
<.0001
36.71
<.0001
313.25
8.3E+03
5.9E+03
constant variance
power
4
<.0001
36.71
<.0001
313.25
8.3E+03
5.9E+03
constant variance, power
restricted >1, bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-77 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
3
<.0001
6.08
0.11
284.61
2.8E+00
9.6E-05
constant variance, power
unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.2.7.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
60
50
40
30
20
10
0
BVIDLBMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
12:27 11/20 2009
H.2.7.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_BMRl_Lung_EROD_wk53.(d)
Gnuplot Plotting File:
Fri Nov 20 12:27:08 2009
Tbl 12, Week 53, Lung Microsomes EROD
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-78 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -0.80064
rho 1.47683
a 2.86045
b 0.000243673
c 16.0581
d 1
Parameter Estimates
Variable Model 4
lnalpha -1.14118
rho 1.62714
a 3.06882
b 0.000715169
c 13.702
d 3.78652
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 3.011 1.584
1354 8 27.15 5.269
3056 8 42.85 11.15
5259 8 36.57 12.99
8918 8 43.75 18.55
1.6e+004 8 43.71 6.322
Estimated Values of Interest
Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-79 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0
3. 061
1.408
-0.1008
1354
27 .14
8 . 377
0. 001867
3056
38.19
11 . 07
1.191
5259
42.16
12 . 01
-1.317
8918
43.23
12 . 25
0.1192
1.6e+004
43.33
12 .28
0.08869
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -135.2677 7 284.5353
A2 -115.6885 12 255.3771
A3 -121.1517 8 258.3034
R -162.0902 2 328.1805
4 -122.9684 5 255.9369
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
92 . 8
39.16
10. 93
3. 633
10
5
4
3
p-value
< 0.0001
< 0.0001
0.0274
0.3039
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 6a is greater than .1. Model 4 seems
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-80 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 52.8515
BMDL = 32.57 06
H.2.7.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
60
50
40
30
20
10
0
BVIDL
BMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
12:27 11/20 2009
H.2.7.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_Unrest_BMRl_Lung_EROD_wk53.(d)
Gnuplot Plotting File:
Fri Nov 20 12:27:17 2009
Tbl 12, Week 53, Lung Microsomes EROD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-81 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'""d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)'""d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -0.80064
rho 1.47683
a 2.86045
b 0.000243673
c 16.0581
d 1
Parameter Estimates
Variable Model 5
lnalpha -1.14118
rho 1.62714
a 3.06882
b 0.000715169
c 13.702
d 3.78652
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 3.011 1.584
1354 8 27.15 5.269
3056 8 42.85 11.15
5259 8 36.57 12.99
8918 8 43.75 18.55
1.6e+004 8 43.71 6.322
Estimated Values of Interest
Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-82 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0
3. 069
1. 407
-0.1162
1354
25. 95
7 . 993
0.4234
3056
42 . 05
i—1
i—1
CO
0.1907
5259
42 . 05
i—1
i—1
CO
-1.309
8918
42 . 05
1—1
1—1
CO
0.4055
6e+004
42 . 05
1—1
1—1
CO
0.3976
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -135.2677 7 284.5353
A2 -115.6885 12 255.3771
A3 -121.1517 8 258.3034
R -162.0902 2 328.1805
5 -122.4411 6 256.8821
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
92 . 8
39.16
10. 93
2.579
10
5
4
2
p-value
< 0.0001
< 0.0001
0.0274
0.2755
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-83 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
The p-value for Test 7a is greater than .1. Model 5 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 584.449
BMDL = 36.24 97
H.2.7.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
60
50
40
0)
w
20
10
0
B
12:27 11/20
H.2.7.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
WDLJBMD
2000 4000 6000 8000 10000 12000 14000 16000
dose
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Hill_Unrest_BMRl_Lung_EROD_wk53.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_Lung_EROD_wk53.plt
Fri Nov 20 12:27:19 2009
Tbl 12, Week 53, Lung Microsomes EROD
The form of the response function is:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-84 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.76968
rho = 0
intercept = 3.011
v = 40.735
n = 2.49974
k = 1565.11
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -k
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha rho intercept v n
lalpha 1 -0.96 -0.5 0.17 NA
rho -0.96 1 0.47 -0.25 NA
intercept
-0.5
0.17
NA
0.47
-0.25
NA
1
-0.25
NA
-0.25
1
NA
NA
NA
NA
NA - This parameter's variance has been estimated as zero or less.
THE MODEL HAS PROBABLY NOT CONVERGED!!!
Parameter Estimates
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
NA NA
NA NA
NA NA
NA NA
NA NA
At least some variance estimates are negative.
THIS USUALLY MEANS THE MODEL HAS NOT CONVERGED!
Try again from another starting point.
Variable Estimate Std. Err.
lalpha -1.07501 NA
rho 1.68859 NA
intercept 3.011 NA
v 35.7 938 NA
n 5.85653 NA
k 2.91999e-005 NA
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-85 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0
8
3. 01
3. 01
1354
8
27 .1
CD
00
CO
3056
8
CO
C\]
CO
CO
CO
5259
8
CO
CO
CO
CO
8918
8
43.7
38 . 8
1.6e+004 8 43.7 38.8
I.58 1.48 -2.78e-009
5.27 12.8 -2.57
II.2 12.8 0.891
13 12.8 -0.493
18.5 12.8 1.09
6.32 12.8 1.08
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -135.267662 7 284.535325
A2 -115.688533 12 255.377067
A3 -121.151707 8 258.303413
fitted -129.200555 5 268.401110
R -162.090242 2 328.180484
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
92.8034
39.1583
10.9263
16.0977
10
5
4
3
<.0001
<.0001
0.0274
0.001083
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-86 DRAFT—DO NOT CITE OR QUOTE
-------
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1.70749e-005
BMDL = 1.7074 9e-005
H.2.7.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
60
50
40
30
20
10
0
BMD
0
2000
4000
6000
8000
10000
12000
14000
16000
dose
12:27 11/20 2009
H.2.7.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_Lung_EROD_wk53.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Pwr_Unrest_BMRl_Lung_EROD_wk53.plt
Fri Nov 20 12:27:20 2009
Tbl 12, Week 53, Lung Microsomes EROD
The form of the response function is:
Y[dose] = control + slope * doseApower
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-87 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.76968
rho = 0
control = 3.011
slope = 7 .10636
power = 0.187 655
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho control slope power
lalpha 1 -0.96 -0.49 0.062 -0.046
rho -0.96 1 0.45 -0.074 0.051
control -0.49 0.45 1 -0.075 0.049
slope 0.062 -0.074 -0.075 1 -1
power -0.046 0.051 0.049 -1 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
-1. 02691
1.6303
3.01554
7.64061
0.18001
Std. Err.
0. 818371
0.240525
0.519298
4 . 22038
0.0639858
Lower Conf. Limit
-2.63089
1.15888
1.99773
-0.631172
0. 0546001
Upper Conf. Limit
0.577065
2.10172
4 . 03334
15.9124
0.30542
Table of Data and Estimated Values of Interest
2 N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 3.01 3.02 1.58 1.47 -0.00872
1354 8 27.1 31 5.27 9.83 -1.11
3056 8 42.8 35.4 11.2 11 1.92
5259 8 36.6 38.7 13 11.8 -0.52
8918 8 43.7 42.3 18.5 12.7 0.323
1.6e+004 8 43.7 46.7 6.32 13.7 -0.607
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-88 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -135.267662 7 284.535325
A2 -115.688533 12 255.377067
A3 -121.151707 8 258.303413
fitted -125.533162 5 261.066325
R -162.090242 2 328.180484
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.'
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 92.8034 10 <.0001
Test 2 39.1583 5 <.0001
Test 3 10.9263 4 0.0274
Test 4 8.76291 3 0.03261
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.000106161
BMDL = 0.000106161
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-89 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.8. National Toxicology Program. (2006): Tblll Index Week 31
H.2.8.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)c
4
<0.0001
20.59
0.00
46.55
4.8E+03
3.8E+03
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
20.59
0.00
46.55
4.8E+03
3.8E+03
nonconstant variance,
power restricted >1
exponential (M4)
3
<0.0001
23.01
<0.0001
50.97
1.7E+03
1.0E+03
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
23.01
<0.0001
50.97
1.7E+03
1.0E+03
nonconstant variance,
power restricted >1
Hill
3
<0.0001
23.01
<0.0001
50.97
1.7E+03
error
nonconstant variance, n
restricted >1, bound hit
linear
4
<0.0001
23.01
0.00
48.97
1.7E+03
1.0E+03
nonconstant variance
polynomial
3
<0.0001
22.24
<.0001
50.20
3.4E+03
1.1E+03
nonconstant variance
power
4
<0.0001
23.01
0.00
48.97
1.7E+03
1.0E+03
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
4
<0.0001
1.20
0.88
101.67
1.0E+04
9.0E+03
constant variance,
power restricted >1
exponential (M3)
3
<0.0001
0.97
0.81
103.44
1.1E+04
9.0E+03
constant variance,
power restricted >1
exponential (M4)
3
<0.0001
5.31
0.15
107.78
6.8E+03
5.2E+03
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
1.08
0.58
105.55
1.1E+04
7.5E+03
constant variance,
power restricted >1
Hill
2
<0.0001
1.08
0.58
105.55
1.1E+04
7.5E+03
constant variance, n
restricted >1
linear
4
<0.0001
5.31
0.26
105.78
6.8E+03
5.2E+03
constant variance
polynomial
4
<0.0001
1.44
0.84
101.91
1.0E+04
8.9E+03
constant variance
power
3
<0.0001
1.08
0.78
103.55
1.1E+04
7.5E+03
constant variance,
power restricted >1
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-90 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.2.8.2. Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
BMDL
10000
12000
14000
16000
11:23 11/19 2009
H.2.8.3. Output File for Selected Model: Exponential (M2'), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_BMRl_Tblll_31wk.(d)
Gnuplot Plotting File:
Thu Nov 19 11:23:48 2009
Tbl 11, 31wk, Hep Cell Proliferation Labeling Index
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-91 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
lnalpha -0.674004
rho 2.29189
a 0.31065
b 3.44 963e-005
c 24.761
d 1
Parameter Estimates
Variable Model 2
lnalpha
rho
-0.467457
2.1664
0.394038
5.38146e-009
78344.1
1
Table of Stats From Input Data
Dose
0
1284
2932
5075
8629
1. 55e + 004
9
10
10
10
10
10
Obs Mean
0.327
0. 852
0. 956
0.792
1. 333
3.846
Obs Std Dev
0.189
0.6514
0.7368
0.4617
1.123
3. 08
Dose
0
1284
2932
5075
8629
1.55e+004
Estimated Values of Interest
Est Mean Est Std Scaled Residual
0.5305
0.6219
0.7627
0.9 9 4 6
1. 545
3. 619
0. 4275
0. 4 983
0.6067
0.7837
1.198
2 .722
-1.428
1.46
1. 007
-0.8176
-0.5587
0.2635
Other models for which
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
likelihoods are calculated:
= Mu(i) + e(ij)
= Sigma^2
= Mu(i) + e(ij)
= Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-92 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -47.23498 7 108.47
A2 -8.679256 12 41.35851
A3 -8.980651 8 33.9613
R -63.44829 2 130.8966
2 -19.27346 4 46.54692
Additive constant for all log-likelihoods = -54.22. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Test
Test 1
Test 2
Test 3
Test 4
Tests of Interest
-2*log(Likelihood Ratio)
109.5
77 .11
0.6028
20.59
D. F.
10
5
4
4
p-value
< 0.0001
< 0.0001
0.9628
0.0003826
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is less than .1. Model 2 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 4772.05
BMDL = 3816.47
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-93 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.9. Van Birgelen et al. (1995b): T4 UGT
H.2.9.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
<0.0001
33.51
<0.0001
36.92
6.3E+04
4.4E+04
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
33.51
<0.0001
36.92
6.3E+04
4.4E+04
nonconstant variance,
power restricted >1
exponential (M4)c
3
<0.0001
1.50
0.68
6.90
2.7E+03
1.5E+03
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
1.14
0.57
8.55
3.5E+03
1.6E+03
nonconstant variance,
power restricted >1
exponential (M5)d
2
<0.0001
1.14
0.57
8.55
3.5E+03
1.6E+03
nonconstant variance,
power unrestricted
Hill
2
<.0001
1.22
0.54
8.63
3.7E+03
1.7E+03
nonconstant variance, n
restricted >1
Hilld
2
<.0001
1.22
0.54
8.63
3.7E+03
1.5E+03
nonconstant variance, n
unrestricted
linear
4
<.0001
19.72
0.00
23.13
1.8E+04
9.1E+03
nonconstant variance
polynomial
4
<.0001
19.72
0.00
23.13
1.8E+04
9.1E+03
nonconstant variance
power
4
<.0001
19.72
0.00
23.13
1.8E+04
9.1E+03
nonconstant variance,
power restricted >1,
bound hit
power d
3
<.0001
6.02
0.11
11.42
1.3E+03
2.1E+02
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
13.46
0.01
38.87
8.2E+04
6.9E+04
constant variance,
power restricted >1
exponential (M3)
4
<0.0001
13.46
0.01
38.87
8.2E+04
6.9E+04
constant variance,
power restricted >1
exponential (M4)
3
<0.0001
0.11
0.99
27.51
1.3E+04
6.7E+03
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
0.07
0.97
29.47
1.5E+04
6.8E+03
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
0.07
0.97
29.47
1.5E+04
6.8E+03
constant variance,
power unrestricted
Hill
2
<.0001
0.10
0.95
29.50
1.4E+04
5.6E+03
constant variance, n
restricted >1
Hill
2
<.0001
0.10
0.95
29.50
1.4E+04
5.1E+03
constant variance, n
unrestricted
linear
4
<.0001
8.58
0.07
33.98
5.1E+04
3.9E+04
constant variance
polynomial
4
<.0001
8.58
0.07
33.98
5.1E+04
3.9E+04
constant variance
power
4
<.0001
8.58
0.07
33.98
5.1E+04
3.9E+04
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-94 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
3
<.0001
2.70
0.44
30.10
1.1E+04
2.6E+03
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.2.9.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
3.5
2.5
0.5
EjMDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:32 11/20 2009
H.2.9.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_BMRl_T4_UGT.(d)
Gnuplot Plotting File:
Fri Nov 20 12:32:06 2009
Tbl2, T4 UGT
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-95 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -0.937573
rho 1.54913
a 0.3135
b 2.19381e-005
c 8.67464
d 1
Parameter Estimates
Variable Model 4
lnalpha -0.934825
rho 1.69365
a 0.293644
b 5.48685e-005
c 7.66316
d 1.27403
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 0.33 0.198
3969 8 0.6 0.4243
6479 8 0.64 0.4525
9968 8 0.87 0.9051
4.7 61e + 004 8 2.08 1.329
1.37 8e + 005 8 2.59 0.8768
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-96 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
o
3969
647 9
9968
4.761e+004
1.37 8e + 005
0.2841
0.5945
0.7663
0.9778
2 . 062
2 . 351
0.2186
0. 4063
0.5029
0.6171
1.155
1.289
0.594
0. 03836
-0.7106
-0.4939
0. 04516
0.5245
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -9.701316 7 33.40263
A2 4.934967 12 14.13007
A3 2.296438 8 11.40712
R -29.51921 2 63.03841
4 1.548351 5 6.903297
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
68 . 91
29.27
5.277
1.496
10
5
4
3
p-value
< 0.0001
< 0.0001
0.26
0.6832
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-97 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 2726.3
BMDL = 14 91.73
H.2.9.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
3.5
3
2.5
2
1.5
1
0.5
0
EiMDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:32 11/20 2009
H.2.9.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_Unrest_BMRl_T4_UGT.(d)
Gnuplot Plotting File:
Fri Nov 20 12:32:13 2009
Tbl2, T4 UGT
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-98 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'""d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)'""d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -0.937573
rho 1.54913
a 0.3135
b 2.19381e-005
c 8.67464
d 1
Parameter Estimates
Variable Model 5
lnalpha -0.934825
rho 1.69365
a 0.293644
b 5.48685e-005
c 7.66316
d 1.27403
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 0.33 0.198
3969 8 0.6 0.4243
6479 8 0.64 0.4525
9968 8 0.87 0.9051
4.7 61e + 004 8 2.08 1.329
1.37 8e + 005 8 2.59 0.8768
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-99 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
o
3969
647 9
9968
4.761e+004
1.37 8e + 005
0.2936
0.555
0.7533
1. 019
2 .185
2 . 25
0.222
0.3806
0. 4929
0.6369
1. 215
1.245
0. 4632
0.334
-0.6498
-0.6636
-0.2441
0 .7717
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -9.701316 7 33.40263
A2 4.934967 12 14.13007
A3 2.296438 8 11.40712
R -29.51921 2 63.03841
5 1.725713 6 8.548574
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
68 . 91
29.27
5.277
1.141
10
5
4
2
p-value
< 0.0001
< 0.0001
0.26
0.5651
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-100 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
The p-value for Test 7a is greater than .1. Model 5 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 3460.45
BMDL = 1550.03
H.2.9.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
Hill
3.5
2.5
0.5
EiMDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:32 11/20 2009
H.2.9.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Hill_Unrest_BMRl_T4_UGT.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_T4_UGT.plt
Fri Nov 20 12:32:14 2009
Tbl2, T4 UGT
The form of the response function is:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-101 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = -0.462247
rho = 0
intercept = 0.33
v = 2.26
n = 0.525864
k = 66891.8
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 0.035 -0.26 -0.18 -0.017 0.038
rho 0.035 1 0.48 -0.49 0.023 -0.21
intercept -0.26 0.48 1 -0.37 0.26 -0.14
v -0.18 -0.49 -0.37 1 -0.59 0.77
n -0.017 0.023 0.26 -0.59 1 -0.84
k 0.038 -0.21 -0.14 0.77 -0.84 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
-0.933225
1.68188
0.294743
2.10713
1.51694
14931.4
Std. Err.
0.25643
0.441442
0.0705015
0.497534
0.601141
7059.91
Lower Conf. Limit
-1.43582
0. 816665
0.156563
1.13198
0.33872
1094.23
Upper Conf. Limit
-0.430632
2.54709
0. 432924
3.08228
2 .69515
28768.6
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 e
3969 8
647 9 8
9968 8
4.761e+004
1.37 8e + 005
0.33
0.6
0. 64
0. 87
2 . 08
2 .59
0.295
0.544
0.758
1. 04
2 .09
2 . 33
198
424
453
905
1. 33
0 . 877
224
376
497
646
1 . 17
1.28
0.444
0.424
-0.672
-0.723
-0.0297
0.571
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-102 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2 : Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -9.701316 7 33.402631
A2 4.934967 12 14.130066
A3 2.296438 8 11.407124
fitted 1.684209 6 8.631582
R -29.519205 2 63.038411
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
68.9083
29.2726
5.27706
1. 22446
10
5
4
2
<.0001
<.0001
0.26
0.5421
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 3674.98
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-103 DRAFT—DO NOT CITE OR QUOTE
-------
1 BMDL = 14 63.(
2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-104 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.2.9.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
3.5
2.5
0.5
S/IDL
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:32 11/20 2009
H.2.9.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_T4_UGT.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Pwr_Unrest_BMRl_T4_UGT.plt
Fri Nov 20 12:32:14 2009
Tbl2, T4 UGT
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-105 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
lalpha
rho
control
slope
power
-0.462247
0
0.33
0. 00102277
0.650735
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
0. 031
-0.26
-0.13
0. 074
rho
0. 031
1
0.58
0.11
-0.18
control
-0.26
0.58
1
-0.13
0. 067
slope
-0.13
0.11
-0.13
1
-0. 99
power
0. 074
-0.18
0. 067
-0. 99
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
-0.836884
1. 68473
0.2748
0.00549254
0.516485
Std. Err.
0.261303
0. 453932
0. 0676712
0. 00532631
0.0924979
Lower Conf. Limit
-1.34903
0.795038
0.142167
-0.00494685
0.335192
Upper Conf. Limit
-0.324739
2.57442
0. 407433
0. 0159319
0 . 697777
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 E
3969 8
647 9 8
9968 8
4.761e+004
1.37 8e + 005
0.33
0.6
0. 64
0. 87
08
59
275
671
786
913
1 .71
2 .75
198
424
453
905
1. 33
0 . 877
0.222
0.471
0.537
0. 61
1. 03
1. 54
0.704
-0.43
-0.767
-0.2
1. 02
-0.299
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model
Log(likelihood)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-106 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
A1
-9.701316
7
33.402631
A2
4 . 934 967
12
14 .130066
A3
2 .296438
8
11.407124
fitted
-0.712209
5
11.424417
R
-29.519205
2
63.038411
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 68.9083 10 <.0001
Test 2 29.2726 5 <.0001
Test 3 5.27706 4 0.26
Test 4 6.01729 3 0.1108
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1286.41
BMDL = 212.2 64
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-107 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.10. Van Birgelen et al. (1995b): UGT 1A1
H.2.10.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
3
0.00
29.05
<0.0001
166.85
8.1E+04
2.4E+04
nonconstant variance,
power restricted >1
exponential (M3)
3
0.00
29.05
<0.0001
166.85
8.1E+04
2.4E+04
nonconstant variance,
power restricted >1
exponential (M4)c
2
0.00
1.04
0.60
140.83
4.0E+02
2.2E+02
nonconstant variance,
power restricted >1
exponential (M5)
1
0.00
0.97
0.32
142.77
5.0E+02
2.2E+02
nonconstant variance,
power restricted >1
exponential (M5)d
1
0.00
0.97
0.32
142.77
5.0E+02
2.2E+02
nonconstant variance,
power unrestricted
Hill
1
0.00
1.27
0.26
143.07
8.2E+02
error
nonconstant variance, n
restricted >1
Hilld
1
0.00
1.27
0.26
143.07
8.2E+02
error
nonconstant variance, n
unrestricted
linear
3
0.00
26.47
<.0001
164.27
1.8E+04
5.3E+02
nonconstant variance
polynomial
3
0.00
31.07
<.0001
168.87
3.5E+05
4.9E+02
nonconstant variance
power
3
0.00
26.47
<.0001
164.27
1.8E+04
5.3E+02
nonconstant variance,
power restricted >1,
bound hit
power d
2
0.00
5.95
0.05
145.75
3.8E+00
2.3E-04
nonconstant variance,
power unrestricted
exponential (M2)
3
0.00
22.21
<0.0001
165.71
1.6E+05
8.3E+04
constant variance,
power restricted >1
exponential (M3)
3
0.00
22.21
<0.0001
165.71
1.6E+05
8.3E+04
constant variance,
power restricted >1
exponential (M4)
2
0.00
8.05
0.02
153.55
2.6E+03
1.2E+03
constant variance,
power restricted >1
exponential (M5)
1
0.00
7.88
0.00
155.38
3.3E+03
1.2E+03
constant variance,
power restricted >1
exponential (M5)
1
0.00
7.88
0.00
155.38
3.3E+03
1.2E+03
constant variance,
power unrestricted
Hill
1
0.00
8.12
0.00
155.61
3.7E+03
9.6E+02
constant variance, n
restricted >1
Hill
1
0.00
8.12
0.00
155.61
3.7E+03
8.8E+02
constant variance, n
unrestricted
linear
3
0.00
21.83
<.0001
165.32
1.3E+05
6.2E+04
constant variance
polynomial
3
0.00
21.83
<.0001
165.32
1.3E+05
6.2E+04
constant variance
power
3
0.00
21.83
<.0001
165.32
1.3E+05
6.2E+04
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-108 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
2
0.00
13.23
0.00
158.73
7.2E+01
1.6E-06
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fltting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.2.10.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
700
600
500
400
300
200
100
-100
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:43 11/20 2009
H.2.10.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_BMRl_UGT_lAl.(d)
Gnuplot Plotting File:
Fri Nov 20 12:43:51 2009
Tbl2, UGT_1A1
The form of the response function by Model:
Model 2: Y[dose] = a * exp{sign * b * dose}
Model 3: Y[dose] = a * exp{sign * (b * dose)^d}
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-109 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 4:
Model 5:
Y[dose] = a * [c-(c-l) * exp{-b * dose}]
Y[dose] = a * [c-(c-l) * exp{-(b * dosej^d}]
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -1.53604
rho 1.59958
a 95.95
b 1.154 99e-005
c 4.94633
d 1
Parameter Estimates
Variable Model 4
lnalpha -10.3642
rho 3.29138
a 101.591
b 0.000102786
c 3.84125
d 1.08913
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 3 101 15.59
3969 3 194 36.37
9968 3 304 17.32
4.7 61e + 004 3 452 48.5
1.37 8e + 005 3 296 149
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 101.5 11.18 -0.07335
3969 194.7 32.89 -0.03837
9968 282.1 60.75 0.6236
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
4 . 761e + 004
1.37 8e + 005
389.3
392 .1
103.5
104 . 7
1.049
-1.589
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma"2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + log(mean(i)) ^ rho)
Model R: Yij = Mu + e(1)
Var{e(ij)} = Sigma"2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -68.74833 6 149.4967
A2 -58.69126 10 137.3825
A3 -64.89907 7 143.7981
R -80.72265 2 165.4453
4 -65.41669 5 140.8334
Additive constant for all log-likelihoods = -13.78. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
44.06
20.11
12 .42
1. 035
D. F.
p-value
< 0.0001
0.0004741
0.006087
0.5959
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 11 DRAFT—DO NOT CITE OR QUOTE
-------
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 402.539
BMDL = 221.776
H.2.10.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
700
600
500
400
300
200
100
-100
Exponential
20000
40000
60000 80000
dose
100000
120000
140000
12:44 11/20 2009
H.2.10.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Exp_Unrest_BMRl_UGT_lAl.(d)
Gnuplot Plotting File:
Fri Nov 20 12:44:01 2009
Tbl2, UGT_1A1
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign
exp{sign
[c—(c—1)
[c—(c—1)
b * dose}
(b * dose)^d}
exp{-b * dose}]
exp{-(b * dose)^d}
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 12 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -1.53604
rho 1.59958
a 95.95
b 1.154 99e-005
c 4.94633
d 1
Parameter Estimates
Variable Model 5
lnalpha -10.3642
rho 3.29138
a 101.591
b 0.000102786
c 3.84125
d 1.08913
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 3 101 15.59
3969 3 194 36.37
9968 3 304 17.32
4.7 61e + 004 3 452 48.5
1.37 8e + 005 3 296 149
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 101.6 11.28 -0.09081
3969 192.2 32.19 0.09829
9968 286.9 62.23 0.4773
4.7 61e + 004 389.2 102.8 1.058
1.37 8e + 005 390.2 103.3 -1.581
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -68.74833 6 149.4967
A2 -58.69126 10 137.3825
A3 -64.89907 7 143.7981
R -80.72265 2 165.4453
5 -65.38628 6 142.7726
Additive constant for all log-likelihoods = -13.78. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
44.06
20.11
12 .42
0.9744
p-value
< 0.0001
0.0004741
0.006087
0.3236
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is greater than .1. Model 5 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 14 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 504.638
BMDL = 223.156
H.2.10.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
Hill
700
600
500
400
300
200
100
-100
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:44 11/20 2009
H.2.10.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Hill_Unrest_BMRl_UGT_lAl.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Hill_Unrest_BMRl_UGT_lAl.plt
Fri Nov 20 12:44:02 2009
Tbl2 f UGT_1A1
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * In(mean(i)))
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha
rho
intercept
8 . 57191
0
101
351
0.350477
11467.2
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
-0. 99
-0.19
0.14
0.12
-0.0083
rho
-0. 99
1
0.18
-0.17
-0.12
-0.0038
intercept
-0.19
0.18
1
-0.12
0. 031
0.1
v
0.14
-0.17
-0.12
1
-0.57
0.79
n
0.12
-0.12
0. 031
-0.57
1
-0.73
k
-0.0083
-0.0038
0.1
0.79
-0.73
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
-10.4997
3.31877
101.641
296.324
1. 52651
6852.92
Std. Err.
3.7002
0.67548
6.48455
52.8989
0.645076
2333.57
Lower Conf. Limit
-17.752
1.99485
88 . 9319
192.644
0.262182
2279.2
Upper Conf. Limit
-3.24748
4 . 64269
114.351
400.003
2.79084
11426.6
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 3
3969 3
9968 3
4.761e+004
1.37 8e + 005
101
194
304
102
191
291
452
296
383
395
15. 6
36. 4
17 . 3
48.5
149
11. 2
32 .1
64 . 4
102
107
-0.0989
0.141
0.348
1.17
-1. 6
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)/S2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -68.748326 6 149.496653
A2 -58.691256 10 137.382511
A3 -64.899072 7 143.798144
fitted -65.536514 6 143.073028
R -80.722651 2 165.445302
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
44 . 0628
20.1141
12 .4156
1.27488
<.0001
0.0004741
0.006087
0.2589
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 823.8
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.2.10.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
700
600
500
400
300
200
100
-100
BMD
0
20000
40000
60000
80000
100000
120000
140000
dose
12:44 11/20 2009
H.2.10.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Blood\Pwr_Unrest_BMRl_UGT_lAl.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Blood\Pwr_Unrest_BMRl_UGT_lAl.plt
Fri Nov 20 12:44:03 2009
Tbl2, UGT_1A1
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
lalpha
rho
control
slope
power
8 . 57191
0
101
19.8524
0.225107
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 99
-0.22
0.0058
0. 021
rho
-0. 99
1
0.21
0. 024
-0.057
control
-0.22
0.21
1
-0.14
0.11
slope
0.0058
0. 024
-0.14
1
-0. 99
power
0. 021
-0.057
0.11
-0. 99
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
-11.5995
3.55351
101.406
7 . 06329
0.337328
Std. Err.
3. 46108
0. 629824
6.37341
7 . 31729
0.106575
Lower Conf. Limit
-18.3831
2.31908
88.9147
-7.27833
0.128444
Upper Conf. Limit
-4 . 81588
4.78795
113.898
21.4049
0.546212
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 3
3969 3
9968 3
4.761e+004
1.37 8e + 005
101
194
304
3 452
3 296
101
217
259
369
484
15. 6
36. 4
17 . 3
11.1
42 . 9
58 . 8
-0.0634
-0.928
1. 32
48.5
149
110
178
1. 31
-1. 82
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi
Var{e(i)}
Mu + e(i)
Sigma'*-2
Likelihoods of Interest
Model
A1
Log(likelihood)
-68.748326
AIC
149.496653
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-l 19 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
A2 -58.691256 10 137.382511
A3 -64.899072 7 143.798144
fitted -67.875596 5 145.751193
R -80.722651 2 165.445302
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 44.0628 8 <.0001
Test 2 20.1141 4 0.0004741
Test 3 12.4156 3 0.006087
Test 4 5.95305 2 0.05097
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 3.82374
BMDL = 0.000231902
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-120 DRAFT—DO NOT CITE OR QUOTE
-------
H.2.11. Vanden Heuvel et al. (1994): Hepatic CYP1A1 mRNA Expression
H.2.11.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
5
<0.0001
518.90
<0.0001
1114.48
4.2E+03
3.2E+03
nonconstant variance,
power restricted >1
exponential (M3)
5
<0.0001
518.90
<0.0001
1114.48
4.2E+03
3.2E+03
nonconstant variance,
power restricted >1
exponential (M4)
4
<0.0001
71.31
<0.0001
668.92
2.2E+01
1.0E+01
nonconstant variance,
power restricted >1
exponential (M5)c
3
<0.0001
35.23
<0.0001
634.84
4.5E+02
3.3E+02
nonconstant variance,
power restricted >1
Hill
3
<.0001
33.65
<.0001
633.26
5.3E+02
error
nonconstant variance,
n restricted >1
linear
5
<.0001
79.92
<.0001
675.53
1.6E+01
8.5E+00
nonconstant variance
polynomial
5
<.0001
235.66
<.0001
831.27
1.4E+05
3.0E+02
nonconstant variance
power
4
<.0001
77.35
<.0001
674.96
2.1E+01
1.1E+01
nonconstant variance,
power restricted >1
exponential (M2)
5
<0.0001
27.27
<0.0001
1178.21
6.7E+04
5.9E+04
constant variance,
power restricted >1
exponential (M3)
4
<0.0001
62.38
<0.0001
1215.33
1.6E+09
6.0E+06
constant variance,
power restricted >1
exponential (M4)
4
<0.0001
0.86
0.93
1153.81
5.8E+03
4.1E+03
constant variance,
power restricted >1
exponential (M5)
3
<0.0001
0.00
1.00
1154.95
9.0E+03
4.4E+03
constant variance,
power restricted >1
Hill
3
<.0001
0.00
1.00
1154.95
8.4E+03
3.5E+03
constant variance, n
restricted >1
linear
5
<.0001
19.42
0.00
1170.37
3.0E+04
2.4E+04
constant variance
polynomial
5
<.0001
26.27
<.0001
1177.21
2.4E+04
2.1E+04
constant variance
power
5
<.0001
19.32
0.00
1170.27
3.1E+04
2.4E+04
constant variance,
power restricted >1,
bound hit
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-121 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.2.11.2. Figure for Selected Model: Exponential (M5), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 5 with 0.95 Confidence Level
70000
60000
50000
40000
30000
20000
10000
MDLBMD
Exponential
20000
40000
60000
dose
80000
100000
120000
11:31 11/19 2009
H.2.11.3. Output File for Selected Model: Exponential (M5), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\AD\Blood\Exp_BMRl_hepatic_CYPlAl_mRNA_expression.(d)
Gnuplot Plotting File:
Thu Nov 19 11:31:49 2009
[insert study notes]
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-122 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 7
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -0.89532
rho 2.01401
a 5.13
b 2.68046e-005
c 7511.7
d 1
Parameter Estimates
Variable Model 5
lnalpha
rho
0.166401
1.90534
9.80088
7.30524e-005
3246.67
2 . 37353
Table of Stats From Input Data
Dose
N
0 13
3.805 5
35.91 12
301.9 7
2149 7
1. 43e + 004
1.147e + 005
11
Obs Mean
5.4
7 . 2
14 . 8
12 . 8
536
1. 8e + 004
3. 67e + 004
Obs Std Dev
3. 606
5.59
14 . 9
4 . 498
320.1
1. 522e + 004
2 . 214e + 004
Dose
0
3. 805
35. 91
301. 9
2149
1. 43e + 004
1.147e + 005
Estimated Values of Interest
Est Mean Est Std Scaled Residual
9. 801
9. 801
9. 825
13.52
400.1
2 .133e + 004
3 .182e + 004
9.561
9.561
9.583
12 . 99
327 . 4
1. 446e + 004
117e+004
2
-1. 66
-0.6083
1.799
-0.1474
1.099
-0.7638
0.5154
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-123 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
Var{e(ij)} = Sigma(i)/X2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -572.4744 8 1160.949
A2 -290.7965 14 609.5929
A3 -293.806 9 605.6119
R -603.6646 2 1211.329
5 -311.4203 6 634.8406
Additive constant for all log-likelihoods = -55.14. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1
Test 2
Test 3
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 625.7 12 < 0.0001
Test 2 563.4 6 < 0.0001
Test 3 6.019 5 0.3044
Test 7a 35.23 3 < 0.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 449.252
BMDL = 332.057
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-124 DRAFT—DO NOT CITE OR QUOTE
-------
H.3. ADMINISTERED DOSE BMDS RESULTS
H.3.1. Hassoun et al. (2000): CytC Liver
H.3.1.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
0.39
15.15
0.00
-140.98
2.8E+01
1.9E+01
nonconstant variance,
power restricted >1
exponential (M3)
4
0.39
15.15
0.00
-140.98
2.8E+01
1.9E+01
nonconstant variance,
power restricted >1
exponential (M4)
3
0.39
1.73
0.63
-152.40
7.5E+00
4.6E+00
nonconstant variance,
power restricted >1
exponential (M5)
2
0.39
0.56
0.76
-151.57
1.2E+01
5.2E+00
nonconstant variance,
power restricted >1
exponential (M5)
2
0.39
0.56
0.76
-151.57
1.2E+01
5.2E+00
nonconstant variance,
power unrestricted
Hill
2
0.39
0.67
0.72
-151.46
1.3E+01
error
nonconstant variance, n
restricted >1
Hill
2
0.39
0.67
0.72
-151.46
1.3E+01
4.5E+00
nonconstant variance, n
unrestricted
linear
4
0.39
7.87
0.10
-148.27
1.5E+01
1.0E+01
nonconstant variance
polynomial
4
0.39
7.87
0.10
-148.27
1.5E+01
1.0E+01
nonconstant variance
power
4
0.39
7.87
0.10
-148.27
1.5E+01
1.0E+01
nonconstant variance,
power restricted >1,
bound hit
power
3
0.39
3.95
0.27
-150.18
5.6E+00
1.7E+00
nonconstant variance,
power unrestricted
exponential (M2)
4
0.39
16.43
0.00
-139.08
3.9E+01
3.3E+01
constant variance, power
restricted >1
exponential (M3)
4
0.39
16.43
0.00
-139.08
3.9E+01
3.3E+01
constant variance, power
restricted >1
exponential (M4)c
3
0.39
1.70
0.64
-151.81
9.1E+00
5.9E+00
constant variance,
power restricted >1
exponential (M5)
2
0.39
0.48
0.79
-151.02
1.4E+01
6.5E+00
constant variance, power
restricted >1
exponential (M5)d
2
0.39
0.48
0.79
-151.02
1.4E+01
6.5E+00
constant variance, power
unrestricted
Hill
2
0.39
0.60
0.74
-150.90
1.5E+01
6.3E+00
constant variance, n
restricted >1
Hilld
2
0.39
0.60
0.74
-150.90
1.5E+01
5.9E+00
constant variance, n
unrestricted
linear
4
0.39
10.56
0.03
-144.95
2.5E+01
1.9E+01
constant variance
polynomial
4
0.39
10.56
0.03
-144.95
2.5E+01
1.9E+01
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-125 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Model
Degrees
of
Freedom
Variance
/>-Value3
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
4
0.39
10.56
0.03
-144.95
2.5E+01
1.9E+01
constant variance, power
restricted >1, bound hit
power d
3
0.39
4.52
0.21
-148.99
6.6E+00
2.0E+00
constant variance, power
unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.1.2. Figure for Selected Model: Exponential (M4), Constant Variance, Power Restricted
>1
Exponential_beta Model 4 with 0.95 Confidence Level
0.6
Exponential
0.5
0.4
0.3
0.2
BMDL
BMD
0
20
40
60
80
100
dose
13:45 11/23 2009
H.3.1.3. Output File for Selected Model: Exponential (M4), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Exp_CV_BMRl_CytC_Liver.(d)
Gnuplot Plotting File:
Mon Nov 23 13:45:24 2009
TBARs, liver only (Table 2)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-126 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'M}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
Specified
lnalpha -5.48625
rho(S) 0
a 0.1387
b 0.027423
c 3.36121
d 1
Parameter Estimates
Variable Model 4
lnalpha -5.47287
rho 0
a 0.156285
b 0.0293581
c 2.85125
d 1.56807
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0
6
0
146
0.06614
3
6
0
177
0.05389
10
6
0
191
0.05634
22
6
0
271
0.05634
46
6
0
388
0.06369
100
6
0
444
0.1102
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-127 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0
0.1413
0. 06591
0
1762
3
0.1646
0.06591
0
4609
10
0.2131
0. 06591
-0
8196
22
0.2796
0. 06591
-0
3199
46
0.3676
0. 06591
0
7587
100
0. 4509
0. 06591
-0
2564
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 80.75258 7 -147.5052
A2 83.37355 12 -142.7471
A3 80.75258 7 -147.5052
R 55.82002 2 -107.64
4 79.90337 4 -151.8067
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
55.11
5.242
5.242
1.698
D. F.
10
5
5
3
p-value
< 0.0001
0.3871
0.3871
0.6373
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-128 DRAFT—DO NOT CITE OR QUOTE
-------
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 9.0851
BMDL = 5.88612
H.3.1.4. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
0.6
Exponential
0.5
0.4
0.3
0.2
BMDL
BMD
0
20
40
60
80
100
dose
13:45 11/23 2009
H.3.1.5. Output File for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Exp_CV_Unrest_BMRl_CytC_Liver.(d)
Gnuplot Plotting File:
Mon Nov 23 13:45:31 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-129 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
TBARs, liver only (Table 2)
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
Specified
lnalpha -5.48625
rho(S) 0
a 0.1387
b 0.027423
c 3.36121
d 1
Parameter Estimates
Variable Model 5
lnalpha -5.47287
rho 0
a 0.156285
b 0.0293581
c 2.85125
d 1.56807
Table of Stats From Input Data
Obs Mean
0
3
10
22
46
146
177
191
271
388
Obs Std Dev
06614
05389
05634
05634
06369
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-130 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
100 6 0.444 0.1102
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0
0.1563
0
0648
-0.3888
3
0.1626
0
0648
0.5434
10
0.1957
0
0648
-0.1766
22
0.2708
0
0648
0. 007576
46
0.3873
0
0648
0.02644
100
0.4443
0
0648
-0.01203
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 80.75258 7 -147.5052
A2 83.37355 12 -142.7471
A3 80.75258 7 -147.5052
R 55.82002 2 -107.64
5 80.51171 5 -151.0234
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
55.11
5.242
5.242
0.4817
D. F.
10
5
5
2
p-value
< 0.0001
0.3871
0.3871
0.7859
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-131 DRAFT—DO NOT CITE OR QUOTE
-------
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 14.1987
BMDL = 6.537 38
H.3.1.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
0.6
Hill
0.5
0.4
0.3
0.2
BMDL
BMD
0
20
40
60
80
100
dose
13:45 11/23 2009
H.3.1.7. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Hill_CV_Unrest_BMRl_CytC_Liver.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Hill_CV_Unrest_BMRl_CytC_Liver.plt
Mon Nov 23 13:45:33 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-132 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
TBARs, liver only (Table 2)
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
rho
intercept
0.004972
0
0.146
0.298
17 . 5689
65.0769
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
8 . 6e-008
-3.4e-008
3. 4e-008
8 . 6e-008
intercept
8.6e-008
1
-0. 61
0.53
0. 069
v
-3.4e-008
-0. 61
1
-0.84
0. 64
n
3. 4e-008
0.53
-0.84
1
-0.52
k
J. 6e-008
0. 069
0. 64
-0.52
1
Parameter Estimates
Variable
alpha
intercept
Estimate
. 00421303
0.159748
0.305175
2 .11196
28 .1195
Std. Err.
0.00099302
0.0202818
0. 0615956
1. 024
6.8986
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
. 00226674
0.119997
0.18445
0.104959
14.5985
. 00615931
0.1995
0.4259
4 .11895
41. 6405
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
0.146
0.177
0.16
0.162
0.0661
0.0539
0.0649
0.0649
-0.519
0.55
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-133 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
10 6 0.191
22 6 0.271
46 6 0.388
100 6 0.444
0.191 0.0563
0.274 0.0563
0.385 0.0637
0.445 0.11
0.0649 0.0134
0.0649 -0.1
0.0649 0.106
0.0649 -0.0503
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma'"2
Model A2: Yij = Mu(i) + e(ij
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 80.752584 7 -147.505168
A2 83.373547 12 -142.747094
A3 80.752584 7 -147.505168
fitted 80.452332 5 -150.904663
R 55.820023 2 -107.640047
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
55.107
5.24193
5.24193
0.600505
10
5
5
2
<.0001
0.3871
0.3871
0.7406
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-134 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 15.1313
BMDL = 5.93521
H.3.1.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
0.5
0.4
0.3
0.2
BMDL
BMD
0
20
40
60
80
100
dose
13:45 11/23 2009
H.3.1.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Pwr_CV_Unrest_BMRl_CytC_Liver.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Pwr_CV_Unrest_BMRl_CytC_Liver.plt
Mon Nov 23 13:45:33 2009
TBARs, liver only (Table 2)
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-135 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The power is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 0.004972
rho = 0 Specified
control = 0.146
slope = 0.0109242
power = 0.717 914
Asymptotic Correlation Matrix of Parameter Estimates
alpha
control
The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
1
]. 8e-010
control
3. 8e-010
1
slope
-3.8e-009
-0.77
power
4.5e-009
0. 68
slope -3.8e-009 -0.77 1 -0.98
power 4.5e-009 0.68 -0.98 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
0. 00469717
0.135495
0.0232652
0.573772
Std. Err.
0.00110713
0.0246289
0.013381
0.119032
Lower Conf. Limit
0. 00252723
0.0872229
-0.00296103
0.340474
Upper Conf. Limit
0.00686711
0.183766
0.0494915
0.80707
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
6
0
146
0.135
0.0661
0
0685
0.375
3
6
0
177
0.179
0.0539
0
0685
-0.0784
10
6
0
191
0.223
0.0563
0
0685
-1.13
22
6
0
271
0.273
0.0563
0
0685
-0.056
46
6
0
388
0.345
0.0637
0
0685
1. 54
100
6
0
444
0. 462
0.11
0
0685
-0.653
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-136 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 80.752584 7 -147.505168
A2 83.373547 12 -142.747094
A3 80.752584 7 -147.505168
fitted 78.494318 4 -148.988637
R 55.820023 2 -107.640047
Explanation of Tests
Test 1:
Test
2
Test
3
Test
4
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.'
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 55.107 10 <.0001
Test 2 5.24193 5 0.3871
Test 3 5.24193 5 0.3871
Test 4 4.51653 3 0.2108
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 6.57302
BMDL = 1.96558
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-137 DRAFT—DO NOT CITE OR QUOTE
-------
H.3.2. Hassoun et al. (2000): DNA SSB
H.3.2.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
0.75
47.92
<0.0001
121.75
3.8E+01
2.5E+01
nonconstant variance,
power restricted >1
exponential (M3)
4
0.75
47.92
<0.0001
121.75
3.8E+01
2.5E+01
nonconstant variance,
power restricted >1
exponential (M4)
3
0.75
8.98
0.03
84.81
3.7E+00
2.2E+00
nonconstant variance,
power restricted >1
exponential (M5)
3
0.75
8.98
0.03
84.81
3.7E+00
2.2E+00
nonconstant variance,
power restricted >1
exponential (M5)
3
0.75
8.98
0.03
84.81
3.7E+00
2.2E+00
nonconstant variance,
power unrestricted
Hill
3
0.75
7.46
0.06
83.29
2.6E+00
1.4E+00
nonconstant variance, n
restricted >1, bound hit
Hill
2
0.75
3.76
0.15
81.60
6.6E-01
1.8E-01
nonconstant variance, n
unrestricted
linear
4
0.75
39.32
<.0001
113.16
1.9E+01
1.0E+01
nonconstant variance
polynomial
4
0.75
39.32
<.0001
113.16
1.9E+01
1.0E+01
nonconstant variance
power
4
0.75
39.32
<.0001
113.16
1.9E+01
1.0E+01
nonconstant variance,
power restricted >1,
bound hit
power
3
0.75
4.68
0.20
80.52
3.0E-01
8.5E-02
nonconstant variance,
power unrestricted
exponential (M2)
4
0.75
48.54
<0.0001
120.83
3.0E+01
2.5E+01
constant variance, power
restricted >1
exponential (M3)
4
0.75
48.54
<0.0001
120.83
3.0E+01
2.5E+01
constant variance, power
restricted >1
exponential (M4)
3
0.75
8.53
0.04
82.81
3.7E+00
2.8E+00
constant variance, power
restricted >1
exponential (M5)
3
0.75
8.53
0.04
82.81
3.7E+00
2.8E+00
constant variance, power
restricted >1
exponential (M5)d
3
0.75
8.53
0.04
82.81
3.7E+00
2.8E+00
constant variance, power
unrestricted
Hillc
3
0.75
7.12
0.07
81.41
2.9E+00
2.0E+00
constant variance, n
restricted >1, bound hit
Hilld
2
0.75
4.03
0.13
80.32
9.6E-01
2.1E-01
constant variance, n
unrestricted
linear
4
0.75
38.88
<.0001
111.16
1.8E+01
1.5E+01
constant variance
polynomial
4
0.75
38.88
<.0001
111.16
1.8E+01
1.5E+01
constant variance
power
4
0.75
38.88
<.0001
111.16
1.8E+01
1.5E+01
constant variance, power
restricted >1, bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-138 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Model
Degrees
of
Freedom
Variance
/>-Value3
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power d
3
0.75
5.10
0.16
79.39
4.4E-01
1.5E-01
constant variance, power
unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.2.2. Figure for Selected Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
25
20
15
10
5
BMDL
BMD
0
20
40
60
80
100
dose
13:44 11/23 2009
H.3.2.3. Output File for Selected Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Exp_CV_Unrest_BMRl_DNA_SSB.(d)
Gnuplot Plotting File:
Mon Nov 23 13:44:02 2009
DNA single-strand breaks, liver only (Table 3)
The form of the response function by Model:
Model 2: Y[dose] = a * exp{sign * b * dose}
Model 3: Y[dose] = a * exp{sign * (b * dose)^d}
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-139 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 4:
Model 5:
Y[dose] = a * [c-(c-l) * exp{-b * dose}]
Y[dose] = a * [c-(c-l) * exp{-(b * dosej^d}]
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
Specified
lnalpha 0.841244
rho(S) 0
a 7.0395
b 0.0279582
c 3.50522
d 1
Parameter Estimates
Variable Model 5
lnalpha 1.07816
rho 0
a 8.47733
b 0.0311513
c 2.84178
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 6 7.41 1.543
3 6 10.78 1.249
10 6 13.6 1.69
22 6 15.3 1.715
46 6 20.4 2.254
100 6 23.5 1.372
Estimated Values of Interest
Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-140 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0 8.477 1.714 -1.525
3 9.87 1.714 1.3
10 12.66 1.714 1.348
22 16.22 1.714 -1.318
46 20.37 1.714 0.04957
100 23.4 1.714 0.1459
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma"2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -33.14239 7 80.28478
A2 -31.81197 12 87.62394
A3 -33.14239 7 80.28478
R -80.44209 2 164.8842
5 -37.40682 4 82.81364
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
97 .26
2 . 661
2 . 661
8 . 529
10
5
5
3
p-value
< 0.0001
0.7521
0.7521
0.03626
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-141 DRAFT—DO NOT CITE OR QUOTE
-------
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 3.73387
BMDL = 2.78339
H.3.2.4. Figure for Unrestricted Model: Hill, Constant Variance, n Restricted >1, Bound
Hit
Hill Model with 0.95 Confidence Level
Hill
25
20
15
10
5
BMDL
BMD
0
20
40
60
80
100
dose
13:43 11/23 2009
H.3.2.5. Output File for Unrestricted Model: Hill, Constant Variance, n Restricted >1,
Bound Hit
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Hill_CV_BMRl_DNA_SSB.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Hill_CV_BMRl_DNA_SSB.plt
Mon Nov 23 13:43:57 2009
DNA single-strand breaks, liver only (Table 3)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-142 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter restricted to be greater than 1
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 2.7831
rho = 0 Specified
intercept = 7.41
v = 16.09
n = 0.174831
k = 69.2706
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha intercept v k
alpha 1 l.le-007 1.9e-007 1.9e-007
intercept l.le-007 1 0.099 0.61
v 1.9e-007 0.099 1 0.79
k 1.9e-007 0.61 0.79 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
alpha 2.82659 0.666233 1.5208 4.13238
intercept 8.16404 0.581043 7.02522 9.30286
v 20.1253 1.69013 16.8127 23.4379
n 1 NA
k 31.702 8.35815 15.3203 48.0836
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
3
10
7 .41
10.8
13. 6
3.16
9. 9
13
1. 54
1. 25
1.69
-1.1
1.28
0.889
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-143 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
22 6 15.3
46 6 20.4
100 6 23.5
16.4 1.71
20.1 2.25
23.4 1.37
1.68 -1.62
1.68 0.469
1.68 0.0802
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)~2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -33.142389 7 80.284779
A2 -31.811970 12 87.623940
A3 -33.142389 7 80.284779
fitted -36.703273 4 81.406545
R -80.442086 2 164.884172
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
97.2602
2.66084
2.66084
7.12177
10
5
5
3
<.0001
0.7521
0.7521
0.06812
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-144 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 2.88 97 6
BMDL = 2.00669
H.3.2.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
Hill
25
20
15
10
5
0
20
40
60
80
100
dose
13:44 11/23 2009
H.3.2.7. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Hill_CV_Unrest_BMRl_DNA_SSB.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Hill_CV_Unrest_BMRl_DNA_SSB.plt
Mon Nov 23 13:44:03 2009
DNA single-strand breaks, liver only (Table 3)
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-145 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
rho
intercept
2 .7831
0
7 .41
16.09
0.174831
69.2706
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
-2 . 2e-008
-4 . 6e-008
8 . 4e-009
-4 . 3e-008
intercept
-2 . 2e-008
1
-0.33
0.47
-0.29
v
-4 . 6e-008
-0.33
1
-0. 95
1
n
J. 4e-009
0.47
-0. 95
1
-0. 96
k
-4 . 3e-008
-0.29
1
-0. 96
1
Parameter Estimates
Variable
alpha
intercept
Estimate
2 .5942
7 . 47627
36.9014
0.612877
148.104
Std. Err.
0. 611459
0. 665055
25.5466
0.190055
303.532
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
1.39576
6.17278
-13.1689
0.240376
-446.809
3.79264
8 . 77 975
86.9718
0.985377
743.016
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 6 7.41 7.48
3 6 10.8 10.6
10 6 13.6 13.4
22 6 15.3 16.2
46 6 20.4 19.6
100 6 23.5 23.7
1.54 1.61 -0.101
1.25 1.61 0.313
1.69 1.61 0.286
1.71 1.61 -1.41
2.25 1.61 1.24
1.37 1.61 -0.33
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-146 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/S2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -33.142389 7 80.284779
A2 -31.811970 12 87.623940
A3 -33.142389 7 80.284779
fitted -35.159023 5 80.318046
R -80.442086 2 164.884172
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
97 .2602
2.66084
2.66084
4.03327
10
5
5
2
<.0001
0.7521
0.7521
0.1331
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.961789
BMDL = 0.211403
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-147 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
H.3.2.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
25
20
15
10
5
BMD
0
20
40
60
80
100
dose
13:44 11/23 2009
H.3.2.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Pwr_CV_Unrest_BMRl_DNA_SSB.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Pwr_CV_Unrest_BMRl_DNA_SSB.plt
Mon Nov 23 13:44:04 2009
DNA single-strand breaks, liver only (Table 3)
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 2.7831
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-148 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
rho
control
slope
power
0 Specified
7 .41
1.99047
0.4538
Asymptotic Correlation Matrix of Parameter Estimates
( *** The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
control
slope
power
alpha
1
le-010
3. 4e-009
-3.5e-009
control
le-010
1
-0.79
0. 66
slope
3. 4e-009
-0.79
1
-0. 97
power
-3.5e-009
0. 66
-0. 97
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
2.67247
7.29122
2.31759
0.428335
Std. Err.
0. 629906
0. 640201
0.501503
0. 0441267
Lower Conf. Limit
1.43787
6. 03645
1.33466
0.341848
Upper Conf. Limit
3. 90706
8 . 54599
3.30051
0.514821
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
3
10
22
46
100
7 .41
10.8
13. 6
15.3
20.4
23.5
7.29
11
13.5
16
19.2
24
1. 54
1. 25
1.69
1.71
2 . 25
1. 37
1. 63
1. 63
1. 63
1. 63
1. 63
1. 63
0.178
-0.332
0.142
-1. 05
1.74
-0.678
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model
Log(likelihood)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-149 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
A1 -33.142389 7 80.284779
A2 -31.811970 12 87.623940
A3 -33.142389 7 80.284779
fitted -35.694033 4 79.388067
R -80.442086 2 164.884172
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
97.2602
2.66084
2.66084
5.10329
10
5
5
3
<.0001
0.7521
0.7521
0.1644
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.442709
BMDL = 0.149473
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-150 DRAFT—DO NOT CITE OR QUOTE
-------
H.3.3. Hassoun et al. (2000): TBARs Liver
H.3.3.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
0.33
20.31
0.00
-4.29
7.0E+01
3.4E+01
nonconstant variance,
power restricted >1
exponential (M3)
4
0.33
20.31
0.00
-4.29
7.0E+01
3.4E+01
nonconstant variance,
power restricted >1
exponential (M4)
3
0.33
3.08
0.38
-19.53
4.3E+00
1.9E+00
nonconstant variance,
power restricted >1
exponential (M5)
2
0.33
2.78
0.25
-17.82
5.5E+00
2.0E+00
nonconstant variance,
power restricted >1
exponential (M5)
2
0.33
2.78
0.25
-17.82
5.5E+00
2.0E+00
nonconstant variance,
power unrestricted
Hill
2
0.33
2.52
0.28
-18.08
5.7E+00
2.0E+00
nonconstant variance, n
restricted >1
Hill
2
0.33
2.52
0.28
-18.08
5.7E+00
error
nonconstant variance, n
unrestricted
linear
4
0.33
19.16
0.00
-5.44
5.2E+01
2.2E+01
nonconstant variance
polynomial
4
0.33
19.16
0.00
-5.44
5.2E+01
2.2E+01
nonconstant variance
power
4
0.33
19.16
0.00
-5.44
5.2E+01
2.2E+01
nonconstant variance,
power restricted >1,
bound hit
power
3
0.33
8.22
0.04
-14.38
1.2E+00
5.2E-03
nonconstant variance,
power unrestricted
exponential (M2)
4
0.33
20.40
0.00
-6.14
8.0E+01
5.3E+01
constant variance,
power restricted >1
exponential (M3)
4
0.33
20.40
0.00
-6.14
8.0E+01
5.3E+01
constant variance,
power restricted >1
exponential (M4)c
3
0.33
3.36
0.34
-21.18
4.9E+00
2.3E+00
constant variance,
power restricted >1
exponential (M5)
2
0.33
2.86
0.24
-19.68
6.7E+00
2.5E+00
constant variance,
power restricted >1
exponential (M5)d
2
0.33
2.86
0.24
-19.68
6.7E+00
2.5E+00
constant variance,
power unrestricted
Hill
2
0.33
2.61
0.27
-19.93
6.3E+00
2.6E+00
constant variance, n
restricted >1
Hilld
2
0.33
2.61
0.27
-19.93
6.3E+00
2.6E+00
constant variance, n
unrestricted
linear
4
0.33
19.52
0.00
-7.02
6.9E+01
4.4E+01
constant variance
polynomial
4
0.33
19.52
0.00
-7.02
6.9E+01
4.4E+01
constant variance
power
4
0.33
19.52
0.00
-7.02
6.9E+01
4.4E+01
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-151 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
X2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power d
3
0.33
9.55
0.02
-14.99
2.9E+00
6.1E-02
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.3.2. Figure for Selected Model: Exponential (M4), Constant Variance, Power Restricted
>1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
3
2.5
2
.5
1
BMDL
BMD
0
20
40
60
80
100
dose
13:44 11/23 2009
H.3.3.3. Output File for Selected Model: Exponential (M4), Constant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Exp_CV_BMRl_TBARs_Liver.(d)
Gnuplot Plotting File:
Mon Nov 23 13:44:41 2009
TBARs, liver only (Table 2)
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-152 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
Specified
lnalpha -1.90388
rho(S) 0
a 1.39555
b 0.0194898
c 1.97051
d 1
Parameter Estimates
Variable Model 4
lnalpha -1.82448
rho 0
a 1.46519
b 0.113543
c 1.63661
d 2.13652
Table of Stats From Input Data
use N Obs Mean Obs Std Dev
0 6 1.469 0.2915
3 6 1.549 0.5389
10 6 2.15 0.3625
22 6 2.28 0.2474
46 6 2.619 0.5168
100 6 2.292 0.4874
Estimated Values of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-153 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dose Est Mean Est Std Scaled Residual
0
1.404
0
4044
0.3915
3
1. 674
0
4044
-0.7582
10
2 . 063
0
4044
0.527
22
2 . 332
0
4044
-0.3134
46
2 .438
0
4044
1.099
100
2.448
0
4044
-0.9458
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(1)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 16.26977 7 -18.53954
A2 19.12783 12 -14.25565
A3 16.26977 7 -18.53954
R 2.44294 2 -0.8858799
4 14.5907 4 -21.18141
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
33.37
5.716
5.716
3.358
p-value
10
5
5
3
0. 000236
0.3348
0.3348
0.3396
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-154 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 4.91639
BMDL = 2.2 9 952
H.3.3.4. Figure for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
3
2.5
2
.5
1
BMDL
BMD
0
20
40
60
80
100
dose
13:44 11/23 2009
H.3.3.5. Output File for Unrestricted Model: Exponential (M5), Constant Variance, Power
Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov23\Exp_CV_Unrest_BMRl_TBARs_Liver.(d)
Gnuplot Plotting File:
Mon Nov 23 13:44:47 2009
TBARs, liver only (Table 2)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-155 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'M}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
rho is set to 0.
A constant variance model is fit.
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
Specified
lnalpha -1.90388
rho(S) 0
a 1.39555
b 0.0194898
c 1.97051
d 1
Parameter Estimates
Variable Model 5
lnalpha -1.82448
rho 0
a 1.46519
b 0.113543
c 1.63661
d 2.13652
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 6 1.469 0.2915
3 6 1.549 0.5389
10 6 2.15 0.3625
22 6 2.28 0.2474
46 6 2.619 0.5168
100 6 2.292 0.4874
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-156 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 1.465 0.4016 0.02326
3 1.554 0.4016 -0.03103
10 2.147 0.4016 0.02011
22 2.397 0.4016 -0.7145
46 2.398 0.4016 1.348
100 2.398 0.4016 -0.6461
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 16.26977 7 -18.53954
A2 19.12783 12 -14.25565
A3 16.26977 7 -18.53954
R 2.44294 2 -0.8858799
5 14.8407 5 -19.68141
Additive constant for all log-likelihoods = -33.08. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
33.37
5.716
5.716
2 . 858
D. F.
10
5
5
2
p-value
0.000236
0.3348
0.3348
0.2395
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is greater than .1. A homogeneous
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-157 DRAFT—DO NOT CITE OR QUOTE
-------
variance model appears to be appropriate here.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
to adeguately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 6.73152
BMDL = 2.47 02 9
H.3.3.6. Figure for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
Hill
3
2.5
2
.5
1
BMDL
BMD
0
20
40
60
80
100
dose
13:44 11/23 2009
H.3.3.7. Output File for Unrestricted Model: Hill, Constant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Hill_CV_Unrest_BMRl_TBARs_Liver.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Hill_CV_Unrest_BMRl_TBARs_Liver.plt
Mon Nov 23 13:44:49 2009
TBARs, liver only (Table 2)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-158 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
Power parameter is not restricted
A constant variance model is fit
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha
rho
intercept
0.178788
0
1.469
1.15
0.921061
11.2346
Specified
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
intercept
alpha
1
4 . 6e-010
-1. 2e-008
2 . 8e-009
3. 8e-009
intercept
4 . 6e-010
1
-0. 82
0.48
0.52
v
-1.2e-008
-0. 82
1
-0. 61
-0.22
n
2 . 8e-009
0.48
-0. 61
1
0.29
k
3. 8e-009
0.52
-0.22
0.29
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
intercept
Estimate
0.160182
1.4615
0.962989
2 .4861
7 .18099
Std. Err.
0.0377552
0.152914
0.202872
1.76422
2.79941
Lower Conf. Limit
0.0861829
1.16179
0.565367
-0.971707
1.69424
Upper Conf. Limit
0.23418
1.76121
1. 36061
5.94391
12.6677
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
3
10
1.47
1. 55
2 .15
1.46
1. 56
2 .13
0.291
0.539
0.363
0.4
0.4
0.4
0.0459
-0.0685
0.118
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-159 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
22 6 2.28
46 6 2.62
100 6 2.29
2.37 0.247
2.42 0.517
2.42 0.487
0.4 -0.541
0.4 1.25
0.4 -0.802
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)~2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)) = Sigma^2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 16.269770 7 -18.539539
A2 19.127827 12 -14.255654
A3 16.269770 7 -18.539539
fitted 14.966039 5 -19.932079
R 2.442940 2 -0.885880
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
33.3698
5.71611
5.71611
2.60746
10
5
5
2
0.000236
0.3348
0.3348
0.2715
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-160 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 6.26103
BMDL = 2.57 4 65
H.3.3.8. Figure for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
2.5
BMD
0
20
40
60
80
100
dose
13:44 11/23 2009
H.3.3.9. Output File for Unrestricted Model: Power, Constant Variance, Power Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov23\Pwr_CV_Unrest_BMRl_TBARs_Liver.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov23\Pwr_CV_Unrest_BMRl_TBARs_Liver.plt
Mon Nov 23 13:44:49 2009
TBARsf liver only (Table 2)
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
rho is set to 0
The power is not restricted
A constant variance model is fit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-161 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
alpha = 0.178788
rho = 0 Specified
control = 1.469
slope = 0.0756538
power = 0.652114
Asymptotic Correlation Matrix of Parameter Estimates
alpha
control
The model parameter(s) -rho
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
alpha
1
1. le-008
control
1. le-008
1
slope
-1. le-009
-0.75
power
-1. 5e-008
0.47
slope -1.le-009 -0.75 1 -0.91
power -1.5e-008 0.47 -0.91 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
alpha
control
slope
power
Estimate
0.194232
1.42104
0.333105
0.262735
Std. Err.
0. 0457809
0.171077
0.166768
0. 0983956
Lower Conf. Limit
0.104503
1.08573
0. 00624603
0. 0698836
Upper Conf. Limit
0.283961
1.75634
0.659963
0. 455587
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 6 1.47 1.42 0.291 0.441 0.267
3 6 1.55 1.87 0.539 0.441 -1.76
10 6 2.15 2.03 0.363 0.441 0.661
22 6 2.28 2.17 0.247 0.441 0.603
46 6 2.62 2.33 0.517 0.441 1.6
100 6 2.29 2.54 0.487 0.441 -1.37
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-162 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
Var{e(ij)} = Sigma/N2
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 16.269770 7 -18.539539
A2 19.127827 12 -14.255654
A3 16.269770 7 -18.539539
fitted 11.496634 4 -14.993268
R 2.442940 2 -0.885880
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 33.3698 10 0.000236
Test 2 5.71611 5 0.3348
Test 3 5.71611 5 0.3348
Test 4 9.54627 3 0.02284
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is greater than .1. A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 2 . 90232
BMDL = 0.0614 971
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-163 DRAFT—DO NOT CITE OR QUOTE
-------
H.3.4. Kitchin et al. (1979): BaP Hydrolase Activity
H.3.4.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
9
<0.0001
247.00
<0.0001
452.67
2.6E+03
1.2E+03
nonconstant variance,
power restricted >1
exponential (M3)
9
<0.0001
247.00
<0.0001
452.67
2.6E+03
1.2E+03
nonconstant variance,
power restricted >1
exponential (M4)
8
<0.0001
18.96
0.02
226.60
1.8E+00
1.4E+00
nonconstant variance,
power restricted >1
exponential (M5)c
7
<0.0001
16.75
0.02
226.40
3.4E+00
1.6E+00
nonconstant variance,
power restricted >1
exponential (M5)d
7
<0.0001
16.75
0.02
226.40
3.4E+00
1.6E+00
nonconstant variance,
power unrestricted
Hill
7
<.0001
296.88
<.0001
506.53
error
error
nonconstant variance, n
restricted >1
Hilld
7
<.0001
296.88
<.0001
506.53
error
error
nonconstant variance, n
unrestricted
linear
9
<.0001
94.11
<.0001
299.75
2.8E+00
2.0E+00
nonconstant variance
polynomial
9
<.0001
-197.64
<.0001
8.00
error
error
nonconstant variance
power
9
<.0001
94.11
<.0001
299.75
2.8E+00
2.0E+00
nonconstant variance,
power restricted >1,
bound hit
power d
8
<.0001
63.59
<.0001
271.23
3.0E-01
1.3E-01
nonconstant variance,
power unrestricted
exponential (M2)
9
<0.0001
129.40
<0.0001
451.61
3.6E+03
3.1E+03
constant variance,
power restricted >1
exponential (M3)
9
<0.0001
129.40
<0.0001
451.61
3.6E+03
3.1E+03
constant variance,
power restricted >1
exponential (M4)
8
<0.0001
6.93
0.54
331.19
2.7E+01
2.1E+01
constant variance,
power restricted >1
exponential (M5)
8
<0.0001
6.93
0.54
331.19
2.7E+01
2.1E+01
constant variance,
power restricted >1
exponential (M5)
8
<0.0001
6.93
0.54
331.19
2.7E+01
2.1E+01
constant variance,
power unrestricted
Hill
7
<.0001
67.64
<.0001
393.90
5.7E+02
5.2E+00
constant variance, n
restricted >1
Hill
7
<.0001
2.70
0.91
328.96
2.0E+01
1.1E+01
constant variance, n
unrestricted
linear
9
<.0001
120.31
<.0001
442.57
1.9E+03
1.4E+03
constant variance
polynomial
9
<.0001
120.31
<.0001
442.57
1.9E+03
1.4E+03
constant variance
power
9
<.0001
120.31
<.0001
442.57
1.9E+03
1.4E+03
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-164 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
8
<.0001
51.05
<.0001
375.31
1.2E+00
2.5E-01
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.4.2. Figure for Selected Model: Exponential (M5), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 5 with 0.95 Confidence Level
250
200
150
100
50
Exponential
S/IDL
BMD
1000
2000
3000
dose
4000
5000
6000
7000
14:26 11/20 2009
H.3.4.3. Output File for Selected Model: Exponential (M5), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_BaP_hydro_act.(d)
Gnuplot Plotting File:
Fri Nov 20 14:26:45 2009
Kitchin 1979, Tbl3, BaP hydrolase activity
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-165 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 11
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -3.27793
rho 1.92227
a 4.655
b 0.000532066
c 42.6316
d 1
Parameter Estimates
Variable Model 5
lnalpha -2.6425
rho 1.93734
a 5.43493
b 0.00574894
c 31.1998
d 1.21529
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 9 4.9 1.11
0.2 4 4.9 1.18
0.667 4 6.7 1.4
1.33 4 7.2 1.8
6.67 4 8.3 0.26
20 4 14 5
66.7 4 59 6.8
200 4 96 46
667 4 155 16.4
1670 4 182 26
6670 4 189 26
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-166 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dose
Estimated Values of Interest
Est Mean Est Std Scaled Residual
0
5. 435
1.
375
-1.167
0.2
5.479
1.
386
-0.8354
0. 667
5. 625
1.
422
1. 513
1. 33
5. 874
1.
483
1.789
6. 67
8 . 524
2 .
127
-0.211
20
16.86
4 .
118
-1.391
66.7
49.42
11
. 67
1. 642
200
119. 4
27
.42
-1.705
667
168 . 6
38
. 31
-0.7095
1670
169.6
38
. 52
0.6454
6670
169.6
38
. 52
1. 009
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -158.1306 12 340.2613
A2 -84.80028 22 213.6006
A3 -98.82189 13 223.6438
R -234.6252 2 473.2504
5 -107.1994 6 226.3987
Additive constant for all log-likelihoods = -45.03. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
299.6
146.7
28 . 04
16.75
20
10
9
7
p-value
< 0.0001
< 0.0001
0.0009381
0.01905
The p-value for Test 1 is less than .05. There appears to be a
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-167 DRAFT—DO NOT CITE OR QUOTE
-------
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is less than .1. Model 5 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 3.41185
BMDL = 1.60436
H.3.4.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
250
Exponential
200
150
100
S/IDLBMD
0
1000
2000
3000
4000
5000
6000
7000
dose
14:27 11/20 2009
H.3.4.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-168 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_Unrest_BMRl_BaP_hydro_act.(d)
Gnuplot Plotting File:
Fri Nov 20 14:27:02 2009
Kitchin 1979, Tbl3, BaP hydrolase activity
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'""d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * doseJ'M}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 11
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -3.27793
rho 1.92227
a 4.655
b 0.000532066
c 42.6316
d 1
Parameter Estimates
Variable Model 5
lnalpha -2.6425
rho 1.93734
a 5.43493
b 0.00574894
c 31.1998
d 1.21529
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 9 4.9 1.11
0.2 4 4.9 1.18
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-169 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0. 667
4
6.7
1. 4
1. 33
4
7 . 2
1. 8
6. 67
4
8 . 3
0.26
20
4
14
5
66.7
4
59
6.8
200
4
96
46
667
4
155
16.4
1670
4
182
26
6670
4
189
26
Estimated
Values
of
Interest
Dose
Est Mean
Est
Std
Scaled Residi
0
5. 435
1.
375
-1.167
0.2
5.479
1.
386
-0.8354
0. 667
5. 625
1.
422
1. 513
1. 33
5. 874
1.
483
1.789
6. 67
8 . 524
2 .
127
-0.211
20
16.86
4 .
118
-1.391
66.7
49.42
11
. 67
1. 642
200
119. 4
27
.42
-1.705
667
168 . 6
38
. 31
-0.7095
1670
169.6
38
. 52
0.6454
6670
169.6
38
. 52
1. 009
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -158.1306 12 340.2613
A2 -84.80028 22 213.6006
A3 -98.82189 13 223.6438
R -234.6252 2 473.2504
5 -107.1994 6 226.3987
Additive constant for all log-likelihoods = -45.03. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-170 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
-2*log(Likelihood Ratio) D. F. p-value
Test 1 299.6 20 < 0.0001
Test 2 146.7 10 < 0.0001
Test 3 28.04 9 0.0009381
Test 7a 16.75 7 0.01905
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is less than .1. Model 5 may not adeguately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 3.41185
BMDL = 1.60436
H.3.4.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
250
200
150
0
C/5
c
o
Q.
C/5
CD
a:
« 100
o
50
Hill Model
0 1000 2000 3000 4000 5000 6000 7000
dose
14:27 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-171 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
H.3.4.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_BaP_hydro_act.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_Unrest_BMRl_BaP_hydro_act.plt
Fri Nov 20 14:27:04 2009
Kitchin 1979, Tbl3, BaP hydrolase activity
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 11
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 5.7 0855
rho = 0
intercept = 4.9
v = 184 .1
n = 18
k = 1126.48
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha NA NA NA NA NA NA
rho NA NA NA NA NA NA
intercept NA NA 1 -0.012 NA NA
v NA NA -0.012 1 NA NA
n NA NA NA NA NA NA
k NA NA NA NA NA NA
NA - This parameter's variance has been estimated as zero or less.
THE MODEL HAS PROBABLY NOT CONVERGED!!!
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha 8.0472 NA NA NA
rho -0.0780259 NA NA NA
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-172 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
intercept
-1. 52215e-006
NA
NA
NA
V
185.167
NA
NA
NA
n
17 . 997 9
NA
NA
NA
k
117036
NA
NA
NA
At least some variance estimates are negative.
THIS USUALLY MEANS THE MODEL HAS NOT CONVERGED!
Try again from another starting point.
Table of Data and Estimated Values of Interest
5 N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
9
4 . 9
-1.52e-006
1.11
94 . 3
0.156
0.2
4
4 . 9
-1. 52e-006
1.18
94 . 3
0.104
0. 667
4
6.7
-1.52e-006
1. 4
94 . 3
0.142
1 . 33
4
7 . 2
-1. 52e-006
1. 8
94 . 3
0.153
6. 67
4
8 . 3
-1.52e-006
0.26
94 . 3
0.176
20
4
14
-1.52e-006
5
94 . 3
0.297
66.7
4
59
-1. 52e-006
6.8
94 . 3
1. 25
200
4
96
-1. 52e-006
46
94 . 3
2 . 04
667
4
155
-1. 52e-006
16.4
94 . 3
3.29
1670
4
182
-1.52e-006
26
94 . 3
3.86
6670
4
189
-1. 52e-006
26
94 . 3
4 . 01
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -158.130647 12 340.261294
A2 -84.800279 22 213.600558
A3 -98.821893 13 223.643786
fitted -247.263464 6 506.526928
R -234.625213 2 473.250426
Explanation of Tests
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
Test 1:
Test 2
Test 3
Test 4
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-173 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
Test 1 299.65 20 <.0001
Test 2 146.661 10 <.0001
Test 3 28.0432 9 0.0009381
Test 4 296.883 7 <.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1.#QNAN
BMDL computation failed.
H.3.4.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
250
Power
200
150
100
0
1000
2000
3000
4000
5000
6000
7000
dose
14:27 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-174 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
H.3.4.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_BaP_hydro_act.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Pwr_Unrest_BMRl_BaP_hydro_act.plt
Fri Nov 20 14:27:04 2009
Kitchin 1979, Tbl3, BaP hydrolase activity
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(
exp(lalpha + log(mean(i)
rho)
Total number of dose groups = 11
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-006
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 5.7 0855
rho = 0
control = 4.9
slope = 0.98 4 853
power = 0.59404
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 9
-0.45
0.26
-0.23
rho
-0. 9
1
0.35
-0.24
0.12
control
-0.45
0.35
1
-0.45
0.42
slope
0.26
-0.24
-0.45
1
-0. 92
power
-0.23
0.12
0.42
-0. 92
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
-3.42083
2 .42943
4.52619
2 .4104
0. 619986
Std. Err.
0.570828
0.164289
0.315826
0.540821
0. 0482232
Lower Conf. Limit
-4.53963
2 .10743
3.90719
1. 35041
0.525471
Upper Conf. Limit
-2.30202
2 .75143
5.1452
3. 47039
0.714502
Table of Data and Estimated Values of Interest
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-175 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dose
Obs Mean
Est Mean
Obs Std Dev Est Std Dev Scaled Res.
0 9 4.9 4.53
0.2 4 4.9 5.41
0.667 4 6.7 6.4
1 . 33 4 7.2 7.4
6.67 4 8.3 12.3
20 4 14 20
66.7 4 59 37.1
200 4 96 68.9
667 4 155 140
1670 4 182 244
6670 4 189 571
1.11 1.13 0.991
1.18 1.41 -0.732
1.4 1.72 0.346
1.8 2.06 -0.197
0.26 3.83 -2.11
5 6.87 -1.74
6.8 14.6 3
46 30.9 1.75
16.4 73.4 0.399
26 144 -0.868
26 403 -1.89
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -158.130647 12 340.261294
A2 -84.800279 22 213.600558
A3 -98.821893 13 223.643786
fitted -130.616947 5 271.233893
R -234.625213 2 473.250426
Explanation of Tests
Test 1:
Test 2
Test 3
Test 4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
299.65
146.661
28.0432
63.5901
20
10
9
<.0001
<.0001
0.0009381
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than
model appears to be appropriate
.1. A non-homogeneous variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-176 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.29535
BMDL = 0.127 27
H.3.5. National Toxicology Program. (2006): EROD Liver Week 53
H.3.5.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
p-Value a
x2 Test
Statistic
xV
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
<0.0001
121.00
<0.0001
210.78
5.7E+01
4.0E+01
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
121.00
<0.0001
210.78
5.7E+01
4.0E+01
nonconstant variance,
power restricted >1
exponential (M4)
3
<0.0001
7.05
0.07
98.86
2.7E-01
1.9E-01
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
6.44
0.04
100.25
3.4E-01
2.0E-01
nonconstant variance,
power restricted >1
Hillc
2
<0001
3.05
0.22
96.86
5.4E-01
3.3E-01
nonconstant variance,
n restricted >1
Hilld
2
<.0001
3.05
0.22
96.86
5.4E-01
3.3E-01
nonconstant variance, n
unrestricted
linear
4
<.0001
113.79
<.0001
203.61
2.9E+01
1.1E+01
nonconstant variance
polynomial
4
<.0001
113.79
<.0001
203.61
2.9E+01
1.1E+01
nonconstant variance
power
4
<.0001
113.79
<.0001
203.61
2.9E+01
1.1E+01
nonconstant variance,
power restricted >1,
bound hit
exponential (M2)
4
<0.0001
85.26
<0.0001
209.43
5.0E+01
4.1E+01
constant variance, power
restricted >1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-177 DRAFT—DO NOT CITE OR QUOTE
-------
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M3)
4
<0.0001
85.26
<0.0001
209.43
5.0E+01
4.1E+01
constant variance, power
restricted >1
exponential (M4)
3
<0.0001
4.50
0.21
130.67
1.5E+00
1.2E+00
constant variance, power
restricted >1
exponential (M5)
3
<0.0001
4.50
0.21
130.67
1.5E+00
1.2E+00
constant variance, power
restricted >1
Hill
2
<.0001
2.30
0.32
130.48
1.7E+00
9.3E-01
constant variance, n
restricted >1
Hill
2
<.0001
2.30
0.32
130.48
1.7E+00
9.3E-01
constant variance, n
unrestricted
linear
4
<.0001
77.49
<.0001
201.66
3.2E+01
2.5E+01
constant variance
polynomial
4
<.0001
77.49
<.0001
201.66
3.2E+01
2.5E+01
constant variance
power
4
<.0001
77.49
<.0001
201.66
3.2E+01
2.5E+01
constant variance, power
restricted >1, bound hit
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fltting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.5.2. Figure for Selected Model: Hill, Nonconstant Variance, n Restricted >1
Hill Model with 0.95 Confidence Level
Hill
25
20
15
10
5
BMDL
BMD
0
20
40
60
80
100
dose
16:50 11/20 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-178 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
H.3.5.3. Output File for Selected Model: Hill, Nonconstant Variance, n Restricted >1
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_BMRl_Tbll2_wk53_EROD_liv.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_BMRl_Tbll2_wk53_EROD_liv.plt
Fri Nov 20 16:50:09 2009
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter restricted to be greater than 1
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 1.59547
rho = 0
intercept = 3.614
v = 17.599
n = 1.38584
k = 12.1933
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -0.96 -0.16 0.086 -0.057 0.041
rho -0.96 1 0.14 -0.11 0.06 -0.045
intercept -0.16 0.14 1 -0.18 0.13 0.069
v 0.086 -0.11 -0.18 1 -0.72 0.84
n -0.057 0.06 0.13 -0.72 1 -0.79
k 0.041 -0.045 0.069 0.84 -0.79 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha -4.86544 0.741662 -6.31907 -3.4118
rho 2.26969 0.287261 1.70667 2.83271
intercept 3.62908 0.133826 3.36679 3.89138
v 17.9785 0.989021 16.0401 19.917
n 1.43249 0.162632 1.11374 1.75124
k 7.81956 1.00384 5.85206 9.78706
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-179 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Table of Data and Estimated Values of Interest
5 N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 3.61 3.63 0.486 0.379 -0.113
3 8 7.27 7.27 0.557 0.833 0.0201
10 8 14.8 14.2 1.61 1.78 0.912
22 8 17.3 18.3 1.59 2.37 -1.19
46 8 20.6 20.3 3.05 2.67 0.306
100 8 21.2 21.2 3.82 2.8 0.061
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -59.086537 7 132.173073
A2 -37.515858 12 99.031716
A3 -40.906180 8 97.812359
fitted -42.430348 6 96.860697
R -116.710291 2 237.420582
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
158.389
43.1414
6.78064
3.04834
10
5
4
2
<.0001
<.0001
0.1479
0.2178
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-180 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.536614
BMDL = 0.328003
H.3.5.4. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
EMDL
16:50 11/20 2009
H.3.5.5. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_Tbll2_wk53_EROD_liv.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_Unrest_BMRl_Tbll2_wk53_EROD_liv.plt
Fri Nov 20 16:50:14 2009
The form of the response function is:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-181 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variadle = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 1.59547
rho = 0
intercept = 3.614
v = 17.599
n = 1.38584
k = 12.1933
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -0.96 -0.16 0.086 -0.057 0.041
rho -0.96 1 0.14 -0.11 0.06 -0.045
intercept -0.16 0.14 1 -0.18 0.13 0.069
v 0.086 -0.11 -0.18 1 -0.72 0.84
n -0.057 0.06 0.13 -0.72 1 -0.79
k 0.041 -0.045 0.069 0.84 -0.79 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
-4 .86544
2 .26969
3. 62908
17 . 9785
1.43249
7 . 81956
Std. Err.
0.741662
0.287261
0.133826
0.989021
0.162632
1.00384
Lower Conf. Limit
-6.31907
1.70667
3.36679
16.0401
1.11374
5.85206
Upper Conf. Limit
-3.4118
2 . 83271
3. 89138
19.917
1.75124
9.78706
Table of Data and Estimated Values of Interest
Dose
N
Obs Mean
Est Mean
Obs Std Dev
Est Std Dev
Scaled Res
0
8
3. 61
3. 63
0.486
0.379
-0.113
3
8
7 . 27
7 . 27
0.557
0. 833
0. 0201
10
8
14 . 8
14 . 2
1. 61
1.78
0. 912
22
8
17 . 3
18 . 3
1.59
2 . 37
-1.19
46
8
20.6
20.3
3. 05
2 . 67
0.306
100
8
21. 2
21. 2
3. 82
2 . 8
0. 061
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-182 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model
A1
A2
A3
fitted
R
Log(likelihood)
-59.086537
-37.515858
-40.906180
-42.430348
-116.710291
Param'
7
12
AIC
132.173073
99.031716
97.812359
96.860697
237.420582
Explanation of Tests
Test
1
Test
2
Test
3
Test
4
(Note:
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adeguately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
158.389
43.1414
6.78064
3.04834
10
5
4
2
<.0001
<.0001
0.1479
0.2178
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adeguately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.536614
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-183 DRAFT—DO NOT CITE OR QUOTE
-------
BMDL = 0.328003
H.3.6. National Toxicology Program. (2006): Lung EROD Week 31
H.3.6.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
p-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
<0.0001
129.30
<0.0001
396.45
8.9E+01
6.2E+01
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
129.30
<0.0001
396.45
8.9E+01
6.2E+01
nonconstant variance,
power restricted >1
exponential (M4)c
3
<0.0001
20.80
0.00
289.99
1.0E-01
6.9E-02
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
20.80
0.00
289.99
1.0E-01
6.9E-02
nonconstant variance,
power restricted >1
exponential (M5)d
3
<0.0001
20.80
0.00
289.99
1.0E-01
6.9E-02
nonconstant variance,
power unrestricted
Hill
3
<.0001
78.84
<.0001
348.02
9.6E+00
error
nonconstant variance, n
restricted >1, bound hit
Hilld
3
<.0001
78.84
<.0001
348.02
9.6E+00
error
nonconstant variance, n
unrestricted
linear
4
<.0001
125.96
<.0001
393.15
6.3E+01
3.1E+01
nonconstant variance
polynomial
4
<.0001
128.35
<.0001
395.53
1.0E+02
2.3E+01
nonconstant variance
power
4
<.0001
125.96
<.0001
393.15
6.3E+01
3.1E+01
nonconstant variance,
power restricted >1,
bound hit
power d
3
<.0001
22.50
<.0001
291.68
1.9E-06
1.9E-06
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
87.28
<0.0001
397.44
6.6E+01
5.3E+01
constant variance,
power restricted >1
exponential (M3)
4
<0.0001
87.28
<0.0001
397.44
6.6E+01
5.3E+01
constant variance,
power restricted >1
exponential (M4)
3
<0.0001
15.56
0.00
327.72
1.7E+00
1.2E+00
constant variance,
power restricted >1
exponential (M5)
3
<0.0001
15.56
0.00
327.72
1.7E+00
1.2E+00
constant variance,
power restricted >1
exponential (M5)
3
<0.0001
15.56
0.00
327.72
1.7E+00
1.2E+00
constant variance,
power unrestricted
Hill
2
<.0001
34.01
<.0001
348.17
2.8E+00
2.4E-01
constant variance, n
restricted >1
Hill
2
<.0001
34.01
<.0001
348.17
2.8E+00
5.0E-05
constant variance, n
unrestricted
linear
4
<.0001
81.72
<.0001
391.88
4.6E+01
3.5E+01
constant variance
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-184 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
polynomial
4
<.0001
81.72
<.0001
391.88
4.6E+01
3.5E+01
constant variance
power
4
<.0001
81.72
<.0001
391.88
4.6E+01
3.5E+01
constant variance,
power restricted >1,
bound hit
power
3
<.0001
22.22
<.0001
334.38
1.0E-02
6.9E-04
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be
selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.6.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
60
50
40
30
20
10
0
i/IDLBMD
0
20
40
60
80
100
dose
14:28 11/20 2009
H.3.6.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_Lung_EROD_wk31.(d)
Gnuplot Plotting File:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-185 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Fri Nov 20 14:28:10 2009
Tbl 12, Week 31, Lung Microsomes EROD
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -1.42653
rho 1.46168
a 1.96745
b 0.034997
c 26.7857
d 1
Parameter Estimates
Variable Model 4
lnalpha -1.46439
rho 1.61106
a 2.12443
b 0.19145
c 22.311
d 1
Table of Stats From Input Data
Dose
N
Obs Mean
Obs Std Dev
0
10
2 . 071
0.9708
3
10
25.34
2.549
10
10
30.39
5. 831
22
10
50.19
8 . 68
46
10
49. 07
13. 91
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-186 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
100 10 48.42 8.933
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 2.124 0.8823 -0.1915
3 21.91 5.779 1.88
10 40.72 9.524 -3.432
22 46.73 10.64 1.029
46 47.39 10.76 0.4921
100 47.4 10.76 0.3006
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -152.0793 7 318.1586
A2 -123.367 12 270.734
A3 -129.5911 8 275.1823
R -206.5175 2 417.0349
4 -139.9927 5 289.9853
Additive constant for all log-likelihoods = -55.14. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
166.3
57 .42
12 .45
20.8
D. F.
10
5
4
3
p-value
< 0.0001
< 0.0001
0.01431
0.0001157
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-187 DRAFT—DO NOT CITE OR QUOTE
-------
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 6a is less than .1. Model 4 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 0.102798
BMDL = 0.069311
H.3.6.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
60
50
40
30
20
10
0
i/IDLBMD
0
20
40
60
80
100
dose
14:28 11/20 2009
H.3.6.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_Unrest_BMRl_Lung_EROD_wk31.(d)
Gnuplot Plotting File:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-188 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Fri Nov 20 14:28:17 2009
Tbl 12, Week 31, Lung Microsomes EROD
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -1.42653
rho 1.46168
a 1.96745
b 0.034997
c 26.7857
d 1
Parameter Estimates
Variable Model 5
lnalpha -1.46439
rho 1.61106
a 2.12443
b 0.19145
c 22.311
d 1
Table of Stats From Input Data
Dose
N
Obs Mean
Obs Std Dev
0
10
2 . 071
0.9708
3
10
25.34
2.549
10
10
30.39
5. 831
22
10
50.19
8 . 68
46
10
49. 07
13. 91
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-189 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
100 10 48.42 8.933
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 2.124 0.8823 -0.1915
3 21.91 5.779 1.88
10 40.72 9.524 -3.432
22 46.73 10.64 1.029
46 47.39 10.76 0.4921
100 47.4 10.76 0.3006
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)"2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -152.0793 7 318.1586
A2 -123.367 12 270.734
A3 -129.5911 8 275.1823
R -206.5175 2 417.0349
5 -139.9927 5 289.9853
Additive constant for all log-likelihoods = -55.14. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
166.3
57 .42
12 .45
20.8
D. F.
10
5
4
3
p-value
< 0.0001
< 0.0001
0.01431
0.0001157
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-190 DRAFT—DO NOT CITE OR QUOTE
-------
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 0.102798
BMDL = 0.0693109
H.3.6.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
60
50
40
o
Q.
20
10
0
14:28 11/20
H.3.6.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
Hill
1
T 1
_L
:
: :e
BMD
0 20 40 60 80 100
dose
2009
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_Lung_EROD_wk31.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_Unrest_BMRl_Lung_EROD_wk31.plt
Fri Nov 20 14:28:19 2009
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-191 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Tbl 12, Week 31, Lung Microsomes EROD
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.17467
rho = 0
intercept = 2.071
v = 48.119
n = 18
k = 15.9059
Asymptotic Correlation Matrix of Parameter Estimates
( The model parameter(s) -n
have been estimated at a boundary point, or have been specified by the user,
and do not appear in the correlation matrix )
lalpha rho intercept v k
lalpha 1 -0.98 0.04 -0.054 -0.092
rho -0.98 1 -0.027 0.046 0.096
intercept 0.04 -0.027 1 -0.82 0.36
v -0.054 0.046 -0.82 1 -0.16
k -0.092 0.096 0.36 -0.16 1
Parameter Estimates
95.0% Wald Confidence Interval
Variable Estimate Std. Err. Lower Conf. Limit Upper Conf. Limit
lalpha 5.40756 1.03169 3.38548 7.42964
rho -0.228427 0.299513 -0.815462 0.358608
intercept 13.5736 2.48089 8.71112 18.436
v 35.6207 3.03486 29.6725 41.5689
n 18 NA
k 10.0457 0.22238 9.60987 10.4816
NA - Indicates that this parameter has hit a bound
implied by some inequality constraint and thus
has no standard error.
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-192 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0
10
2 . 07
13. 6
0. 971
11.1
-3.28
3
10
25.3
13. 6
2 . 55
11.1
3.36
10
10
30. 4
30.7
5. 83
10.1
-0.0833
22
10
50.2
49.2
8 . 68
9.57
0.329
46
10
49.1
49.2
13. 9
9.57
-0.0424
100
10
48.4
49.2
8 . 93
9.57
-0.255
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2 : Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -152.079318 7 318.158637
A2 -123.366985 12 270.733969
A3 -129.591134 8 275.182269
fitted -169.011448 5 348.022896
R -206.517459 2 417.034919
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
166.301
57 .4247
12.4483
78.8406
10
5
4
3
<.0001
<.0001
0.01431
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-193 DRAFT—DO NOT CITE OR QUOTE
-------
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 9.61218
BMDL computation failed.
H.3.6.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
60
50
40
30
20
10
0
i/IDLBMD
0
20
40
60
80
100
dose
14:28 11/20 2009
H.3.6.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_Lung_EROD_wk31.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_Lung_EROD_wk31.plt
Fri Nov 20 14:28:19 2009
Tbl 12, Week 31, Lung Microsomes EROD
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-194 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.17467
rho = 0
control = 2.071
slope = 18.9386
power = 0.224076
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 94
-0.42
0.15
-0.13
rho
-0. 94
1
0.38
-0.19
0.14
control
-0.42
0.38
1
-0.15
0. 093
slope
0.15
-0.19
-0.15
1
-0. 94
power
-0.13
0.14
0. 093
-0. 94
1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
-1.53358
1.6412
2.10983
18 . 5389
0.233238
95.0% Wald Confidence Interval
Std. Err. Lower Conf. Limit Upper Conf. Limit
0.571219 -2.65315 -0.414007
0.166321 1.31521 1.96718
0.270093 1.58046 2.6392
2.01491 14.5897 22.488
0.0324661 0.169605 0.29687
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
3
10
22
46
100
10
10
10
10
10
10
2 . 07
25.3
30. 4
50.2
49.1
48.4
2 .11
26.1
33. 8
40.2
47 . 4
56. 4
0. 971
2 . 55
5. 83
8 . 68
13. 9
8 . 93
0. 857
6.74
8 . 35
9. 63
11
12 . 7
-0.143
-0.338
-1. 3
3.27
0. 481
-1. 98
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
Mu(i) + e(ij ;
Sigma/N2
Mu(i) + e(ij ;
Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-195 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -152.079318 7 318.158637
A2 -123.366985 12 270.733969
A3 -129.591134 8 275.182269
fitted -140.838955 5 291.677909
R -206.517459 2 417.034919
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
166.301
57 .4247
12.4483
22.4956
10
5
4
3
<.0001
<.0001
0.01431
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1. 88864e-006
BMDL = 1. 88864e-006
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-196 DRAFT—DO NOT CITE OR QUOTE
-------
H.3.7. National Toxicology Program. (2006): Lung EROD Week 53
H.3.7.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
<0.0001
66.01
<0.0001
316.32
1.3E+02
8.1E+01
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
66.01
<0.0001
316.32
1.3E+02
8.1E+01
nonconstant variance,
power restricted >1
exponential (M4)c
3
<0.0001
2.82
0.42
255.12
1.2E-01
7.5E-02
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
16.10
0.00
270.40
2.6E-01
1.5E-04
nonconstant variance,
power restricted >1
exponential (M5)d
2
<0.0001
16.10
0.00
270.40
2.6E-01
1.5E-04
nonconstant variance,
power unrestricted
Hill
2
<.0001
81.88
<.0001
336.18
3.0E+02
error
nonconstant variance, n
restricted >1
Hilld
2
<.0001
81.88
<.0001
336.18
3.0E+02
error
nonconstant variance, n
unrestricted
linear
4
<.0001
65.65
<.0001
315.96
1.2E+02
6.3E+01
nonconstant variance
polynomial
4
<.0001
65.65
<.0001
315.96
1.2E+02
6.3E+01
nonconstant variance
power
4
<.0001
65.65
<.0001
315.96
1.2E+02
6.3E+01
nonconstant variance,
power restricted >1,
bound hit
power d
3
<.0001
8.50
0.04
260.80
3.8E-10
3.8E-10
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
43.26
<0.0001
319.80
8.0E+01
6.0E+01
constant variance,
power restricted >1
exponential (M3)
4
<0.0001
43.26
<0.0001
319.80
8.0E+01
6.0E+01
constant variance,
power restricted >1
exponential (M4)
3
<0.0001
3.04
0.39
281.57
9.2E-01
5.5E-01
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
2.71
0.26
283.24
2.2E+00
5.7E-01
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
2.71
0.26
283.24
2.2E+00
5.7E-01
constant variance,
power unrestricted
Hill
2
<.0001
2.71
0.26
283.24
2.7E+00
3.2E-01
constant variance, n
restricted >1
Hill
2
<.0001
2.71
0.26
283.24
2.7E+00
1.2E-02
constant variance, n
unrestricted
linear
4
<.0001
41.45
<.0001
317.99
6.5E+01
4.4E+01
constant variance
polynomial
4
<.0001
41.45
<.0001
317.99
6.5E+01
4.4E+01
constant variance
power
4
<.0001
41.45
<.0001
317.99
6.5E+01
4.4E+01
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-197 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
3
<.0001
5.93
0.11
284.47
5.3E-04
5.3E-04
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.7.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
60
50
40
30
20
10
0
0
20
40
60
80
100
dose
14:29 11/20 2009
H.3.7.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_Lung_EROD_wk53.(d)
Gnuplot Plotting File:
Fri Nov 20 14:29:03 2009
Tbl 12, Week 53, Lung Microsomes EROD
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-198 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2: Y[dose]
Model 3: Y[dose]
Model 4: Y[dose]
Model 5: Y[dose]
= a * exp{sign *
= a * exp{sign *
= a * [c-(c — 1) *
= a * [c-(c — 1) *
b * dose}
(b * dose ) "M}
exp{-b * dose}]
exp{-(b * dosej^d}]
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -0.80064
rho 1.47683
a 2.86045
b 0.0390303
c 16.0581
d 1
Parameter Estimates
Variable Model 4
lnalpha -1.07501
rho 1.68859
a 3.011
b 3.22004
c 12.8877
d 18
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 3.011 1.584
3 8 27.15 5.269
10 8 42.85 11.15
22 8 36.57 12.99
46 8 43.75 18.55
100 8 43.71 6.322
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-199 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0
3. 068
1.404
-0.1156
3
26.26
8 . 088
0.3111
10
40.44
11. 5
0.5912
22
42.43
11. 96
-1. 386
46
42.49
i—1
i—1
>£>
CO
0.2969
100
42.49
1—1
1—1
VD
CO
0.2891
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -135.2677 7 284.5353
A2 -115.6885 12 255.3771
A3 -121.1517 8 258.3034
R -162.0902 2 328.1805
4 -122.5608 5 255.1215
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
92 . 8
39.16
10. 93
2 . 818
10
5
4
3
p-value
< 0.0001
< 0.0001
0.0274
0.4205
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 6a is greater than .1. Model 4 seems
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-200 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 0.122595
BMDL = 0.0752795
H.3.7.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
60
50
40
0
C/5
c
o
Q.
20
10
0
B
14:29 11/20
H.3.7.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
J/IDL
BMD
0
2009
20
40
60
80
100
dose
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C :\USEPA\E.MDS21\Nov2 0\E::p_Unrest_E.MRl_Lung_EROD_wk53 . (d)
Gnup'lO't Plotting File:
Fri Nov 2 0 14:29:09 2 00 9
Tbl 12, Week 53, Lung Microsomes ERuD
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-201 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)'""d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)'""d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -0.80064
rho 1.47683
a 2.86045
b 0.0390303
c 16.0581
d 1
Parameter Estimates
Variable Model 5
lnalpha -1.07501
rho 1.68859
a 3.011
b 3.22004
c 12.8877
d 18
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 3.011 1.584
3 8 27.15 5.269
10 8 42.85 11.15
22 8 36.57 12.99
46 8 43.75 18.55
100 8 43.71 6.322
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-202 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0 3.011 1.482 -4.539e-008
3 38.8 12.82 -2.571
10 38.8 12.82 0.8915
22 38.8 12.82 -0.4931
46 38.8 12.82 1.09
100 38.8 12.82 1.082
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i) ) 'k rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma"2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -135.2677 7 284.5353
A2 -115.6885 12 255.3771
A3 -121.1517 8 258.3034
R -162.0902 2 328.1805
5 -129.2006 6 270.4011
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
92 . 8
39.16
10. 93
16.1
10
5
4
2
p-value
< 0.0001
< 0.0001
0.0274
0.0003195
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-203 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 0.260501
BMDL = 0.000148718
H.3.7.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
60
50
40
30
20
10
dose
14:29 11/20 2009
H.3.7.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_Lung_EROD_wk53.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_Unrest_BMRl_Lung_EROD_wk53.plt
Fri Nov 20 14:29:11 2009
Tbl 12, Week 53, Lung Microsomes EROD
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-204 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.76968
rho = 0
intercept = 3.011
v = 40.735
n = 1. 63324
k = 3.46862
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho intercept v n k
lalpha 1 -1 0.00098 -0.015 NA NA
rho -1 1 -0.00098 0.015 NA NA
intercept 0.00098 -0.00098 1 -1.5e-005 NA NA
v -0.015 0.015 -1.5e-005 1 NA NA
n NA NA NA NA NA NA
k NA NA NA NA NA NA
NA - This parameter's variance has been estimated as zero or less.
THE MODEL HAS PROBABLY NOT CONVERGED!!!
Parameter Estimates
95.0% Wald Confidence Interval
Variable
Estimate
Std. Err.
Lower Conf. Limit
Upper Conf. Limi'
lalpha
16.2956
NA
NA
NA
rho
-3.01917
NA
NA
NA
intercept
32 . 8392
NA
NA
NA
V
81.7793
NA
NA
NA
n
17.5977
NA
NA
NA
k
324 . 4 91
NA
NA
NA
At least some variance
THIS USUALLY MEANS THE
Try again from another
estimates are negative.
MODEL HAS NOT CONVERGED!
starting point.
Table of Data and Estimated Values of Interest
Dose N Obs Mean Est Mean Obs Std Dev Est Std Dev Scaled Res.
0 8 3.01 32.8 1.58 17.8 -4.75
3 8 27.1 32.8 5.27 17.8 -0.906
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-205 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
10 8 42.8
22 8 36.6
46 8 43.7
100 8 43.7
32.8 11.2
32.8 13
32 . 8 18.5
32.8 6.32
17.8 1.59
17.8 0.594
17.8 1.74
17.8 1.73
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model
A1
A2
A3
fitted
R
Log(likelihood)
-135.267662
-115.688533
-121.151707
-162.090242
-162.090242
Param's AIC
7 284.535325
12 255.377067
8 258.303413
6 336.180484
2 328.180484
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.)
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
92 .8034
39.1583
10.9263
81. 8771
10
5
4
2
<.0001
<.0001
0.0274
<.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-206 DRAFT—DO NOT CITE OR QUOTE
-------
Risk Type =
Confidence level =
BMD =
Estimated standard deviations from the control mean
0. 95
301.687
BMDL computation failed.
H.3.7.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
60
50
40
20
10
0
B
14:29 11/20
Power
BMD
0
20
40
60
80
100
dose
2009
H.3.7.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_Lung_EROD_wk53.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Pwr_Unrest_BMRl_Lung_EROD_wk53.plt
Fri Nov 20 14:29:12 2009
Tbl 12, Week 53, Lung Microsomes EROD
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-207 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha = 4.76968
rho = 0
control = 3.011
slope = 23.6162
power = 0.133025
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
-0. 96
-0.48
0.11
-0.048
rho
-0. 96
1
0.45
-0.14
0. 053
control
-0.48
0.45
1
-0.14
0. 05
slope
0.11
-0.14
-0.14
1
-0. 93
power
-0.048
0. 053
0. 05
-0. 93
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
-1. 03233
1. 63033
3. 01788
24 . 0756
0.128899
Std. Err.
0. 815925
0.23978
0.518168
3.58644
0. 0448635
Lower Conf. Limit
-2.63152
1.16037
2 . 00229
17 . 0463
0.040968
Upper Conf. Limit
0.566849
2 .10029
4.03347
31.1049
0.21683
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
3
10
22
46
100
3. 01
27 .1
42 . 8
3 6.6
43.7
43.7
3. 02
30. 8
35. 4
38 . 9
42 . 5
46.6
I. 58
5.27
II. 2
13
18 . 5
6.32
I.47
9.74
10. 9
II. 8
12 . 7
13.7
-0.0133
-1. 05
1. 92
-0.554
0.288
-0.599
Model Descriptions for likelihoods calculated
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
Model A3:
Mu(i) + e(ij;
Sigma/N2
Mu(i) + e(ij ;
Sigma(i)^2
Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-208 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -135.267662 7 284.535325
A2 -115.688533 12 255.377067
A3 -121.151707 8 258.303413
fitted -125.400472 5 260.800944
R -162.090242 2 328.180484
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 92.8034 10 <.0001
Test 2 39.1583 5 <.0001
Test 3 10.9263 4 0.0274
Test 4 8.49753 3 0.03677
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 3.7 6923e-010
BMDL = 3.76923e-010
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-209 DRAFT—DO NOT CITE OR QUOTE
-------
H.3.8. National Toxicology Program. (2006): Tblll Index Week 31
H.3.8.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)c
4
<0.0001
21.34
0.00
47.30
3.3E+01
2.6E+01
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
21.34
0.00
47.30
3.3E+01
2.6E+01
nonconstant variance,
power restricted >1
exponential (M4)
3
<0.0001
25.36
<0.0001
53.32
1.7E+01
1.1E+01
nonconstant variance,
power restricted >1
exponential (M5)
2
<0.0001
21.10
<0.0001
51.06
4.6E+01
2.9E+01
nonconstant variance,
power restricted >1
Hill
3
<.0001
21.10
0.00
49.06
4.6E+01
error
nonconstant variance, n
restricted >1, bound hit
linear
4
<.0001
25.36
<.0001
51.32
1.7E+01
1.1E+01
nonconstant variance
polynomial
3
<.0001
21.87
<.0001
49.83
3.9E+01
1.7E+01
nonconstant variance
power
3
<.0001
21.87
<.0001
49.83
4.5E+01
2.4E+01
nonconstant variance,
power restricted >1
exponential (M2)
4
<0.0001
1.02
0.91
101.49
6.4E+01
5.7E+01
constant variance,
power restricted >1
exponential (M3)
3
<0.0001
1.00
0.80
103.47
6.7E+01
5.7E+01
constant variance,
power restricted >1
exponential (M4)
3
<0.0001
3.38
0.34
105.85
4.4E+01
3.3E+01
constant variance,
power restricted >1
exponential (M5)
2
<0.0001
1.10
0.58
105.57
6.6E+01
3.9E+01
constant variance,
power restricted >1
Hill
2
<.0001
1.10
0.58
105.57
6.6E+01
3.9E+01
constant variance, n
restricted >1
linear
4
<.0001
3.38
0.50
103.85
4.4E+01
3.3E+01
constant variance
polynomial
3
<.0001
1.08
0.78
103.55
6.4E+01
3.9E+01
constant variance
power
3
<.0001
1.10
0.78
103.57
6.6E+01
3.9E+01
constant variance,
power restricted >1
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-210 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.3.8.2. Figure for Selected Model: Exponential (M2), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 2 with 0.95 Confidence Level
Exponential
BMDL BMD
100
dose
16:50 11/20 2009
H.3.8.3. Output File for Selected Model: Exponential (M2'), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_Tblll_31wk.(d)
Gnuplot Plotting File:
Fri Nov 20 16:50:52 2009
Tbl 11, 31wk, Hep Cell Proliferation Labeling Index
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-211 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 2
lnalpha -0.674004
rho 2.29189
a 0.31065
b 0.024912
c 12.9995
d 1
Parameter Estimates
Variable Model 2
lnalpha -0.495833
rho 1.97486
a 0.740304
b 0.0199927
c 5.16751
d 18
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 9
3 10
10 10
22 10
46 10
100 10
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0.327
0. 852
0. 956
0.792
1. 333
3.846
0.189
0.6514
0.7368
0.4617
1.123
3. 08
0 0.6166 0.4987 -1.742
3 0.651 0.5251 1.21
10 0.7391 0.5925 1.158
22 0.9186 0.7287 -0.5493
46 1.419 1.102 -0.2466
100 3.775 2.796 0.08069
Other models for which
Model A1: Yij
Var{e(ij)}
Model A2: Yij
Var{e(ij)}
likelihoods are calculated:
= Mu(i) + e(ij)
= Sigma^2
= Mu(i) + e(ij)
= Sigma(i)^2
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-212 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Model A3:
Yij
(ij ) }
Mu(i) + e(ij)
exp(lalpha + log(mean(
rho)
Model R: Yij
Var{e(ij)}
Mu + e(i
Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -47.23498 7 108.47
A2 -8.679256 12 41.35851
A3 -8.980651 8 33.9613
R -63.44829 2 130.8966
2 -19.6508 4 47.30161
Additive constant for all log-likelihoods = -54.22. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Does response and/or variances differ among Dose levels? (A2
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Does Model 2 fit the data? (A3 vs. 2)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 109.5 10 < 0.0001
Test 2 77.11 5 < 0.0001
Test 3 0.6028 4 0.9628
Test 4 21.34 4 0.0002711
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 4 is less than .1. Model 2 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 32.7092
BMDL = 2 6.1405
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-213 DRAFT—DO NOT CITE OR QUOTE
-------
H.3.9. Van Birgelen et al. (1995b): T4 UGT
H.3.9.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
4
<0.0001
35.16
<0.0001
38.57
5.0E+02
3.5E+02
nonconstant variance,
power restricted >1
exponential (M3)
4
<0.0001
35.16
<0.0001
38.57
5.0E+02
3.5E+02
nonconstant variance,
power restricted >1
exponential (M4)c
3
<0.0001
1.01
0.80
6.42
1.2E+01
6.2E+00
nonconstant variance,
power restricted >1
exponential (M5)
3
<0.0001
1.01
0.80
6.42
1.2E+01
6.2E+00
nonconstant variance,
power restricted >1
exponential (M5)d
3
<0.0001
1.01
0.80
6.42
1.2E+01
6.2E+00
nonconstant variance,
power unrestricted
Hill
2
<.0001
1.12
0.57
8.52
1.3E+01
6.1E+00
nonconstant variance, n
restricted >1
Hilld
2
<.0001
1.12
0.57
8.52
1.3E+01
3.7E+00
nonconstant variance, n
unrestricted
linear
4
<.0001
23.17
0.00
26.57
1.7E+02
9.4E+01
nonconstant variance
polynomial
4
<.0001
23.17
0.00
26.57
1.7E+02
9.4E+01
nonconstant variance
power
4
<.0001
23.17
0.00
26.57
1.7E+02
9.4E+01
nonconstant variance,
power restricted >1,
bound hit
power d
3
<.0001
5.05
0.17
10.45
4.0E+00
4.8E-01
nonconstant variance,
power unrestricted
exponential (M2)
4
<0.0001
14.22
0.01
39.63
6.1E+02
5.2E+02
constant variance,
power restricted >1
exponential (M3)
4
<0.0001
14.22
0.01
39.63
6.1E+02
5.2E+02
constant variance,
power restricted >1
exponential (M4)
3
<0.0001
0.16
0.98
27.56
8.8E+01
3.6E+01
constant variance,
power restricted >1
exponential (M5)
3
<0.0001
0.16
0.98
27.56
8.8E+01
3.6E+01
constant variance,
power restricted >1
exponential (M5)
3
<0.0001
0.16
0.98
27.56
8.8E+01
3.6E+01
constant variance,
power unrestricted
Hill
2
<.0001
0.10
0.95
29.51
7.1E+01
2.7E+01
constant variance, n
restricted >1
Hill
2
<.0001
0.10
0.95
29.51
7.1E+01
1.9E+01
constant variance, n
unrestricted
linear
4
<.0001
9.68
0.05
35.08
4.0E+02
3.0E+02
constant variance
polynomial
4
<.0001
9.68
0.05
35.08
4.0E+02
3.0E+02
constant variance
power
4
<.0001
9.68
0.05
35.08
4.0E+02
3.0E+02
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-214 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
3
<.0001
2.09
0.55
29.49
5.7E+01
9.9E+00
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.9.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
3.5
2.5
0.5
EMDL
BMD
0
200
400
600
800
1000
dose
14:33 11/20 2009
H.3.9.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_T4_UGT.(d)
Gnuplot Plotting File:
Fri Nov 20 14:33:47 2009
Tbl2, T4 UGT
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-215 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2: Y[dose]
Model 3: Y[dose]
Model 4: Y[dose]
Model 5: Y[dose]
= a * exp{sign *
= a * exp{sign *
= a * [c-(c — 1) *
= a * [c-(c — 1) *
b * dose}
(b * dose ) "M}
exp{-b * dose}]
exp{-(b * dosej^d}]
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -0.937573
rho 1.54913
a 0.3135
b 0.00297568
c 8.67464
d 1
Parameter Estimates
Variable Model 4
lnalpha -0.937201
rho 1.6967
a 0.294922
b 0.0100397
c 7.64822
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 0.33 0.198
14 8 0.6 0.4243
26 8 0.64 0.4525
47 8 0.87 0.9051
320 8 2.08 1.329
1024 8 2.59 0.8768
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-216 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0
0.2949
0.2221
0. 4466
14
0.552
0.3781
0.3589
26
0.7454
0. 4878
-0.6111
47
1. 032
0.6431
-0.7146
320
2 .177
1. 211
-0.2259
1024
2 . 256
1.248
0.758
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -9.701316 7 33.40263
A2 4.934967 12 14.13007
A3 2.296438 8 11.40712
R -29.51921 2 63.03841
4 1.790563 5 6.418874
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
68 . 91
29.27
5.277
1. 012
10
5
4
3
p-value
< 0.0001
< 0.0001
0.26
0.7984
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 6a is greater than .1. Model 4 seems
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-217 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 11.9766
BMDL = 6.23544
H.3.9.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
Exponential
3.5
2.5
0.5
EMDLBMD
0
200
400
600
800
1000
dose
14:33 11/20 2009
H.3.9.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_Unrest_BMRl_T4_UGT.(d)
Gnuplot Plotting File:
Fri Nov 20 14:33:53 2009
Tbl2, T4 UGT
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-218 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2: Y[dose]
Model 3: Y[dose]
Model 4: Y[dose]
Model 5: Y[dose]
= a * exp{sign *
= a * exp{sign *
= a * [c-(c — 1) *
= a * [c-(c — 1) *
b * dose}
(b * dose ) "M}
exp{-b * dose}]
exp{-(b * dosej^d}]
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -0.937573
rho 1.54913
a 0.3135
b 0.00297568
c 8.67464
d 1
Parameter Estimates
Variable Model 5
lnalpha -0.937201
rho 1.6967
a 0.294922
b 0.0100397
c 7.64822
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 8 0.33 0.198
14 8 0.6 0.4243
26 8 0.64 0.4525
47 8 0.87 0.9051
320 8 2.08 1.329
1024 8 2.59 0.8768
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-219 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
0
0.2949
0.2221
0. 4466
14
0.552
0.3781
0.3589
26
0.7454
0. 4878
-0.6111
47
1. 032
0.6431
-0.7146
320
2 .177
1. 211
-0.2259
1024
2 . 256
1.248
0.758
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)rt2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -9.701316 7 33.40263
A2 4.934967 12 14.13007
A3 2.296438 8 11.40712
R -29.51921 2 63.03841
5 1.790563 5 6.418874
Additive constant for all log-likelihoods = -44.11. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
68 . 91
29.27
5.277
1. 012
10
5
4
3
p-value
< 0.0001
< 0.0001
0.26
0.7984
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is greater than .1. Model 5 seems
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-220 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
to adequately describe the data.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 11.9766
BMDL = 6.23544
H.3.9.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model with 0.95 Confidence Level
Hill
3.5
2.5
0.5
BVIDL
BMD
0
200
400
600
800
1000
dose
14:33 11/20 2009
H.3.9.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_T4_UGT.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_Unrest_BMRl_T4_UGT.plt
Fri Nov 20 14:33:54 2009
Tbl2, T4 UGT
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-221 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
The variance is to be modeled as Var(i)
exp(lalpha + rho * ln(mean(i)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha
rho
intercept
-0.462247
0
0.33
2 .26
0. 430022
459.884
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
0. 036
-0.26
-0.16
-0.017
0. 037
rho
0. 036
1
0.48
-0.46
0. 02
-0.2
intercept
-0.26
0.48
1
-0.37
0.26
-0.15
v
-0.16
-0.46
-0.37
1
-0. 64
0. 81
n
-0.017
0. 02
0.26
-0. 64
1
-0. 85
k
0. 037
-0.2
-0.15
0. 81
-0. 85
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
-0.935113
1.68648
0.295265
2.14661
1.16336
80.2777
Std. Err.
0.256585
0.441197
0.0703668
0.547941
0.46393
52.4068
Lower Conf. Limit
-1.43801
0. 821746
0.157348
1. 07267
0.25407
-22 .4378
Upper Conf. Limit
-0.432217
2.55121
0.433181
3.22056
2.07264
182.993
Table of Data and Estimated Values of Interest
Dose
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
14
26
47
320
1024
0.33
0.6
0. 64
0. 87
2 . 08
2 .59
0.295
0.544
0.751
1. 04
2 . 08
2 . 34
0.198
0.424
0. 453
0. 905
1. 33
0 . 877
0.224
0.375
0. 492
0. 65
1.16
1.28
0.439
0. 422
-0.637
-0.76
-0.00947
0.56
Model Descriptions for likelihoods calculated
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-222 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2 : Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -9.701316 7 33.402631
A2 4.934967 12 14.130066
A3 2.296438 8 11.407124
fitted 1.738274 6 8.523453
R -29.519205 2 63.038411
Test
1
Test
2
Test
3
Test
4
(Note:
Explanation of Tests
Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
68.9083
29.2726
5.27706
1.11633
10
5
4
2
<.0001
<.0001
0.26
0.5723
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 12.6477
BMDL = 3.73502
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-223 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.3.9.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
3.5
2.5
0.5
BMD
0
200
400
600
800
1000
dose
14:33 11/20 2009
H.3.9.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_T4_UGT.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Pwr_Unrest_BMRl_T4_UGT.plt
Fri Nov 20 14:33:55 2009
Tbl2, T4 UGT
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 6
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-224 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
lalpha
rho
control
slope
power
-0.462247
0
0.33
0. 0542809
0.537973
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
control
slope
power
lalpha
1
0. 032
-0.26
-0.19
0. 071
rho
0. 032
1
0.57
0. 021
-0.19
control
-0.26
0.57
1
-0.23
0 . 077
slope
-0.19
0. 021
-0.23
1
-0. 94
power
0. 071
-0.19
0 . 077
-0. 94
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
control
slope
power
Estimate
-0.85465
1. 67517
0.275898
0.12137
0. 43322
Std. Err.
0.259915
0. 448857
0.0675474
0.0517127
0.0764873
Lower Conf. Limit
-1.36407
0.795428
0.143507
0.0200146
0.283308
Upper Conf. Limit
-0.345225
2.55492
0. 408288
0.583132
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
14
26
47
320
1024
0.33
0.6
0. 64
0. 87
2 . 08
2 .59
0.276
0. 657
0.774
0. 919
1 .75
2 .72
0.198
0.424
0. 453
0. 905
1. 33
0 . 877
0.222
0.459
0.526
0. 608
1. 04
1. 51
0.69
-0.349
-0.719
-0.229
0.886
-0.245
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i)} = Sigma/'2
Likelihoods of Interest
Model
Log(likelihood)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-225 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
A1
-9.701316
7
33.402631
A2
4 . 934 967
12
14 .130066
A3
2 .296438
8
11.407124
fitted
-0.226526
5
10.453053
R
-29.519205
2
63.038411
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Test
2
Test
3
Test
4
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 68.9083 10 <.0001
Test 2 29.2726 5 <.0001
Test 3 5.27706 4 0.26
Test 4 5.04593 3 0.1685
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is greater than .1. The modeled variance appears
to be appropriate here
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 4.02257
BMDL = 0.480637
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-226 DRAFT—DO NOT CITE OR QUOTE
-------
H.3.10. Van Birgelen et al. (1995b): UGT 1A1
H.3.10.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
3
0.00
29.54
<0.0001
167.34
7.4E+02
2.1E+02
nonconstant variance,
power restricted >1
exponential (M3)
3
0.00
29.54
<0.0001
167.34
7.4E+02
2.1E+02
nonconstant variance,
power restricted >1
exponential (M4)c
2
0.00
1.28
0.53
141.08
1.6E+00
8.5E-01
nonconstant variance,
power restricted >1
exponential (M5)
2
0.00
1.28
0.53
141.08
1.6E+00
8.5E-01
nonconstant variance,
power restricted >1
exponential (M5)d
2
0.00
1.28
0.53
141.08
1.6E+00
8.5E-01
nonconstant variance,
power unrestricted
Hill
1
0.00
1.31
0.25
143.11
1.8E+00
error
nonconstant variance, n
restricted >1
Hilld
1
0.00
1.31
0.25
143.11
1.8E+00
error
nonconstant variance, n
unrestricted
linear
3
0.00
27.83
<.0001
165.63
2.1E+02
5.8E+01
nonconstant variance
polynomial
3
0.00
30.93
<.0001
168.73
1.8E+03
2.9E+01
nonconstant variance
power
3
0.00
27.83
<.0001
165.63
2.1E+02
5.8E+01
nonconstant variance,
power restricted >1,
bound hit
power d
2
0.00
5.39
0.07
145.19
3.4E-03
3.4E-03
nonconstant variance,
power unrestricted
exponential (M2)
3
0.00
22.45
<0.0001
165.95
1.3E+03
6.4E+02
constant variance,
power restricted >1
exponential (M3)
3
0.00
22.45
<0.0001
165.95
1.3E+03
6.4E+02
constant variance,
power restricted >1
exponential (M4)
2
0.00
7.89
0.02
153.38
1.1E+01
4.7E+00
constant variance,
power restricted >1
exponential (M5)
2
0.00
7.89
0.02
153.38
1.1E+01
4.7E+00
constant variance,
power restricted >1
exponential (M5)
2
0.00
7.89
0.02
153.38
1.1E+01
4.7E+00
constant variance,
power unrestricted
Hill
1
0.00
8.15
0.00
155.65
1.3E+01
3.0E+00
constant variance, n
restricted >1
Hill
1
0.00
8.15
0.00
155.65
1.3E+01
1.9E+00
constant variance, n
unrestricted
linear
3
0.00
22.15
<.0001
165.65
1.1E+03
4.9E+02
constant variance
polynomial
3
0.00
22.15
<.0001
165.65
1.1E+03
4.9E+02
constant variance
power
3
0.00
22.15
<.0001
165.65
1.1E+03
4.9E+02
constant variance,
power restricted >1,
bound hit
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-227 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Model
Degrees
of
Freedom
Variance
/>-Value3
x2 Test
Statistic
x2 p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
power
2
0.00
12.90
0.00
158.40
1.6E-01
5.2E-06
constant variance,
power unrestricted
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
dAltemate model also presented in this appendix
H.3.10.2. Figure for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 4 with 0.95 Confidence Level
Exponential
700
600
500
400
300
200
100
-100
0
200
400
600
800
1000
dose
14:34 11/20 2009
H.3.10.3. Output File for Selected Model: Exponential (M4), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_UGT_lAl.(d)
Gnuplot Plotting File:
Fri Nov 20 14:34:36 2009
Tbl2, UGT_1A1
The form of the response function by Model:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-228 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dosej^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dosej^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 4
lnalpha -1.53604
rho 1.59958
a 95.95
b 0.00148532
c 4.94633
d 1
Parameter Estimates
Variable Model 4
lnalpha -10.1636
rho 3.25851
a 101.863
b 0.0256373
c 3.78343
d 1
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 3 101 15.59
14 3 194 36.37
47 3 304 17.32
320 3 452 48.5
1024 3 296 149
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 101.9 11.6 -0.1288
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-229 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
14 187.4 31.32 0.3668
47 300.4 67.58 0.09183
320 385.3 101.4 1.139
1024 385.4 101.4 -1.527
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(1j)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(1j)
Var{e(ij)} = exp(lalpha + log(mean(i)) ^ rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -68.74833 6 149.4967
A2 -58.69126 10 137.3825
A3 -64.89907 7 143.7981
R -80.72265 2 165.4453
4 -65.54073 5 141.0815
Additive constant for all log-likelihoods = -13.78. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 6a: Does Model 4 fit the data? (A3 vs 4)
Test
Test 1
Test 2
Test 3
Test 6a
Tests of Interest
-2*log(Likelihood Ratio)
44.06
20.11
12 .42
1.283
D. F.
p-value
< 0.0001
0.0004741
0.006087
0.5264
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 6a is greater than .1. Model 4 seems
to adeguately describe the data.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-230 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1.62983
BMDL = 0.853335
H.3.10.4. Figure for Unrestricted Model: Exponential (M5), Nonconstant Variance, Power
Unrestricted
Exponential_beta Model 5 with 0.95 Confidence Level
700
600
500
400
300
200
100
-100
Exponential
4
l/IDL
BMD
200
400
600
800
1000
dose
14:34 11/20 2009
H.3.10.5. Output File for Unrestricted Model: Exponential (M5), Nonconstant Variance,
Power Unrestricted
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_Unrest_BMRl_UGT_lAl.(d)
Gnuplot Plotting File:
Fri Nov 20 14:34:44 2009
Tbl2, UGT 1A1
The form of the response function by Model:
Model 2
Model 3
Model 4
Y[dose]
Y[dose]
Y[dose]
exp{sign
exp{sign
[c—(c—1)
b * dose}
(b * dose)^d}
exp{-b * dose}
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-231 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Model 5: Y[dose] = a * [c-(c-l) * exp{-(b * dosej^d}]
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -1.53604
rho 1.59958
a 95.95
b 0.00148532
c 4.94633
d 1
Parameter Estimates
Variable Model 5
lnalpha -10.1636
rho 3.25851
a 101.863
b 0.0256373
c 3.78343
d 1
Table of Stats From Input Data
N Obs Mean Obs Std Dev
0
3
101
15.59
14
3
194
36.37
47
3
304
17 . 32
320
3
452
48.5
1024
3
296
149
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0 101.9 11.6 -0.1288
14 187.4 31.32 0.3668
47 300.4 67.58 0.09183
320 385.3 101.4 1.139
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-232 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
1024 385.4 101.4 -1.527
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/N2
Model A2 : Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma/N2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -68.74833 6 149.4967
A2 -58.69126 10 137.3825
A3 -64.89907 7 143.7981
R -80.72265 2 165.4453
5 -65.54073 5 141.0815
Additive constant for all log-likelihoods = -13.78. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1: Does response and/or variances differ among Dose levels? (A2 vs. R)
Test 2: Are Variances Homogeneous? (A2 vs. A1)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Test
Test 1
Test 2
Test 3
Test 7a
Tests of Interest
-2*log(Likelihood Ratio)
44.06
20.11
12 .42
1.283
p-value
< 0.0001
0.0004741
0.006087
0.5264
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is less than .1. You may want to
consider a different variance model.
The p-value for Test 7a is greater than .1. Model 5 seems
to adeguately describe the data.
Benchmark Dose Computations:
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-233 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 1.62983
BMDL = 0.853335
H.3.10.6. Figure for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model
700
600
500
400
300
200
100
-100
200
400 600
dose
800
1000
14:34 11/20 2009
H.3.10.7. Output File for Unrestricted Model: Hill, Nonconstant Variance, n Unrestricted
Hill Model. (Version: 2.14; Date: 06/26/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Hill_Unrest_BMRl_UGT_lAl.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Hill_Unrest_BMRl_UGT_lAl.plt
Fri Nov 20 14:34:45 2009
Tbl2, UGT_1A1
The form of the response function is:
Y[dose] = intercept + v*doseAn/(kAn + doseAn)
Dependent variable = Mean
Independent variable = Dose
Power parameter is not restricted
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-234 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
The variance is to be modeled as Var(i) = exp(lalpha + rho * ln(mean(i)))
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
lalpha
rho
intercept
8 . 57191
0
101
351
0.273231
55.25
Asymptotic Correlation Matrix of Parameter Estimates
lalpha
rho
intercept
lalpha
1
-0. 99
-0.19
0.12
0.12
-0.017
rho
-0. 99
1
0.18
-0.16
-0.12
0. 0049
intercept
-0.19
0.18
1
-0.11
0. 031
0. 096
v
0.12
-0.16
-0.11
1
-0. 61
0. 82
n
0.12
-0.12
0. 031
-0. 61
1
-0.76
k
-0.017
0.0049
0. 096
0. 82
-0.76
1
Parameter Estimates
95.0% Wald Confidence Interval
Variable
lalpha
rho
intercept
Estimate
-10.516
3.32204
101.644
298.646
1.1568
29.0772
Std. Err.
3. 69809
0. 675078
6.48157
56.3141
0.50134
13.7717
Lower Conf. Limit
-17.7641
1.99891
88 . 9405
188.273
0.17419
2.08512
Upper Conf. Limit
-3.26786
4.64517
114 . 348
409.02
2 .13941
56.0693
Table of Data and Estimated Values of Interest
Obs Mean
Est Mean Obs Std Dev Est Std Dev Scaled Res.
0
14
47
320
1024
101
194
304
452
296
102
191
291
383
396
15. 6
36. 4
17 . 3
48.5
149
11. 2
32 .1
64 . 6
102
107
-0.0994
0.143
0.338
1.18
-1. 61
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma/'2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-235 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Var{e(ij)} = Sigma(i)/N2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma/'2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -68.748326 6 149.496653
A2 -58.691256 10 137.382511
A3 -64.899072 7 143.798144
fitted -65.554216 6 143.108432
R -80.722651 2 165.445302
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2
Test 3
Test 4
Are Variances Homogeneous? (A1 vs A2)
Are variances adequately modeled? (A2 vs. A3)
Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test
-2*log(Likelihood Ratio) Test df
p-value
Test 1
Test 2
Test 3
Test 4
44 . 0628
20.1141
12 .4156
1.31029
<.0001
0.0004741
0.006087
0.2523
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is greater than .1. The model chosen seems
to adequately describe the data
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 1 .76282
BMDL computation failed.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-236 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.3.10.8. Figure for Unrestricted Model: Power, Nonconstant Variance, Power Unrestricted
Power Model with 0.95 Confidence Level
Power
700
600
500
400
300
200
100
-100
0
200
400
600
800
1000
dose
14:34 11/20 2009
H.3.10.9. Output File for Unrestricted Model: Power, Nonconstant Variance, Power
Unrestricted
Power Model. (Version: 2.15; Date: 04/07/2008)
Input Data File: C:\USEPA\BMDS21\Nov20\Pwr_Unrest_BMRl_UGT_lAl.(d)
Gnuplot Plotting File: C:\USEPA\BMDS21\Nov2 0\Pwr_Unrest_BMRl_UGT_lAl.plt
Fri Nov 20 14:34:47 2009
Tbl2, UGT_1A1
The form of the response function is:
Y[dose] = control + slope * doseApower
Dependent variable = Mean
Independent variable = Dose
The power is not restricted
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 5
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Default Initial Parameter Values
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-237 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
lalpha = 8.57191
rho = 0
control = 101
slope = 7 5.1984
power = 0.19277
Asymptotic Correlation Matrix of Parameter Estimates
lalpha rho control slope power
lalpha 1 -0.99 -0.22 0.054 0.018
rho -0.99 1 0.2 -0.038 -0.052
control -0.22 0.2 1 -0.2 0.11
slope 0.054 -0.038 -0.2 1 -0.95
power 0.018 -0.052 0.11 -0.95 1
Parameter Estimates
Variable
lalpha
rho
control
slope
power
Estimate
-11. 5264
3.53579
101.425
53.9904
0.279427
Std. Err.
3.45692
0.629031
6.34917
20.7283
0.0834726
95.0% Wald Confidence Interval
Lower Conf. Limit Upper Conf. Limit
-18.3019
2.30291
88.981
13.3638
0.115823
-4 .75098
4 .76867
113.869
94.617
0.44303
Table of Data and Estimated Values
Dose N Obs Mean Est Mean
0 3 101 101
14 3 194 214
47 3 304 260
320 3 452 372
1024 3 296 476
of Interest
Obs Std Dev Est Std Dev Scaled Res.
15.6 11.1 -0.0666
36.4 41.5 -0.847
17.3 58.3 1.31
48.5 110 1.26
149 170 -1.83
Model Descriptions for likelihoods calculated
Model A1: Yij = Mu(i) + e(iji
Var{e(ij)} = Sigma'" 2
Model A2: Yij = Mu(i) + e(ij|
Var{e(ij)} = Sigma(i)^2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + rho*ln(Mu(i)))
Model A3 uses any fixed variance parameters that
were specified by the user
Model R: Yi = Mu + e(i)
Var{e(i) } = Sigma'" 2
Likelihoods of Interest
Model Log(likelihood) # Param's AIC
A1 -68.748326 6 149.496653
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-238 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
A2 -58.691256 10 137.382511
A3 -64.899072 7 143.798144
fitted -67.596085 5 145.192170
R -80.722651 2 165.445302
Explanation of Tests
Test 1: Do responses and/or variances differ among Dose levels?
(A2 vs. R)
Test 2: Are Variances Homogeneous? (A1 vs A2)
Test 3: Are variances adeguately modeled? (A2 vs. A3)
Test 4: Does the Model for the Mean Fit? (A3 vs. fitted)
(Note: When rho=0 the results of Test 3 and Test 2 will be the same.
Tests of Interest
Test -2*log(Likelihood Ratio) Test df p-value
Test 1 44.0628 8 <.0001
Test 2 20.1141 4 0.0004741
Test 3 12.4156 3 0.006087
Test 4 5.39403 2 0.06741
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose levels
It seems appropriate to model the data
The p-value for Test 2 is less than .1. A non-homogeneous variance
model appears to be appropriate
The p-value for Test 3 is less than .1. You may want to consider a
different variance model
The p-value for Test 4 is less than .1. You may want to try a different
model
Benchmark Dose Computation
Specified effect = 1
Risk Type = Estimated standard deviations from the control mean
Confidence level = 0.95
BMD = 0.00343319
BMDL = 0.00343312
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-239 DRAFT—DO NOT CITE OR QUOTE
-------
H.3.11. Vanden Heuvel et al. (1994): Hepatic CYP1A1 mRNA Expression
H.3.11.1. Summary Table of BMDS Modeling Results
Model
Degrees
of
Freedom
Variance
/>-Value3
x 2 Test
Statistic
x2p-
Value"
AIC
BMD
(ng/kg-
day)
BMDL
(ng/kg-
day)
Model Notes
exponential (M2)
5
<0.0001
568.80
<0.0001
1164.38
4.7E+03
1.7E+03
nonconstant variance,
power restricted >1
exponential (M3)
5
<0.0001
568.80
<0.0001
1164.38
4.7E+03
1.7E+03
nonconstant variance,
power restricted >1
exponential (M4)
4
<0.0001
63.39
<0.0001
661.01
4.5E-01
2.6E-01
nonconstant variance,
power restricted >1
exponential (M5)c
3
<0.0001
35.71
<0.0001
635.33
1.5E+01
1.0E+01
nonconstant variance,
power restricted >1
Hill
3
<.0001
33.98
<.0001
633.59
1.9E+01
error
nonconstant variance, n
restricted >1
linear
5
<.0001
71.94
<.0001
667.55
5.0E-01
3.1E-01
nonconstant variance
polynomial
5
<.0001
137.66
<.0001
733.28
5.4E+03
1.7E+01
nonconstant variance
power
4
<.0001
71.83
<.0001
669.44
5.6E-01
3.2E-01
nonconstant variance,
power restricted >1
exponential (M2)
5
<0.0001
27.93
<0.0001
1178.88
5.9E+03
5.1E+03
constant variance,
power restricted >1
exponential (M3)
5
<0.0001
27.93
<0.0001
1178.88
5.9E+03
5.1E+03
constant variance,
power restricted >1
exponential (M4)
4
<0.0001
0.34
0.99
1153.28
4.0E+02
2.8E+02
constant variance,
power restricted >1
exponential (M5)
3
<0.0001
0.00
1.00
1154.95
5.7E+02
2.9E+02
constant variance,
power restricted >1
Hill
3
<.0001
0.00
1.00
1154.95
5.3E+02
2.1E+02
constant variance, n
restricted >1
linear
5
<.0001
21.45
0.00
1172.40
2.8E+03
2.2E+03
constant variance
polynomial
5
<.0001
22.53
0.00
1173.48
3.1E+03
2.1E+03
constant variance
power
5
<.0001
21.45
0.00
1172.40
2.8E+03
2.2E+03
constant variance,
power restricted >1,
bound hit
aValues <0.1 means nonconstant variance model should be selected; Values >0.1 means a constant variance model should be selected
bValues <0.1 fail to meet BMDS goodness-of-fit criteria
cBest-fitting model as assessed by lowest-AIC criterion, bolded, presented in this appendix
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-240 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
H.3.11.2. Figure for Selected Model: Exponential (M5), Nonconstant Variance, Power
Restricted >1
Exponential_beta Model 5 with 0.95 Confidence Level
70000
Exponential
60000
50000
40000
30000
20000
10000
0
0
2000
4000
6000
8000
10000
dose
16:52 11/20 2009
H.3.11.3. Output File for Selected Model: Exponential (M5), Nonconstant Variance, Power
Restricted >1
Exponential Model. (Version: 1.5; Date: 4/23/2009)
Input Data File: C:\USEPA\BMDS21\Nov20\Exp_BMRl_hepatic_CYPlAl_mRNA_expression.(d)
Gnuplot Plotting File:
Fri Nov 20 16:52:21 2009
[insert study notes]
The form of the response function by Model:
Model 2
Model 3
Model 4
Model 5
Y[dose]
Y[dose]
Y[dose]
Y[dose]
exp{sign * b * dose}
exp{sign * (b * dose)^d}
[c-(c-l) * exp{-b * dose}]
[c-(c-l) * exp{-(b * dose)^d}
Note: Y[dose] is the median response for exposure = dose;
sign = +1 for increasing trend in data;
sign = -1 for decreasing trend.
Model 2 is nested within Models 3 and 4.
Model 3 is nested within Model 5.
Model 4 is nested within Model 5.
Dependent variable = Mean
Independent variable = Dose
Data are assumed to be distributed: normally
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-241 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Variance Model: exp(lnalpha +rho *ln(Y[dose]))
The variance is to be modeled as Var(i) = exp(lalpha + log(mean(i)) * rho)
Total number of dose groups = 7
Total number of records with missing values = 0
Maximum number of iterations = 250
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
MLE solution provided: Exact
Initial Parameter Values
Variable Model 5
lnalpha -0.89532
rho 2.01401
a 5.13
b 0.000307638
c 7511.7
d 1
Parameter Estimates
Variable Model 5
lnalpha
rho
0.176234
1.90467
9.74751
0.00106447
3247.52
1.96414
Table of Stats From Input Data
Dose N Obs Mean Obs Std Dev
0 13
0.1 5
1 12
10 7
100 7
1000 11
le+004 5
Estimated Values of Interest
Dose Est Mean Est Std Scaled Residual
0
9.748
9.551
-1.641
0.1
9.748
9.551
-0.5965
1
9.793
9.593
1.808
10
13. 97
13. 45
-0.2296
100
395. 9
325.2
1.14
1000
2 .144e + 004
1. 456e + 004
-0.7835
le+004
3 .166e + 004
2 . lle + 004
0.5347
5.4 3.606
7.2 5.59
14.8 14.9
12.8 4.498
536 320.1
1.8e+004 1.522e+004
3.67e+004 2.214e+004
Other models for which likelihoods are calculated:
Model A1: Yij = Mu(i) + e(ij)
Var{e(ij)} = Sigma^2
Model A2: Yij = Mu(i) + e(ij)
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-242 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
Var{e(ij)} = Sigma(i)/X2
Model A3: Yij = Mu(i) + e(ij)
Var{e(ij)} = exp(lalpha + log(mean(i)) * rho)
Model R: Yij = Mu + e(i)
Var{e(ij)} = Sigma^2
Likelihoods of Interest
Model Log(likelihood) DF AIC
A1 -572.4744 8 1160.949
A2 -290.7965 14 609.5929
A3 -293.806 9 605.6119
R -603.6646 2 1211.329
5 -311.6633 6 635.3266
Additive constant for all log-likelihoods = -55.14. This constant added to the
above values gives the log-likelihood including the term that does not
depend on the model parameters.
Explanation of Tests
Test 1
Test 2
Test 3
Does response and/or variances differ among Dose levels? (A2 vs. R)
Are Variances Homogeneous? (A2 vs. A1)
Are variances adequately modeled? (A2 vs. A3)
Test 7a: Does Model 5 fit the data? (A3 vs 5)
Tests of Interest
Test -2*log(Likelihood Ratio) D. F. p-value
Test 1 625.7 12 < 0.0001
Test 2 563.4 6 < 0.0001
Test 3 6.019 5 0.3044
Test 7a 35.71 3 < 0.0001
The p-value for Test 1 is less than .05. There appears to be a
difference between response and/or variances among the dose
levels, it seems appropriate to model the data.
The p-value for Test 2 is less than .1. A non-homogeneous
variance model appears to be appropriate.
The p-value for Test 3 is greater than .1. The modeled
variance appears to be appropriate here.
The p-value for Test 7a is less than .1. Model 5 may not adequately
describe the data; you may want to consider another model.
Benchmark Dose Computations:
Specified Effect = 1.000000
Risk Type = Estimated standard deviations from control
Confidence Level = 0.950000
BMD = 15.1574
BMDL = 10.4625
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-243 DRAFT—DO NOT CITE OR QUOTE
-------
1 H.4. REFERENCES
2 Hassoun, EA; Li, F; Abushaban, A; et al. (2000) The relative abilities of TCDD and its congeners to induce
3 oxidative stress in the hepatic and brain tissues of rats after subchronic exposure. Toxicology 145:103-113.
4 NTP (National Toxicology Program). (2006) Studies of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) in female
5 Harlan Sprague-Dawley rats (gavage studies) Tech. Rep. Ser. No. 521. U.S. Department of Health and Human
6 Services, Public Health Service, Research Triangle Park, NC.
7 Van Birgelen et al. 1995a and 1995b -1 assume below is one of them?
8 Van Birgelen, AP; Van der Kolk, J; Fase, KM; et al. (1995) Subchronic dose-response study of 2,3,7,8-
9 tetrachlorodibenzo-p-dioxin in female Sprague-Dawley rats. Toxicol Appl Pharmacol 132:1-13.
10 van den Heuvel, JP; Clark, GC; Kohn, MC; et al. (1994) Dioxin-responsive genes: examination of dose-response
11 relationships using quantitative reverse transcriptase-polymerase chain reaction. Cancer Res 54:62-68.
This document is a draft for review purposes only and does not constitute Agency policy.
1/15/10 H-244 DRAFT—DO NOT CITE OR QUOTE
------- |