EPA/63 5/R-l 6/3 50Fa
www.epa.gov/iris
Evaluation of the Inhalation Carcinogenicity of
Ethylene Oxide
(CASRN 75-21-8)
In Support of Summary Information on the
Integrated Risk Information System (IRIS)
December 2016
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Washington, DC
-------
DISCLAIMER
This document has been reviewed in accordance with U.S. Environmental Protection
Agency policy and approved for publication. Mention of trade names or commercial
products does not constitute endorsement or recommendation for use.
11
-------
PREFACE
The purpose of this document is to provide scientific support and rationale for the hazard and
dose-response assessment in the Integrated Risk Information System (IRIS) pertaining to
carcinogenicity from chronic inhalation exposure to ethylene oxide (EtO).
In its 2011 review of the U. S. Environmental Protection Agency's (EPA's) draft IRIS assessment
of formaldehyde, the National Research Council (NRC) provided a number of recommendations related
to the IRIS program. In the report, the NRC encouraged the EPA to continue with assessments in
progress while also developing improvements to the IRIS program. Consistent with this advice, the
EPA indicated that it would not go backwards in the assessment development process, but would focus
on moving forward, while also phasing in improvements.
The EtO assessment was one of a group of chemical assessments for which an external peer
review had already been completed at the time the 2011 NRC recommendations were released. For this
group of assessments, the EPA focused on applying some of the short-term NRC
recommendations. Thus, the EtO assessment does not incorporate recent revisions to the IRIS
assessment format, such as the inclusion of a standard Preamble and the revised chapter structure, and it
does not fully implement the longer term NRC recommendations.
The assessment implements a number of other NRC recommendations. For example, the
assessment is streamlined and uses tables, figures, and appendices to increase transparency and
clarity. The assessment is also structured to have separate hazard identification and dose-response
sections. In addition, an updated literature search was conducted using systematic
approaches. Furthermore, consistent with the goal that assessments should provide a scientifically
sound and transparent evaluation of the relevant scientific literature and presentation of the analyses
performed, the assessment contains an expanded discussion on the rationales for study evaluation and
selection, as well as for other key assessment decisions.
In 2013, the IRIS program implemented a suite of enhancements to the IRIS process, one of
which included separating the public comment period from the external peer-review period. This
enhancement allowed the IRIS program to consider public comments and make revisions as needed
before releasing a document for peer review. Thus, current and future assessments will include the
response to public comments as an appendix in the external review draft sent for peer review, and then
include a response to peer-review comments in the final assessment. Given the complex history of the
EtO assessment, which had multiple public comment periods and peer reviews, the EPA is including all
of the responses to comments (both public and peer review) in the appendices of this final
assessment. More details on the development of the assessment can be found in the assessment
introduction in Chapter 2.
in
-------
CONTENTS
LIST OF TABLES vi
LIST OF FIGURES ix
LIST 01 ABBREVIATIONS x
AUTHORS, CONTRIBUTORS, AND REVIEWERS xii
1. EXECUTIVE SUMMARY 1-1
2. INTRODUCTION 2-1
2.1. LITERATURE IDENTIFICATION 2-2
2.2. NATIONAL RESEARCH COUNCIL RECOMMENDATIONS OF 2011 2-3
3. HAZARD IDENTIFICATION 3-1
3.1. EVIDENCE 01 CANCER IN HUMANS 3-1
3.1.1. Conclusions Regardingthe Evidence of Cancer in Humans 3-12
3.2. EVIDENCE OF CANCER IN LABORATORY ANIMALS 3-20
3.2.1. Conclusions Regardingthe Evidence of Cancer in Laboratory Animals 3-23
3.3. SUPPORTING EVIDENCE 3-23
3.3.1. Metabolism and Kinetics 3-23
3.3.2. Protein Adducts 3-26
3.3.3. Genotoxicity 3-28
3.4. MODE OF ACTION 3-59
3.4.1. Possible Mechanisms for Mutagenic Mode of Action 3-59
3.4.2. Evidence and Possible Mechanisms for Alternative Modes of Action 3-64
3.4.3. Analysis of the Mode of Action for Ethylene Oxide Carcinogenicity under the
EPA's Mode-of-Action Framework 3-64
3.5. HAZARD CHARACTERIZATION 3-67
3.5.1. Characterization of Cancer Hazard 3-67
3.5.2. Susceptible Life Stages and Populations 3-71
4. CANCER DOSE-RESPONSE ASSESSMENT FOR INHALATION EXPOSURE 4-1
4.1. INHALATION UNIT RISK ESTIMATES DERIVED FROM HUMAN DATA 4-3
4.1.1. Risk Estimates for Lymphohematopoietic Cancer 4-6
4.1.2. Risk Estimates for Breast Cancer 4-32
4.1.3. Total Cancer Risk Estimates 4-57
4.1.4. Sources of Uncertainty in the Human-Data-Based Cancer Risk Estimates... 4-59
4.1.5. Summary of Unit Risk Estimates Derived from Human Data 4-75
4.2. INHALATION UNIT RISK DERIVED FROM laboratory ANIMAL DATA 4-76
4.2.1. Overall Approach 4-76
4.2.2. Cross-Species Scaling 4-76
4.2.3. Dose-Response Modeling Methods 4-78
4.2.4. Description of Laboratory Animal Studies 4-80
4.2.5. Results of Data Analysis of Laboratory Animal Studies 4-81
4.3. SUMMARY OF INHALATION UNIT RISK ESTIMATES—NOT ACCOUNTING
FOR ASSUMED INCREASED EARLY-LIFE SUSCEPTIBILITY 4-83
4.4. ADJUSTMENTS FOR POTENTIAL INCREASED EARLY-LIFE
SUSCEPTIBILITY 4-85
4.5. INHALATION UNIT RISK ESTIMATES—CONCLUSIONS 4-90
4.6. COMPARISON WITH OTHER PUBLISHED RISK ESTIMATES 4-97
4.6.1. Unit Risk Estimates Based on Human Studies 4-97
iv
-------
4.6.2. Unit Risk Estimates Based on Laboratory Animal Studies 4-99
4.7. RISK ESTIMATES FOR SOME OCCUPATIONAL EXPOSURE SCENARIOS . 4-99
4.7.1. Extra Risk Estimates for Lymphoid Cancer 4-100
4.7.2. Extra Risk Estimates for Breast Cancer 4-104
4.7.3. Extra Risk Estimates for Total Cancer 4-108
4.7.4. Calculation of Extra Risk Estimates for Other Occupational Exposure
Scenarios 4-110
5. REFERENCES R-l
APPENDIX A. CRITICAL REVIEW OF EPIDEMIOLOGIC EVIDENCE A-l
APPENDIX B. REFERENCES I OR FIGURE 3-3 B-l
APPENDIX C. GENOTOXICITY AND MUTAGENICITY OF ETHYLENE OXIDE C-l
APPENDIX D. REANALYSES OF ETHYLENE OXIDE EXPOSURE-RESPONSE DATA. D-l
APPENDIX E. LIFE-TABLE ANALYSIS E-l
APPENDIX F. EQUATIONS USED FOR WEIGHTED LINEAR REGRESSION F-l
APPENDIX G. MODEL PARAMETERS IN THE ANALYSIS OF ANIMAL TUMOR
INCIDENCE G-l
APPENDIX H. SUMMARY OF 2007 EXTERNAL PEER REVIEW AND PUBLIC
COMMENTS AND DISPOSITION II-l
APPENDIX I. EPA RESPONSES TO SAB COMMENTS ON 2014 EXTERNAL REVIEW
DRAFT 1-1
APPENDIX J. SUMMARY OF MAJOR NEW STUDIES SINCE THE LITERATURE
CUTOFF DATE J-l
APPENDIX K. SUMMARY OF PUBLIC COMMENTS RECEIVED ON THE JULY 2013
PUBLIC COMMENT DRAFT AND EPA RESPONSES K-l
v
-------
LIST OF TABLES
Table 1-1. Summary of major findings 1-7
Table 3-1. Epidemiological studies of ethylene oxide and human
cancer—lymphohematopoietic cancer results 3-14
Table 3-2. Summary of epidemiological results on ethylene oxide and female breast
cancer (all sterilizer workers) 3-18
Table 3-3. Tumor incidence data in National Toxicology Program Study of
B6C3Fi mice (NTP, 1987) and exposure-response modeling results 3-21
Table 3-4. Tumor incidence data in Lynch et al. (1984a, c) study of male F344 rats and
exposure-response modeling results 3-22
Table 3-5. Tumor incidence data in Snellings et al. (1984) and Garman et al. (1985)
reports on F344 rats and exposure-response modeling results 3-24
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and
mutations in humans and laboratory animals 3-33
Table 3-7. Ethylene oxide-induced cytogenetic effects in laboratory animals 3-47
Table 3-8. Ethylene oxide-induced cytogenetic effects in humans 3-51
Table 3-9. Summary of exposure and duration patterns for ethylene oxide -induced DNA
adducts and mutations in humans and laboratory animals (inhalation studies) 3-57
Table 3-10. Summary of exposure and duration patterns for ethylene oxide-induced
cytogenetic effects in humans and laboratory animals (inhalation studies) 3-58
Table 4-1. Considerations used in this assessment for selecting epidemiology studies for
quantitative risk estimation 4-5
Table 4-2. Cox regression results for all lymphohematopoietic cancer and lymphoid
cancer mortality in both sexes in the National Institute for Occupational Safety
and Health cohort, for the models presented by Steenland et al. (2004) 4-7
Table 4-3. Linear regression of categorical results—modeling results for all
lymphohematopoietic cancer and lymphoid cancer mortality in both sexes in the
National Institute for Occupational Safety and Health cohort 4-14
Table 4-4. Two-piece spline modeling results for lymphoid cancer and all
lymphohematopoietic cancer mortality in both sexes in the National Institute for
Occupational Safety and Health cohort 4-17
vi
-------
Table 4-5. Exposure-response modeling results for lymphoid cancer mortality in both
sexes in the National Institute for Occupational Safety and Health cohort for
models with square-root transformations of exposure 4-18
Table 4-6. Models considered for modeling the exposure-response data for lymphoid
cancer mortality in both sexes in the National Institute for Occupational Safety
and Health cohort for the derivation of unit risk estimates 4-19
Table 4-7. ECoi, LECoi, and unit risk estimates for lymphoid cancer from various
models 4-23
Table 4-8. ECoi, LECoi, and unit risk estimates for all lymphohematopoietic cancer 4-29
Table 4-9. Cox regression results for breast cancer mortality in females in the National
Institute for Occupational Safety and Health cohort, for models presented in
Steenland et al. (2004) 4-33
Table 4-10. Exposure-response modeling results for breast cancer mortality in females
in the National Institute for Occupational Safety and Health cohort for models not
presented by Steenland et al. (2004) 4-37
Table 4-11. ECoi, LECoi, and unit risk estimates for breast cancer mortality in females 4-38
Table 4-12. Cox regression results for breast cancer incidence in females from the
National Institute for Occupational Safety and Health cohort, for the models
presented by Steenland et al. (2003) 4-40
Table 4-13. Exposure-response modeling results for breast cancer incidence in females
from the National Institute for Occupational Safety and Health cohort for models
not presented by Steenland et al. (2003) 4-47
Table 4-14. Models considered for modeling the exposure-response data for breast
cancer incidence in females in the subcohort with interviews from the National
Institute for Occupational Safety and Health incidence study cohort for the
derivation of unit risk estimates 4-50
Table 4-15. ECoi, LECoi, and unit risk estimates for breast cancer incidence (invasive
and in situ) in females from various models 4-53
Table 4-16. Calculation of ECoi for total cancer risk 4-58
Table 4-17. Calculation of total cancer unit risk estimate 4-58
Table 4-18. Upper-bound unit risks (per (J,g/m3) obtained by combining tumor sites 4-80
Table 4-19. Unit risk values from multistage Weibull time-to-tumor modeling of mouse
tumor incidence in the NTP (1987) study 4-82
vii
-------
Table 4-20. Summary of unit risk estimates (per (J,g/m3) in animal bioassays 4-83
Table 4-21. ECoi, LECoi, and unit risk estimates for adult-only exposures 4-86
Table 4-22. Calculation of ECoi for total cancer (incidence) risk from adult-only
exposure 4-87
Table 4-23. Calculation of total cancer unit risk estimate from adult-only exposure 4-87
Table 4-24. Adult-based unit risk estimates for use in calculations involving
age-dependent adjustment factors and less-than-lifetime exposure scenarios 4-89
Table 4-25. Adult-based extra risk estimates per ppm based on adult-exposure-only
ECois (0.01/ECoi estimates) 4-94
Table 4-26. Summary of key unit risk estimates from this assessment (see Section 4.7
for risk estimates based on occupational exposure scenarios) 4-96
Table 4-27. Comparison of unit risk estimates 4-98
Table 4-28. Extra risk estimates for lymphoid cancer in both sexes for various
occupational exposure levels 4-102
Table 4-29. Extra risk estimates for breast cancer incidence in females for various
occupational exposure levels 4-106
Table 4-30. Extra risk estimates for total cancer incidence for various occupational
exposure levels 4-109
viii
-------
LIST OF FIGURES
Figure 3-1. Metabolism of ethylene oxide 3-25
Figure 3-2. Simulated blood areas under the curve for EtO following a 6-hour exposure
to EtO from the rat, mouse, and human physiologically based pharmacokinetic
models of Fennell and Brown (2001) 3-27
Figure 3-3. Display of 203 data sets, including bacteria, fungi, plants, insects, and
mammals (in vitro and in vivo), measuring the full range of genotoxic endpoints 3-29
Figure 4-1. Flowchart of key steps in the derivation of the unit risk estimate from human
data. Rectangles depict preferred options; ovals depict comparison analyses 4-2
Figure 4-2. RR estimate for lymphoid cancer vs. occupational cumulative exposure (with
15-year lag) 4-11
Figure 4-3. Exposure-response models for lymphoid cancer mortality vs. occupational
cumulative exposure (with 15-year lag) 4-21
Figure 4-4. RR estimate for all lymphohematopoietic cancer vs. occupational cumulative
exposure (with 15-year lag) 4-27
Figure 4-5. RR estimate for breast cancer mortality vs. occupational cumulative exposure
(with 20-year lag) 4-35
Figure 4-6. RR estimate for breast cancer incidence in full cohort vs. occupational
cumulative exposure (with 15-year lag) 4-43
Figure 4-7. RR estimate for breast cancer incidence in subcohort with interviews vs.
occupational cumulative exposure (with 15-year lag) 4-44
Figure 4-8. RR estimate for breast cancer incidence in subcohort with interviews vs.
occupational cumulative exposure (with 15-year lag); select models compared to
deciles 4-51
Figure 4-9. RR estimates for lymphoid cancer from occupational EtO exposures (with
15-year lag) 4-103
Figure 4-10. RR estimates for breast cancer incidence from occupational EtO exposures
(with 15-year lag) 4-107
IX
-------
LIST OF ABBREVIATIONS
ADAF
age-dependent adjustment factor
AIC
Akaike information criterion
AML
acute myeloid leukemia
AP
apurinic/apyrimidinic
AUC
areas under the curve
CASRN
Chemical Abstracts Service Registry Number
COSMIC
Catalogue of Somatic Mutations in Cancer
CI
confidence interval
DSB
double-strand breaks
ECoi
effective concentration (modeled) associated with 1% extra risk
ECio
effective concentration (modeled) associated with 10% extra risk
EPA
U.S. Environmental Protection Agency
ERR
excess relative risk
EtO
ethylene oxide
FIFRA
Federal Insecticide, Fungicide, and Rodenticide Act
GST
glutathione S-transferase
HERO
Health and Environmental Research Online
HEVal adduct
hydroxyethylvaline adduct
hprt
hypoxanthine phosphoribosyl transferase
I ARC
International Agency for Research on Cancer
ICD
International Classification of Diseases
i.p.
intraperitoneal
IRIS
Integrated Risk Information System
LacI
lactose-inducible lac operon transcriptional repressor gene
LEC*
95% (one-sided) lower confidence limit on the EC*
LOAEL
lowest-observed-adverse-effect level
MDS
myelodysplastic syndrome
MF
mutation frequency
MLE
maximum likelihood estimate
MN
micronucleus
N3 HEA
N3-(2-hydroxyethyl)adenine
N7-HEG
N7-(2-hydroxyethyl)guanine
NATA
National-scale Air Toxics Assessment
NCEA
National Center for Environmental Assessment
NHL
non-Hodgkin lymphoma
NIOSH
National Institute for Occupational Safety and Health
NRC
National Research Council
NTP
National Toxicology Program
06-HEG
06-hydroxyethylguanine
Obs
observed number
OR
odds ratios
ORD
Office of Research and Development
x
-------
LIST OF ABBREVIATIONS (CONTINUED)
PBPK
physiologically based pharmacokinetic
PBLs
peripheral blood lymphocytes
POD
point of departure
RR
relative rate (i.e., rate ratio), or more generally, relative risk
SAB
Science Advisory Board
SCE
sister chromatid exchanges
SE
standard error
SEER
Surveillance, Epidemiology, and End Results
SIR
standardized incidence ratio
SMR
standard mortality ratios
TWA
time-weighted average
UCC
Union Carbide Corporation
UCL
upper confidence limit
XI
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS
ASSESSMENT AUTHORS
Jennifer Jinot (Chemical Manager)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
David Bayliss (retired)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Jason Fritz
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Henry D. Kahn (retired)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Nagu Keshava
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Robert McGaughy (retired)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Ravi Subramaniam
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Larry Valcovic (retired)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Xll
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
ASSESSMENT AUTHORS (continued)
Suryanarayana Vulimiri
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
CONTRIBUTORS
James A. Deddens (retired)
National Institute for Occupational Safety and Health
Cincinnati, OH
Kyle Steenland (under contract to the EPA)
Rollins School of Public Health
Emory University
Atlanta, Georgia
Yu-ShengLin
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
PRODUCTION TEAM
Maureen Johnson
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Vicki Soto
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Dahnish Shams
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Xlll
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
EXECUTIVE DIRECTION
Kenneth Olden (Center Director—retired)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Michael Slimak (Acting Center Director)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
John Vandenberg (National Program Director, HHRA)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Lynn Flowers (Associate Director for Health, currently with the Office of Science Policy)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Vincent Cogliano (IRIS Program Director)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Gina Perovich (IRIS Program Deputy Director)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
Samantha Jones (IRIS Associate Director for Science)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
David Bussard (Director Washington Division)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
xiv
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
EXECUTIVE DIRECTION (continued)
Viktor Morozov (Quantitative Risk Methods Group)
National Center for Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC
REVIEWERS
Earlier drafts of this document were provided for review to EPA scientists, interagency reviewers
from other federal agencies and the Executive Office of the President, the public, and independent
scientists external to the EPA. Summaries and the EPA's dispositions of the comments received from
the independent external peer reviewers and from the public are included in Appendices H (for the 2006
external review draft), K (for the 2013 public comment draft), and I (for the 2014 SAB review draft).
EPA REVIEWERS
Office of the Administrator/National Center for Environmental Economics
Office of the Administrator/Office of Children's Health Protection
Office of Air and Radiation/Office of Air Quality and Planning Standards
Office of Land and Emergency Management
Office of Pesticide Programs/Antimicrobials Division
Office of Pesticide Programs/Health Effects Division
Office of Policy
Office of Research and Development/Office of Science Policy
Office of Research and Development/National Health and Environmental Effects Research laboratory
Office of Water
Region 1, Boston
Region 2, New York City
Region 3, Philadelphia
Region 8, Denver
ACKNOWLEDGEMENTS
The authors would like to acknowledge David Bussard, Jane Caldwell, Catherine Gibbons, Linda
Phillips, Leonid Kopylev, Julian Preston, Cheryl Scott, Maria Spassova, and Paul White of the EPA for
their contributions during the draft development process.
xv
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
EXTERNAL PEER REVIEWERS
SCIENCE ADVISORY BOARD ETHYLENE OXIDE REVIEW PANEL (2006-2007)
CHAIR
Dr. Stephen Roberts
University of Florida
OTHER SAB MEMBERS
Dr. Timothy Buckley
Ohio State University
Dr. Montserrat Fuentes
North Carolina State University
Dr. Dale Hattis
Clark University
Dr. James Kehrer
Washington State University
Dr. Mark Miller
California Environmental Protection Agency
Dr. Maria Morandi
University of Texas—Houston Health Science Center
Dr. Robert Schnatter
Exxon Biomedical Sciences, Inc.
Dr. Anne Sweeney
TAMU System Health Science Center
xvi
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
CONSULTANTS SERVING ON THE PANEL
Dr. Steven Alan Belinsky
University of New Mexico
Dr. Norman Drinkwater
University of Wisconsin Medical School
Dr. Steven Heeringa
University of Michigan
Dr. Ulrike Luderer
University of California
Dr. James Swenberg
University of North Carolina
Dr. Vernon Walker
Lovelace Respiratory Research Institute
SCIENCE ADVISORY BOARD ETHYLENE OXIDE REVIEW PANEL (2014-2015)
CHAIR
Dr. Peter S. Thorne
University of Iowa
MEMBERS
Dr. Henry Anderson
Wisconsin Division of Public Health
Dr. James V. Bruckner
University of Georgia
Dr. William Michael Foster
Duke University Medical Center
Dr. Lawrence Lash
Wayne State University
xvii
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (continued)
MEMBERS (continued)
Dr. Maria Morandi
Independent Consultant
Dr. Victoria Persky
University of Illinois at Chicago
Dr. Kenneth Ramos
University of Arizona
Dr. Stephen M. Roberts
University of Florida
CONSULTANTS ON THE PANEL
Dr. Steven Heeringa
University of Michigan
Dr. Peter Infante
Peter F. Infante Consulting, LLC
Dr. Gary Ginsberg
Connecticut Department of Public Health
Dr. Elizabeth A. (Lianne) Sheppard
University of Washington
Dr. Daniel Zelterman
Yale University
xviii
-------
1. EXECUTIVE SUMMARY
Ethylene oxide (EtO) is a gas at room temperature. It is manufactured from ethylene and used
primarily as a chemical intermediate in the manufacture of ethylene glycol. It is also used as a
sterilizing agent for medical equipment and a fumigating agent for spices.
CHARACTERIZATION OF THE CARCINOGENIC HAZARD
The DNA-damaging properties of EtO have been studied since the 1940s. EtO is known to be
mutagenic in a large number of living organisms, ranging from bacteriophage to mammals, and to
induce chromosome damage. It is carcinogenic in mice and rats, inducing tumors of the
lymphohematopoietic system, brain, lung, connective tissue, uterus, and mammary gland. In humans
employed in EtO-manufacturing facilities and in sterilizing facilities, there is strong evidence of an
increased risk of cancer of the lymphohematopoietic system and of breast cancer in females. Increases
in the risk of lymphohematopoietic cancer have been seen in most (but not all) of the epidemiological
studies of EtO-exposed workers, manifested as an increase either in leukemia or in cancer of the
lymphoid tissue. Of note, one large epidemiologic study conducted by the National Institute for
Occupational Safety and Health (NIOSH) of sterilizer workers that had a well-defined exposure
assessment for individuals reported positive exposure-response trends for lymphohematopoietic cancer
mortality, primarily in males and in particular for lymphoid cancer (i.e., non-Hodgkin lymphoma
[NHL], myeloma, and lymphocytic leukemia), and for breast cancer mortality in females (Steenland et
al.. 2004). The positive exposure-response trend for female breast cancer was confirmed in an incidence
study based on the same worker cohort (Steenland et al., 2003). There is supporting evidence for an
association between EtO and breast cancer from other studies, but the database is more limited than that
for lymphohematopoietic cancers, in part because there are not as many studies that include sufficient
numbers of females.
Although the evidence of carcinogenicity from human studies was deemed short of conclusive
on its own, EtO is characterized as "carcinogenic to humans" by the inhalation route of exposure based
on the total weight of evidence, in accordance with the U.S. Environmental Protection Agency's
(EPA's) 2005 Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a). The lines of evidence
supporting this characterization include: (1) strong, but less than conclusive on its own, epidemiological
evidence of lymphohematopoietic cancers and breast cancer in EtO-exposed workers, (2) extensive
evidence of carcinogenicity in laboratory animals, including lymphohematopoietic cancers in rats and
mice and mammary carcinomas in mice following inhalation exposure, (3) clear evidence that EtO is
genotoxic and sufficient weight of evidence to support a mutagenic mode of action for EtO
carcinogenicity, and (4) strong evidence that the key precursor events are anticipated to occur in humans
1-1
-------
and progress to tumors, including evidence of chromosome damage in humans exposed to EtO. Overall,
confidence in the hazard characterization of EtO as "carcinogenic to humans" is high.
DERIVATION OF THE INHALATION UNIT RISK ESTIMATE
Inhalation unit risk estimates were developed for evaluating the potential cancer risks posed by
inhalation exposure to EtO. The unit risk estimates for cancer mortality and incidence were based on the
human data from the NIOSH study (Steenland et al., 2004; Steenland et al., 2003). This study was
selected for the derivation of risk estimates because it is a high-quality study,1 it is the largest of the
available studies, and it has exposure estimates for the individual workers from a high-quality exposure
assessment. Multiple modeling approaches were evaluated for the exposure-response data, including
modeling the cancer response as a function of either categorical exposures or continuous individual
exposure levels. Model selection for each cancer data set was primarily based on a preference for
models of the individual-level continuous exposure data, prioritization of models that are more tuned to
local behavior in the low-exposure data, and a weighing of statistical and biological considerations.
Unit risk estimates based on the human data were first derived under the common assumption
that relative risk is independent of age. This assumption is later superseded by an assumption of
increased early-life susceptibility, and it is the unit risk estimates derived under this latter assumption
that are the ultimate estimates proposed in this assessment (presented further below).
Under the assumption that relative risk is independent of age, an LECoi (lower 95% confidence
limit on the ECoi, the estimated effective concentration associated with 1% extra risk) for excess
lymphoid cancer mortality (Steenland et al., 2004) was calculated using a life-table analysis and the
lower spline segment from a two-piece linear spline model. Linear low-dose extrapolation below the
range of observations is supported by the conclusion that a mutagenic mode of action is operative in EtO
carcinogenicity. Linear low-dose extrapolation from the LECoi for lymphoid cancer mortality yielded a
lifetime extra cancer unit risk estimate of 1.1 x 10~3 per [j,g/m3 (2.0 x 10~3 per ppb)2 of continuous EtO
exposure. Applying the same lower-spline regression coefficient and life-table analysis to background
lymphoid cancer incidence rates and applying linear low-dose extrapolation resulted in a preferred
lifetime extra lymphoid cancer unit risk estimate of 2.9 x 10"3 per [j,g/m3 (5.3 x 10"3 per ppb), as cancer
incidence estimates are generally preferred over mortality estimates.
1 Tlx; NIOSH study (Steenland et al.. 2004: Steenland et al.. 20031 was judged to be a "high-quality" study based on the
attributes discussed in Section 3.1 and in Section A.2.8 of Appendix A, including availability of individual worker exposure
estimates from a high-quality exposure assessment, cohort study design, large size, inclusion of males and females, adequate
follow-up, absence of any known confounding exposures, and use of internal comparisons. The breast cancer incidence
study using the subcohort of female workers with interviews had the additional attribute of investigating and controlling for a
number of breast cancer risk factors C Steenland etal.. 20031.
Conversion equation: 1 ppm= 1,830 |ig/m3.
1-2
-------
Breast cancer incidence risk estimates were calculated directly from the data from a breast cancer
incidence study of the same occupational cohort (Steenland et al.. 2003). Using the same life-table
approach, the lower spline segment from a two-piece linear spline model, and linear low-dose
extrapolation, a unit risk estimate of 8.1 x 10-4 per [j,g/m3 (1.5 x 10-3 per ppb) was obtained for breast
cancer incidence. A unit risk estimate for breast cancer mortality was also calculated from the cohort
mortality data; however, the incidence estimate is preferred over the mortality estimate.
Combining the incidence risk estimates for the two cancer types resulted in a total cancer unit
risk estimate of 3.3 x 10"3 per [j,g/m3 (6.1 x 10"3 per ppb).3
Unit risk estimates (for total cancer) were also derived from the three chronic rodent bioassays
for EtO reported in the literature. These estimates, ranging from 2.2 x 10~5 per [j,g/m3 to 4.6 x 10~5 per
[j,g/m3, are about two orders of magnitude lower than the estimate based on human data. The Agency
takes the position that human data, if adequate data are available, provide a more appropriate basis than
rodent data for estimating population risks (U.S. EPA. 2005a). primarily because uncertainties in
extrapolating quantitative risks from rodents to humans are avoided. Although there is a sizeable
difference between the rodent-based and the human-based estimates, the human data are from a large,
high-quality study, with EtO exposure estimates for the individual workers and little reported exposure
to chemicals other than EtO. Therefore, the estimates based on the human data are the preferred
estimates for this assessment.
Because the weight of evidence supports a mutagenic mode of action for EtO carcinogenicity,
and as there are no chemical-specific data from which to assess early-life susceptibility, increased
early-life susceptibility should be assumed, according to the EPA's Supplemental Guidance for
Assessing Susceptibility from Early-Life Exposure to Carcinogens—hereinafter referred to as the
"EPA's Supplemental Guidance" fU. S. EPA. 2005b). This mode-of-action-based assumption of
increased early-life susceptibility supersedes the assumption of age independence under which the
human data-based estimates presented above were derived. Thus, using the same approach and
exposure-response models as for the estimates discussed above but initiating exposure in the life-table
analysis at age 16 instead of at birth, adult-exposure-only unit risk estimates were calculated for
lymphoid cancer incidence and breast cancer incidence under an alternate assumption that relative risk is
independent of age for adults, which represent the life stage pertaining to the occupational cohort data
which were used for the exposure-response modeling. These adult-exposure-only unit risk estimates
were then rescaled to a 70-year basis for use in the standard age-dependent adjustment factors (ADAFs)
calculations and risk estimate calculations involving less-than-lifetime exposure scenarios. The
resulting adult-based unit risk estimates were 2.6 x 10"3 per [j,g/m3 (4.8 x 10"3 per ppb) for lymphoid
3The method used to derive the total cancer unit risk estimate involves estimating an upper bound on the sum of the
maximum likelihood estimates of risk; see Section4.1.3.
1-3
-------
cancer incidence, 7.0 x io_4 per [j,g/m3 (1.3 x 10"3 per ppb) for breast cancer incidence in females, and
3.0 x 10"3 per [j,g/m3 (5.5 x 10"3 per ppb) for both cancer types combined. The adult-based unit risk
estimates, which were derived under an assumption of increased early-life susceptibility, supersede
those presented earlier that were derived under the assumption that relative risk is independent of age.
When using the adult-based unit risk estimates to estimate extra cancer risks for a given exposure
scenario, the standard ADAFs should be applied, in accordance with the EPA's Supplemental Guidance
(U.S. EPA. 2005b). Applying the ADAFs to obtain a full lifetime total cancer unit risk estimate yields
5.0 x 10"3 per [j,g/m3 (9.1 x 10"3 per ppb), and the commensurate lifetime chronic (lower-bound)
exposure level of EtO corresponding to an increased cancer risk of 10"6 is 2 x 10"4 |j,g/m3 (1 x 10"4 ppb).
The unit risk estimate is intended to provide a reasonable upper bound on cancer risk from
inhalation exposure. The estimate was developed for environmental exposure levels (it is considered
valid for exposures up to about 40 [j,g/m3 [20 ppb]) and is not applicable to higher level exposures, such
as those that may occur occupationally, which appear to have a different exposure-response relationship
(see below for a summary of risk estimates for occupational exposure scenarios).
CONFIDENCE IN THE UNIT RISK ESTIMATE
The primary sources of uncertainty in the unit risk estimates derived from the human data
include the retrospective exposure assessment conducted for the epidemiology study, the
exposure-response modeling of the epidemiological data, and the low-dose extrapolation.4 Despite
uncertainties in the unit risk estimate, confidence in the estimate is relatively high. First, confidence in
the hazard characterization of EtO as "carcinogenic to humans," which is based on strong
epidemiological evidence supplemented by other lines of evidence, is high. Second, the unit risk
estimate is based on human data from a large, high-quality epidemiology study with individual worker
exposure estimates. Retrospective exposure estimation is an inevitable source of uncertainty in this type
of epidemiology study; however, the NIOSH investigators put extensive effort into addressing this issue
by developing a state-of-the-art regression model to estimate unknown historical exposure levels using
variables, such as sterilizer size, for which historical data were available. In addition, the two-piece
spline models used in this assessment to model the supralinear exposure-response relationships are
considered to provide a reasonable basis for the derivation of unit risk estimates. Finally, the use of
linear low-exposure extrapolation is strongly supported by the conclusion that EtO carcinogenicity has a
mutagenic mode of action.
Confidence in the unit risk estimate is particularly high for the breast cancer component, which is
based on over 200 incident cases for which the investigators also had information on other potential
breast cancer risk factors. The selected model for the breast cancer incidence data provided a good
4See Section 4.1.4 for additional discussion of these and other sources of uncertainty in the unit risk estimates.
1-4
-------
global fit as well as a good local fit in the lower exposure range of greatest relevance for the derivation
of a unit risk estimate. The actual unit risk might be higher or lower; however, considering the
continuous-exposure linear model as a lower bound for the supralinear exposure-response relationship
suggests that while a unit risk estimate for breast cancer incidence that is up to fourfold lower is
plausible, unit risk estimates lower than that are considered unlikely from the available data. Sensitivity
analyses for lag time, inclusion of covariates, knot, upper-bound estimation approach, use of the full
incidence cohort, and inclusion of only invasive cancers for the breast cancer background rates in the
life table indicate that the unit risk estimate is not highly influenced by these factors, with comparison
unit risk estimates differing by at most 40%.
There is somewhat less, although still relatively high in general, confidence in the lymphoid
cancer component of the unit risk estimate because it is based on fewer events (53 lymphoid cancer
deaths); incidence risk was estimated from mortality data; and the exposure-response relationship is
exceedingly supralinear, complicating the exposure-response modeling and model selection to a greater
extent than for breast cancer incidence. The actual unit risk might be higher or lower than that from the
selected model, and there were no clear upper or lower bounds for the apparent exposure-response
relationship provided by other models. Sensitivity analyses for lag time, knot, and upper-bound
estimation approach, indicate that the unit risk estimate for lymphoid cancer is more influenced by these
factors than was the estimate for breast cancer incidence. Comparison unit risk estimates from the
sensitivity analyses ranged from about 50% of the preferred unit risk estimate to about three times that
estimate. While there is less confidence in the lymphoid cancer unit risk estimate than in the breast
cancer unit risk estimate, the lymphoid cancer estimate is considered a reasonable estimate from the
available data, and overall, there is relatively high confidence in the total cancer unit risk estimate.
RISK ESTIMATES FOR OCCUPATIONAL EXPOSURE SCENARIOS
As noted above, the inhalation unit risk estimate was developed for environmental exposure
levels (up to about 40 [j,g/m3 [20 ppb]) and is not applicable to higher exposure levels, such as those that
may occur occupationally, which appear to have a different exposure-response relationship. However,
occupational exposure levels of EtO are of concern to the EPA when EtO is used as a pesticide
(e.g., sterilizing agent or fumigant). Therefore, this document also presents estimates of extra risk for
the two cancer types for a range of occupational inhalation exposure scenarios (see Section 4.7).
Maximum likelihood estimates of the extra (incidence) risk of lymphoid cancer and breast cancer
combined for the range of occupational exposure scenarios considered (i.e., 0.1 to 1 ppm 8-hour
time-weighted average [TWA] for 35 years) ranged from 0.037 to 0.11; upper-bound estimates ranged
from 0.081 to 0.22. The uncertainty associated with the extra risk estimates for occupational exposure
scenarios is less than that associated with the unit risk estimates for environmental exposures, and the
overall confidence in the extra risk estimates for occupational exposure is high. The extra risk estimates
1-5
-------
are derived for occupational exposure scenarios that yield cumulative exposures well within the range of
the exposures in the NIOSH study. Moreover, the NIOSH study is a study of sterilizer workers who
used EtO for the sterilization of medical supplies or spices (Steenland et al.. 1991); thus, the results are
directly applicable to workers in these occupations, and these are among the occupations of primary
concern to the EPA.
SUMMARY OF ASSESSMENT FINDINGS
Table 1-1 provides a summary of the major findings in this assessment.
1-6
-------
Table 1-1. Summary of major findings
Hazard conclusions
Hazard characterization
The weight of evidence from
epidemiological studies and supporting
information is sufficient to conclude that
ethylene oxide is carcinogenic to humans.
Mode of action
The weight of evidence is sufficient to
conclude that ethylene oxide carcinogenicity
has a mutagenic mode of action.
Inhalation unit risk estimates (for environmental exposures)3
Basis
Inhalation unit risk estimate3 (per jig/m3)b
Full lifetime unit risk estimate (includes ADAFs)0
Total cancer risk based on human datad—lymphoid cancer incidence
and breast cancer incidence in females
5.0 x 103
Adult-based unit risk estimates (for use with ADAFs)6
Total cancer risk based on human datad—lymphoid cancer incidence
and breast cancer incidence in females
3.0 x 103
Lymphoid cancer incidence inboth sexes based on human data
2.6 x lO-3
Breast cancer incidence in females based on human data
7.0 x 10"4
Total cancer incidence risk estimate from rodent data (female mouse)
4.6 x 10-5
Extra risk estimates for occupational inhalation exposure scenarios (see Section 4.7)
Maximum likelihood estimates of the extra risk of lymphoid cancer
and breast cancer combined for the range of occupational exposure
scenarios considered (i.e., 0.1 to 1 ppm 8-hr TWA for 35 yr)f
0.037-0.11
Upper-bound estimates of the extra risk of lymphoid cancer and
breast cancer combined for the range of occupational exposure
scenarios considered (i.e., 0.1 to 1 ppm 8-hr TWA for 35 yr)f
0.081-0.22
aThese unit risk estimates are not intended for use with continuous lifetime exposure levels above about 40 |xg/m3. See
Section 4.7 for risk estimates based on occupational exposure scenarios. Preferred estimates are in bold.
bTo convert unit risk estimates to (ppm)"1, multiply the (jig/m3)"1 estimates by 1,830 (|ig/m3)/ppm. Also, 1 ppb = 1.83 |xg/m3.
cBecause the weight of evidence supports a mutagenic mode of action for EtO carcinogenicity, and because of the lack of
chemical-specific data, the EPA assumes increased early-life susceptibility and recommends the application of ADAFs, in
accordance with the EPA's Supplemental Guidance (TJ.S. EPA. 2005bl for exposure scenarios that include early-life exposures.
For the full lifetime (upper-bound) unit risk estimate presented here, ADAFs have been applied, as described in Section 4.4.
dTo be precise, this unit risk estimate reflects the total (upper-bound) cancer risk to females and not to the general population
because the breast cancer risk estimate only applies to females. As a practical matter for regulatory purposes, however, females
comprise roughly half the general population and this unit risk estimate enables risk managers to evaluate the individual risk for
this substantial population group. For the purposes of estimating numbers of cancer cases attributable to specific exposure levels
(e. g., for benefits analyses), it would be more appropriate to use the cancer-specific unit risk estimates (or central tendency
estimates), taking sex into account.
eThese (upper-bound) unit risk estimates are intended for use in ADAF calculations and less-than-lifetime adult exposure
scenarios ("U.S. EPA. 2005^. Note that these are not the same as the unit risk estimates derived directly from the human data in
Section 4.1 under the assumption that RRs are independent of age. See Section 4.4 for the derivation of the adult-based unit risk
estimates.
Technically, these sums would reflect the total cancer risk to females and not a mixed-sex workforce because the breast cancer
risk estimate only applies to females. As a practical matter for regulatory purposes, however, females typically comprise a
substantial proportion of the sterilizer workforce and summing these extra risk estimates enables risk managers to evaluate the
individual risk for this substantial workforce group. In a situation in which the workforce of concern is comprised
predominantly of males, it might be appropriate to use a sex-weighted sum of the extra risks from the two cancer types (see
Section 4.7 for the cancer-specific extra risk estimates),
hr = hour; yr = year.
1-7
-------
2. INTRODUCTION
Chapter 2 provides background information for the assessment, including:
• A brief overview about Ethylene oxide (EtO) exposure
The purpose of the assessment
• A history of the development of the assessment and its external reviews
• A summary of literature search approaches (see Section 2.1)
• A discussion of consistency of the assessment with 2011 National Research Council
(NRC) recommendations (see Section 2.2)
Ethylene oxide (EtO) is a gas at room temperature. It is manufactured from ethylene and used
primarily as a chemical intermediate in the manufacture of ethylene glycol. It is also used as a
sterilizing agent for medical equipment and certain other items and as a fumigating agent for spices.
The largest sources of human exposure are in occupations involving contact with the gas in plants that
manufacture or use EtO and in hospitals that sterilize medical equipment. EtO can also be inhaled by
residents living near production or sterilizing/fumigating facilities. Based on the U. S. Environmental
Protection Agency's (EPA's) 2005 National-scale Air Toxics Assessment (NATA) data, the average
environmental exposure concentration from all sources (including concentrations near known sources)
in the United States is 0.0062 (J,g/m3; the average background concentration excluding concentrations
near known sources of EtO is 0.0044 [j,g/m3 (NATA 2005 data,
http://www.epa.gov/ttn/atw/nata2005/tables.html).
EPA offices with an interest in EtO include the Office of Air and Radiation and the Office of
Pesticide Programs. The Office of Air and Radiation has an interest because EtO is one of the
188 hazardous air pollutants listed in the 1990 Clean Air Act Amendments. The Office of Pesticide
Programs has an interest in both environmental and occupational exposures resulting from the
sterilization and fumigation uses of EtO because the EPA is responsible for pesticide labeling and
registration decisions under the Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA).
The purpose of this document is to provide scientific support and rationale for the hazard and
dose-response assessment in the Integrated Risk Information System (IRIS) pertaining to
carcinogenicity from chronic inhalation exposure to EtO (Chemical Abstracts Service Registry Number
[CASRN] 75-21-8). It is not intended to be a comprehensive treatise on the chemical or toxicological
nature of EtO. In general, this IRIS Carcinogenicity Assessment provides information on the
carcinogenic hazard potential of EtO and quantitative estimates of cancer risk from inhalation exposure.
The information includes a weight-of-evidence judgment of the likelihood that the agent is a human
carcinogen and the conditions under which the carcinogenic effects may be expressed. Quantitative risk
2-1
-------
estimates for inhalation exposure (inhalation unit risks) are derived. The definition of an inhalation unit
risk is a plausible upper bound on the estimate of risk per [j,g/m3 air breathed.
Development of the hazard identification and dose-response assessments for EtO has followed
the general guidelines for risk assessment as set forth by the National Research Council (NRC, 1983).
EPA Guidelines and Risk Assessment Forum Technical Panel Reports that were used in the
development of this assessment include the following: Guidelines for Mutagenicity Risk Assessment
(U.S. EPA. 1986). Methods for Derivation of Inhalation Reference Concentrations and Application of
Inhalation Dosimetry (U.S. EPA. 1994). Benchmark Dose Technical Guidance (U.S. EPA. 2012).
Science Policy Council Handbook. Risk Characterization (U.S. EPA. 2000). Guidelines for Carcinogen
Risk Assessment U.S. EPA (2005a), Supplemental Guidance for Assessing Susceptibility from Early-Life
Exposure to Carcinogens (U. S. EPA. 2005b). and Science Policy Council Handbook: Peer Review
(U.S. EPA. 2006b).
An external review draft of this carcinogenicity assessment (U. S. EPA. 2006a) was peer
reviewed by a panel of the EPA's Science Advisory Board (SAB) in 2007 (SAB. 2007). See
Appendix H for a summary and the EPA's disposition of SAB and public comments on the 2006
external review draft. In response to comments from that SAB review, the EPA conducted extensive
new exposure-response modeling of certain epidemiologic data. In addition, the EPA updated the
assessment to reflect new literature through May 2013 (see Appendix J). In July 2013, the EPA released
a revised draft for public comment (U.S. EPA. 2013a. b), and that draft assessment was discussed at the
EPA's December 2013 IRIS Bimonthly Public Meeting. Appendix K contains the EPA's responses to
the public comments that were received on the July 2013 draft. A further revised external review draft
(U.S. EPA. 2014a, b) was reviewed by another panel of the SAB in 2014-2015 (SAB, 2015), primarily
to receive comments on the expanded exposure-response modeling of the epidemiologic data. See
Appendix I for the EPA's disposition of SAB comments on the 2014 draft. Finally, the EPA has further
updated the assessment to reflect new literature through July 2016; this new literature did not
substantively impact the conclusions of the assessment (see Appendix J).
2.1. LITERATURE IDENTIFICATION
The literature search strategy first employed for this assessment was based on the CASRN and at
least one common name. Any pertinent scientific information submitted by the public to the IRIS
Submission Desk was also considered in the development of this document, and references were added
after the first external peer review in response to the reviewers' and public comments.
The cutoff date for literature inclusion into the main body of this carcinogenicity assessment was
30 June 2010. At that time, the analyses and text were largely completed, except for a few focused
issues that remained for discussion and review.
2-2
-------
In preparation for the second external peer review, a well-documented systematic literature
search was conducted for the time frame from January 2006 to May 2013 to ensure that no major studies
were missed from the time of the first external review draft in 2006 until the cutoff date and to
determine whether any significant new studies had been published since the cutoff date that might alter
the findings of the assessment. This systematic literature search is described in Section J.l of
Appendix J. Based on this search, 56 references were identified as potentially relevant to the EtO
assessment. None of the new studies were judged to impact the assessment's major conclusions.
Nonetheless, two new studies of high pertinence to the assessment were identified, and these studies are
reviewed in Section J.2 of Appendix J for transparency and completeness. Furthermore, one of the
studies is a follow-up of an epidemiology study already included in the assessment, and this follow-up
study is also discussed in the main body of the assessment. Reviews of an additional two new studies of
high pertinence were added to Appendix J to address public comments received in October and
December of 2013 (see Section J.3); these studies similarly did not impact the assessment's major
conclusions.
For this current assessment, another well-documented systematic literature search was conducted
using the same approach as in the earlier search (see Section J.l of Appendix J), this time for the time
frame from May 2013 through August 2016. Based on this search, 17 references were identified as
potentially relevant to the EtO assessment and have been added to Section J. 1 of Appendix J. A further
two new studies of high pertinence to the assessment were identified, and these studies are reviewed in
Section J.4 of Appendix J. Once again, none of the new studies were judged to impact the assessment's
major conclusions.
All the references considered and cited in this document, including abstracts, can be found on the
Health and Environmental Research Online (HERO) website.5
2.2. NATIONAL RESEARCH COUNCIL RECOMMENDATIONS OF 2011
In 2011, the National Research Council (NRC), in its review of the EPA's draft IRIS assessment
of formaldehyde, provided a number of recommendations related to the IRIS program (NRC, 2011).
The EtO assessment was one of a group of chemical assessments for which an external peer review had
already been completed at the time the 2011 NRC recommendations were released. For this group of
assessments, the EPA focused on a subset of the short-term recommendations, such as streamlining
documents, increasing transparency and clarity, and using more tables, figures, and appendices to
5HERO is a database of scientific studies and other references used to develop the EPA's risk assessments, which are aimed
at understanding the health and environmental effects of pollutants and chemicals. HERO is developed and managed in the
EPA's Office of Research and Development (ORD) by the National Center for Environmental Assessment (NCEA). The
database includes more than 1,000,000 scientific articles from the peer-reviewed literature. New studies are added
continuously to HERO.
2-3
-------
present information and data in assessments. Thus, the EtO assessment does not incorporate recent
revisions to the IRIS assessment format recommended in the 2011 NRC recommendations [and the more
recent 2014 NRC Review of the IRIS Process (NRC. 2014)1. such as including a standard Preamble and
the revised chapter structure, and does not fully implement the longer term NRC recommendations. The
assessment does, however, conform to a number of the recommendations. For example, the assessment
is streamlined and uses tables, figures, and appendices to increase transparency and clarity. In addition,
the assessment is structured to have separate hazard identification and dose-response sections, and the
update to the literature search was conducted using systematic literature search approaches.
Furthermore, consistent with the goal that assessments should provide a scientifically sound and
transparent evaluation of the relevant scientific literature and presentation of the analyses performed, the
assessment contains an expanded discussion on the rationales for study evaluation and selection, as well
as for other key assessment decisions.
For general information about this assessment or other questions relating to IRIS, the reader
is referred to the EPA's IRIS Hotline at (202) 566-1676 (phone), (202) 566-1749 (fax), or
hotline.iris@epa.gov (email address).
2-4
-------
3. HAZARD IDENTIFICATION
Chapter 3 provides an evaluation of the evidence regarding the carcinogenicity of EtO.
Major findings of Chapter 3:
1. The weight of evidence from epidemiological studies and supporting information
is sufficient to conclude that EtO is "carcinogenic to humans."
2. The weight of evidence is sufficient to conclude that EtO carcinogenicity has a
mutagenic mode of action.
This chapter presents the evidence considered in the hazard identification of EtO carcinogenicity
and the hazard characterization resulting from the weight-of-evidence evaluation. Section 3.1
summarizes the human evidence (a more detailed discussion of the human cancer studies is presented in
Appendix A). Section 3.2 describes the evidence from laboratory animal studies. Section 3.3 discusses
supporting evidence, in particular evidence regarding the genotoxicity of EtO. Section 3.4 provides the
mode-of-action analysis for EtO carcinogenicity. To conclude the chapter, Section 3.5 presents the
hazard characterization for EtO carcinogenicity and a discussion of life stages and populations with
potentially increased susceptibility.
3.1. EVIDENCE OF CANCER IN HUMANS
The literature from 1985, the year of the EPA's previous health assessment document on EtO
(U.S. EPA. 1985), to the present contains numerous epidemiological studies of the carcinogenic effects
of EtO in occupational cohorts; some of these cohorts were the subject of multiple reports. The
conclusions about the human evidence of carcinogenicity in this assessment are based on the following
summary of those studies, which are discussed in more detail and critically reviewed in Appendix A.
Table A-5 in Appendix A provides a tabular summary of the epidemiological studies, including some
study details, results, and limitations. The strengths and weaknesses of these studies were evaluated
individually using standard considerations in evaluating epidemiological studies. The major areas of
concern are study design, exposure assessment, and data analysis. General features of study design
considered include sample size and assessment of the health endpoint. For case-control studies, design
considerations include representativeness of cases, selection of controls, participation rates, use of proxy
respondents, and interview approach (e.g., blinding). For cohort studies, design considerations include
selection of referent population (e.g., internal comparisons are generally preferred to comparisons with
3-1
-------
an external population6), loss to follow-up, and length of follow-up. Exposure assessment issues include
specificity of exposure (exposure misclassification), characterization of exposure (e.g., ever exposed or
quantitative estimate of exposure level), and potential confounders. Analysis considerations include
adjustment for potential confounders or effect modifiers and modeling of exposure-response
relationships.
Two primary sources of exposures to EtO are production facilities and sterilization operations.
There are two types of production facilities (1ARC. 1994b):
1. Those using the older chlorohydrin process, where ethylene is reacted with hypochlorous acid
and then with calcium oxide to make EtO (this method produces unwanted byproducts, the most
toxic of which is ethylene dichloride), and
2. Those producing EtO via direct oxidation of ethylene in a pressurized vessel, which involves less
EtO exposure and eliminates the chemical byproducts of the chlorohydrin process.
Exposure in the sterilization of medical equipment and in the direct oxidation process is predominantly
to EtO, whereas exposure in the chlorohydrin process is to EtO mixed with other chemicals.
Hogstedt et al. (1986) and Hogstedt (1988) summarized findings of three Swedish occupational
cohorts (539 men and 170 women) exposed in a plant where hospital equipment was sterilized, in a
chlorohydrin production facility, and in a direct oxidation production facility. The incidence of
leukemia was elevated in all cohorts, although the risk was not statistically significant in the cohort from
the direct oxidation facility. For the three cohorts combined there were statistically significantly
elevated standard mortality ratios (SMRs) for leukemia (SMR = 9.2; 95% confidence interval
[CI] = 3.7-19), based on 7 deaths, and for stomach cancer (SMR = 5.5; 95% CI = 2.6-10), based on
10 deaths. Although this study reported high SMRs for leukemia, stomach cancer, and total cancer,
there were some limitations, such as multiple exposures to numerous other chemicals, lack of personal
exposure information, and lack of latency analysis. No gender differences were separately analyzed.
No dose-response calculations were possible. This study provides suggestive evidence of the
carcinogenicity of EtO.
Coggon et al. (2004) reported the results of a follow-up study of a cohort originally studied by
Gardner et al. (1989). The cohort included workers in three EtO production facilities in Great Britain
(two using both chlorohydrin and direct oxidation processes and the third using direct oxidation only); in
a fourth facility that used EtO in the manufacture of other chemicals; and in eight hospitals that used
internal comparisons are considered superior to external comparisons in occupational epidemiology studies because internal
comparisons help control for the healthy worker effect and other factors that might be more comparable within a study's
worker population than between the workers and the general population.
3-2
-------
EtO in sterilizing units. The total cohort comprised 1,864 men and 1,012 women. No statistically
significant excesses were observed for any cancer site. Slight increases, based on small numbers, were
observed for the various lymphohematopoietic cancers: Hodgkin lymphoma (2 vs. 1 expected),
non-Hodgkin lymphoma (NHL) (7 vs. 4.8), multiple myeloma (3 vs. 2.5), and leukemia (5 vs. 4.6). The
increases were concentrated in the 1,471 chemical-manufacturing workers, of whom all but 1 were male.
In the chemical-manufacturing workers with "definite" exposure, four leukemias were observed
(1.7 expected) and nine lymphohematopoietic cancers were observed (4.9 expected). A slight deficit in
the risk of breast cancer deaths (11 vs. 13.2) was observed in the cohort. No individual exposure
measurements were obtained from cohort members, and no exposure measurements were available
before 1977. Multiple exposures to other chemicals, small numbers of deaths, and lack of individual
EtO measurements make this study only suggestive of a higher risk of leukemia from exposure to EtO.
A series of retrospective mortality studies of about 2,000 male workers who were assigned to
operations that used or produced EtO in either of two Union Carbide Corporation (UCC) chemical
production facilities in West Virginia (Valdez-Flores et al.. 2010; Swaen et al.. 2009; Teta et al.. 1999;
Benson and Teta. 1993; Teta et al.. 1993; Greenberg et al.. 1990) have been published. EtO was
produced at these facilities until 1971, after which it was imported to the facilities. For EtO production,
the chlorohydrin process was used from 1925 to 1957, and the direct oxidation process was used from
1937 to 1971 (during overlapping years, both processes were in use). The cohort was observed from
1940 through 1978 in the original study (Greenberg et al.. 1990). through 1988 in the Teta et al. (1993).
Benson and Teta (1993). and Teta et al. (1999) studies, and through 2003 in the Swaen et al. (2009) and
Valdez-Flores et al. (2010) studies. A large-scale industrial hygiene survey and monitoring of EtO
concentrations was carried out in 1976, at which time EtO was in use at the facilities but no longer in
production.
Greenberg et al. (1990) found elevated but not statistically significant risks of pancreatic cancer
(SMR = 1.7) and leukemia (SMR = 2.3) (each based on seven cases) in the entire cohort; most of the
cases occurred in the chlorohydrin production unit (note that the chlorohydrin production unit produced
primarily ethylene chlorohydrin, which is used in chlorohydrin-based EtO production, but this unit is not
where chlorohydrin-based EtO production took place). Limitations of this study included multiple
exposures to many different chemicals in the facility through the years and lack of EtO exposure
measurements prior to 1976. Three categories of exposure were established for analysis—low,
intermediate, and high—based on a qualitative characterization of the potential for EtO exposure. The
number of workers in each exposure category was not reported. No significant findings of a
dose-response relationship were discernible. No quantitative estimates of individual exposure were
made in this study, and no latency analysis was conducted (average follow-up was 20 years).
Furthermore, EtO is not the only chemical to which the observed excesses in cancer mortality could be
attributed.
3-3
-------
A follow-up study (Teta et al.. 1993) that extended the observation of this cohort (excluding the
278 chlorohydrin production unit workers, who reportedly had low EtO exposures) for an additional
10 years to 1988 found no significant risk of total cancer; there was a slight trend in the risk of leukemia
with increasing duration of assignment to departments using or processing EtO, but it was not significant
(p = 0.28) and was based on only five cases. The average follow-up was 27 years, and at least 10 years
had elapsed since first exposure for all workers. The same problems of exposure ascertainment exist for
this study as for that of Greenberg et al. (1990). and furthermore, the follow-up did not update work
histories for the workers after 1978. EtO production at the plants was discontinued before 1978, as
noted by Teta et al. (1993); however, according to Greenberg et al. (1990). certain nonproduction areas
had "intermediate" potential for EtO exposure, although estimates of exposure levels suggest that the
levels would also be lower during the update period [<1 ppm 8-hour time-weighted average (TWA),
according to Teta et al. (1993)1. It appears from the Greenberg et al. (1990) publication that the high
potential exposure group was reserved for EtO production workers, and according to Teta et al. (1993).
only 425 EtO production workers were in the cohort. Of these, only 118 worked in the
chlorohydrin-based production process, where exposures were reportedly highest. Essentially, the study
did not support the earlier studies of cancer in EtO workers; however, it was limited by low statistical
power and a crude exposure assessment, and thus, is not very informative regarding whether exposure to
EtO is causally related to cancer.
In a parallel follow-up study through 1988 of only the chlorohydrin production employees,
Benson and Teta (1993) found that pancreatic cancer and lymphohematopoietic cancer cases continued
to accumulate and that the SMRs were statistically significant for pancreatic cancer (SMR = 4.9;
observed number [Obs] = 8,^<0 .05) and for lymphohematopoietic cancer (SMR = 2.9; Obs = 8,
p < 0.05). These investigators interpreted these excesses as possibly due to ethylene dichloride, a
byproduct in the chlorohydrin process. Again, this small study of only 278 workers was limited by the
same problems as the Greenberg et al. (1990) study and the Teta et al. (1993) study. No individual
estimates of exposure were available and the workers were potentially exposed to many different
chemicals (see Table A-5 in Appendix A). Furthermore, the chlorohydrin production unit was
reportedly considered a low potential EtO exposure department. Hence, this study has little weight in
determining the carcinogenicity of EtO.
In a later analysis, Teta et al. (1999) fitted Poisson regression dose-response models to the UCC
data (followed through 1988 and excluding the chlorohydrin production workers) and to data (followed
through 1987) from a study by the National Institute for Occupational Safety and Health (NIOSH)
(described below). Because Teta et al. (1999) did not present risk ratios for the cumulative exposure
categories used to model the dose-response relationships, the only comparison that can be made between
the UCC and NIOSH data is based on the fitted models. These models are almost identical for
leukemia, but for the lymphoid category, the risk—according to the fitted model for the UCC
3-4
-------
data—decreased as a function of exposure, whereas the risk for the modeled NIOSH data increased as a
function of exposure. However, the models are based on small numbers of cases (16 [5 UCC, 11
NIOSH] for leukemia; 22 [3 UCC, 19 NIOSH] for lymphoid cancers), and no statistics are provided to
assess model goodness-of-fit or to compare across models. In any event, this analysis is superseded by
the more recent analysis by the same authors (Valdez-Flores et al.. 2010) of the results of more recent
follow-up studies of these cohorts (see below).
Swaen et al. (2009) studied the same UCC cohort identified by Teta et al. (1993) (i.e., without
the chlorohydrin production workers) but extended the cohort enumeration period from the end of 1978
to the end of 1988, identifying 167 additional workers, and conducted mortality follow-up of the
resulting cohort of 2,063 male workers through 2003. Work histories were also extended through 1988
(exposures after 1988 were considered negligible compared to earlier exposure levels). Swaen et al.
(2009) used an exposure assessment based on the qualitative categorizations of potential EtO exposure
in the different departments developed by Greenberg et al. (1990) and time period exposure estimates
from Teta et al. (1993). This exposure assessment was relatively crude, based on just a small number of
department-specific and time period-specific categories, and with exposure estimates for only a few of
the categories derived from actual measurements (see Appendix A.2.20 for details).
At the end of the 2003 follow-up, 1,048 of the 2,063 workers had died (Swaen et al.. 2009). The
all-cause mortality SMR was 0.85 (95% CI = 0.80, 0.90) and the cancer SMR was 0.95 (95% CI = 0.84,
1.06). None of the SMRs for specific cancer types showed any statistically significant increases. In
analyses stratified by hire date (pre- [inclusive] or post-1956), the SMR for leukemia was elevated but
not statistically significant (1.51; 95% CI 0.69, 2.87) in the early-hire group, based on nine deaths. In
analyses stratified by duration of employment, no trends were apparent for any of the
lymphohematopoietic cancers, although in the 9+ years of employment subgroup, the SMR for NHL
was nonsignificantly increased (1.49; 95% CI 0.48, 3.48), based on five deaths. In SMR analyses
stratified by cumulative exposure, no trends were apparent for any of the lymphohematopoietic cancers
and there were no notable elevations for the highest cumulative exposure category. Note that only
27 lymphohematopoietic cancer deaths (including 12 leukemias and 11 NHLs) were observed in the
cohort.
Swaen et al. (2009) also did internal Cox proportional hazards modeling for some disease
categories (all-cause mortality, leukemia mortality, and lymphoid cancer [NHL, lymphocytic leukemia,
and myeloma] mortality [17 deaths]), using cumulative exposure as the exposure metric. These analyses
showed no evidence of an exposure-response relationship. Alternate Cox proportional hazard analyses
and categorical exposure-response analyses of the UCC data conducted by Valdez-Flores et al. (2010)
for a larger set of cancer endpoints similarly reported an absence of any exposure-response relationships.
Each of these cancer analyses, however, relies on small numbers of cases and a crude exposure
assessment, with a high potential for exposure misclassification.
3-5
-------
In a study of 2,658 male workers at eight chemical plants in West Germany that produced EtO
(manufacturing process not stated), Kiesselbach et al. (1990) found slightly increased SMRs for cancers
of the stomach, esophagus, and lung. A latency analysis was done only for stomach cancer and total
mortality. The investigators considered 71.6% of the cohort to be "weakly" exposed; only 2.6% were
"strongly exposed." No data were provided to explain how these exposure categories were derived. The
workers were followed for a median 15.5 years. Without additional information on exposure to EtO,
this study is of little help at this time in evaluating the carcinogenicity of EtO.
NIOSH conducted an industry-wide study of 18,254 workers (45% male and 55% female) in
14 plants where EtO was used (Steenland et al.. 2004; Stavner et al.. 1993; Steenland et al.. 1991). Most
of the workers were exposed while sterilizing medical supplies and treating spices and in the
manufacture and testing of medical sterilizers. Individual exposure estimates were derived for workers
from 13 of the 14 plants. The procedures for selecting the facilities and defining the cohort are
described in Steenland et al. (1991). and the exposure model and verification procedures are described in
Greife et al. (1988) and Hornung et al. (1994). Briefly, a regression model was developed to allow
estimation of exposure levels for time periods, facilities, and operations for which industrial hygiene
data were unavailable. The data for the model consisted of 2,700 individual time-weighted exposure
values for workers' personal breathing zones, acquired from 18 facilities between 1976 and 1985. The
data were divided into two sets, one for developing the regression model and the second for testing it.
Seven out of 23 independent variables tested for inclusion in the regression model were found to be
significant predictors of EtO exposure and were included in the final model. This model predicted 85%
of the variation in average EtO exposure levels. (See Appendix A, Section A.2.8, for more details on
the NIOSH exposure assessment and its evaluation.) Results of the original follow-up study through
1987 are presented in Steenland et al. (1991) and Stavner et al. (1993). The cohort averaged 26.8 years
of follow-up in the extended follow-up study through 1998, and 16% of the cohort had died (Steenland
et al.. 2004).
The overall SMR for cancer was 0.98, based on 860 deaths (Steenland et al., 2004). The SMR
for (lympho)hematopoietic cancer was 1.00, based on 79 cases. Exposure-response analyses, however,
revealed exposure-related increases in hematopoietic cancer mortality risk, although when analyzed by
sex, the effect was primarily in males. In categorical life-table analysis, men with >13,500 ppm-days of
cumulative exposure had an SMR of 1.46 (Obs = 13). In internal Cox regression analyses (i.e., analyses
in which the referent population is within the cohort) with exposure as a continuous variable, statistically
significant trends in males for all hematopoietic cancer (p = 0.02) and for "lymphoid" cancers (NHL,
lymphocytic leukemia, and myeloma; p = 0.02) were observed using log cumulative exposure
3-6
-------
(ppm-days) with a 15-year lag.7 In internal categorical analyses, statistically significant odds ratios
(ORs) were observed in the highest cumulative exposure quartile (with a 15-year lag) in males for all
hematopoietic cancer (OR= 3.42; 95% CI = 1.09-10.73) and "lymphoid" cancer (OR = 3.76; 95%
CI = 1.03-13.64). The exposure metrics of duration of exposure, average concentration, and maximum
(8-hour TWA) concentration did not predict the hematopoietic cancer results as well as did the
cumulative exposure metric.
Although the overall SMR for female breast cancer was 0.99, based on 102 deaths, the NIOSH
mortality follow-up study reported a significant excess of breast cancer mortality in the highest
cumulative exposure quartile using a 20-year lag period compared to the U.S. population (SMR = 2.07;
95% CI = 1.10-3.54; Obs = 13). Internal exposure-response analyses also noted a significant positive
trend for breast cancer mortality using the log of cumulative exposure and a 20-year lag time (p = 0.01).
In internal categorical analyses, a statistically significant OR for breast cancer mortality was observed in
the highest cumulative exposure quartile with a 20-year lag (OR = 3.13; 95% CI = 1.42-6.92).
In summary, although the overall external comparisons did not demonstrate increased risks, the
NIOSH investigators found significant internal exposure-response relationships between exposure to
EtO and cancers of the hematopoietic system, as well as breast cancer mortality. (As stated earlier,
internal comparisons are considered superior to external comparisons in occupational epidemiology
studies because internal comparisons help control for the healthy worker effect and other factors that
might be more comparable within a study's worker population than between the workers and the general
population.) Exposures to other chemicals in the workplace were believed to be minimal or nonexistent.
This study is the most useful of the epidemiologic studies in terms of carrying out a quantitative
dose-response assessment. It possesses more attributes than the others for performing risk analysis
(e.g., good-quality estimates of individual exposure, lack of exposure to other chemicals, and a large and
diverse cohort of workers).
It should be noted that Steenland et al. (2004) used Cox regression models, which are log-linear
relative rate (RR) models, thus providing some low-dose sublinear curvature for doses expressed in
terms of cumulative exposure. However, the best-fitting dose-response model for both male lymphoid
cancers and male all hematopoietic cancers was for dose expressed in terms of log cumulative exposure,
indicating supralinearity of the low-dose data. Supralinearity of the dose-response data was also
indicated by the categorical exposure results. This is in contrast to the reported results of Rinnan et al.
(2004) based on the Teta et al. (1999) analysis combining the 1993 UCC leukemia data with the 1993
NIOSH leukemia data, which are claimed by the authors to provide empirical evidence supporting a
quadratic dose-response relationship. The 2004 NIOSH dose-response data for hematopoietic cancers
7The sex difference is not statistically significant, however, and the trends for both sexes combined are also statistically
significant [p = 0.01 andp = 0.02, respectively; see Tables D-31 and D-48 in Appendix D],
3-7
-------
clearly do not provide empirical evidence in support of a quadratic dose-response relationship. On the
contrary, the NIOSH data suggest a supralinear dose-response relationship in the observable range.
Wong and Trent (1993) investigated the same cohort as Steenland et al. (1991) but added
474 new subjects (unexplained) and increased the follow-up period by 1 year. The study authors
incremented the total number of deaths by 176 and added 392.2 more expected deaths. The only
positive finding was a statistically significantly increased risk of NHL among men (SMR = 2.5; Obs = 6;
p < 0.05). However, there was a deficit risk of NHL among women. For breast cancer, there was no
trend of increasing risk by duration of employment or by latency. This study has major limitations, not
the least of which is a lack of detailed employment histories, making it impossible to quantify individual
exposures and develop dose-response relationships. Furthermore, the addition of more than twice as
many expected deaths as observed deaths makes the analysis by the authors questionable.
Valdez-Flores et al. (2010) conducted alternative Cox proportional hazards modeling and
categorical exposure-response analyses using data from the UCC cohort (Swaen et al.. 2009). the
NIOSH cohort (Steenland et al.. 2004). and the two cohorts combined, analyzing the sexes both
separately and together. These investigators reported that they found no evidence of exposure-response
relationships for cumulative exposure with either the Cox model or categorical analyses for all of the
cohort/endpoint data sets examined (endpoints included all lymphohematopoietic cancers, lymphoid
cancers, and female breast cancer, the latter in the NIOSH cohort only). Valdez-Flores et al. (2010) did
observe statistically significant increases in response rates in the highest exposure quintile relative to the
lowest exposure quintile for lymphohematopoietic and lymphoid cancers in males in the NIOSH cohort,
consistent with the categorical results of Steenland et al. (2004). as well as a statistically significant
increase in the highest exposure quintile for lymphoid cancers in males and females combined in the
NIOSH cohort, consistent with the results in Appendix D. Because the exposure assessment conducted
for the UCC cohort is much cruder (see above and Appendix A.2.20), especially for the highest
exposures, than the NIOSH exposure assessment (which was based on a validated regression model; see
Appendix A.2.8), the EPA considers the results of exposure-response analyses of the combined cohort
data to have greater uncertainty than those from analyses of the NIOSH cohort alone, despite the
additional cases contributed by the UCC cohort (e.g., the UCC cohort contributes 17 cases of lymphoid
cancer to the 53 from the NIOSH cohort). Furthermore, Valdez-Flores et al. (2010) did not use any log
cumulative exposure models, and these were the models that were statistically significant in the
Steenland et al. (2004) analyses, consistent with the apparent supralinearity of the NIOSH
exposure-response data. See Appendix A.2.20 for a more detailed discussion of the Valdez-Flores et al.
(2010) analyses and how they compared with the Steenland et al. (2004) analyses.
In a mortality study of 1,971 male chemical workers in Italy, 637 of whom were licensed to
handle EtO but not other toxic gases, Bisanti et al. (1993) reported statistically significant excesses of
hematopoietic cancers (SMR = 7.1, Obs = 5,^<0 .05). The study was limited by the lack of exposure
3-8
-------
measurements and by the young age of the cohort. Although this study suggests that exposure to EtO
leads to a significant excess of hematopoietic cancer, the lack of personal exposure measurements and
the fact that members were potentially exposed to other chemicals in the workplace lessen the study's
usefulness for establishing the carcinogenicity of EtO.
Hagmar et al. (Hagmar et al.. 1995; Hagmar et al.. 1991) studied cancer incidence in
2,170 Swedish workers (861 male and 1,309 female) in two medical sterilizing plants. They determined
concentrations in six job categories and estimated cumulative exposures for each worker. They found
hematopoietic cancers in 6 individuals versus 3.4 expected (SMR = 1.8) and a nonsignificant doubling
in the risk when a 10-year latency period was considered. Despite the cohort being young, the follow-up
time being short (median 11.8 years), and only a small fraction of the workers being highly exposed, the
report is suggestive of an association between EtO exposure and hematopoietic cancers. The risk of
breast cancer was less than expected, although with such short follow-up, the total numbers of cases was
small (standardized incidence ratio [SIR] = 0.5, Obs = 5).
More recently, a follow-up study of the Hagmar et al. (Hagmar et al.. 1995; Hagmar et al.. 1991)
cohort was published, with an additional 16 years of follow-up (Mikoczv et al.. 2011).8 For
lymphohematopoietic cancers, nonsignificant increases in SMRs and SIRs were reported, and the
internal incidence analysis showed no exposure-related association. However, this analysis is relatively
uninformative for these cancers, given the small number of cases (five cases in each of the two highest
exposure quartiles and seven cases in the referent group of workers with cumulative exposures below
the median), the generally low estimated cumulative exposures, and the absence of an unexposed
referent group. For breast cancer incidence (41 cases), SIRs were nonsignificantly decreased, both with
and without a 15-year induction period. Internal analyses resulted in statistically significant increases in
the incidence rate ratios for the two highest cumulative exposure quartiles as compared to the 50% of
workers with cumulative exposures below the median, despite having a low-exposed rather than an
unexposed referent group.
In a large chemical manufacturing plant in Belgium (number of employees not stated), Swaen et
al. (1996) performed a nested case-control study of Hodgkin lymphoma to determine whether a cluster
of 10 cases in the active male work force was associated with any particular chemical. They found a
significant association for benzene and EtO. This study is limited by the exclusion of inactive workers
and the potential confounding effect of chemicals other than EtO, and it is not useful for quantitative
dose-response assessment.
01 sen et al. (1997) studied 1,361 male employees working in the ethylene and propylene
chlorohydrin production and processing areas located within the EtO and propylene oxide production
8This follow-up study was published after the general cutoff date for literature inclusion in this assessment and is reviewed in
detail in Section J.2.2 of Appendix J. However, as it is a follow-up of an earlier study and as, with the additional follow-up,
it provides important corroborating evidence, the study is also briefly mentioned here.
3-9
-------
plants at four Dow Chemical Company sites in the United States. Although these investigators found a
nonsignificant positive trend between duration of employment as chlorohydrin workers and
lymphohematopoietic cancer (Obs = 10), they concluded that there was no appreciable risk in these
workers, in contrast to the findings of Benson and Teta (1993). The small cohort size and the lack of
data on EtO exposures limit the usefulness of this study in inferring risks due to EtO.
Ambroise et al. (2005) studied cancer mortality in a small cohort of 181 male municipal
pest-control workers in France, 140 of whom were exposed to EtO, along with a wide variety of other
chemicals, between 1979 and 1994. Because of the small cohort size and limited follow-up, few deaths
were observed or expected for individual cancer sites (e.g., only one lymphohematopoietic cancer death
was observed—1 leukemia vs. 0.23 expected), and the site-specific data were reported only for the full
cohort and not just the EtO-exposed workers; thus, this study was not considered further.
Norman et al. (1995) studied 1,132 workers (204 male and 928 female) in a medical sterilizing
plant in the United States. There was a significant excess incidence of breast cancer (SIR = 2.6,
Obs = 12 ,p< 0.05) in the women; no other cancer sites were elevated. The risk of breast cancer was not
reported to be excessive in the few previous studies that had adequate numbers of females and included
analysis for breast cancer; however, only one of these studies was also an incidence study. The
follow-up time was too short to draw meaningful conclusions, and the study lacks the power to
determine whether risks for cancers other than breast cancer are statistically significantly elevated. The
study has no information regarding historical exposure, and some subjects with breast cancer had
worked for less than 1 month.
Tompa et al. (1999) reported a cluster of 8 breast cancers and 8 other cancers in 98 nurses
exposed to EtO in a hospital in Hungary; however, the expected number of cases cannot be identified.
The NIOSH investigators used the NIOSH cohort to conduct a study of breast cancer incidence
and exposure to EtO (Steenland et al., 2003). The researchers identified 7,576 women from the initial
cohort who had been employed in the commercial sterilization facilities for at least 1 year (76% of the
original cohort). Breast cancer incidence was determined from interviews (questionnaires), death
certificates, and cancer registries. Interviews were obtained for 5,139 women (68% of the study cohort).
The main reason for nonresponse was inability to locate the study subject (22% of cohort). The average
duration of exposure for the cohort was 10.7 years. For the full study cohort, 319 incident breast cancer
cases were identified, including 20 cases of carcinoma in situ. Overall, the SIR was 0.87
(0.94 excluding the in situ cases) using Surveillance, Epidemiology, and End Results (SEER) reference
rates for comparison. Results with the full cohort are expected to be underestimated, however, due to
case under-ascertainment in the women without interviews. A significant exposure-response trend was
observed for SIR across cumulative exposure quintiles, using a 15-year lag time (p = 0.002). In internal
Cox regression analyses, with exposure as a continuous variable, a significant trend for breast cancer
incidence was obtained for log cumulative exposure with a 15-year lag (p = 0.05), taking age, race, and
3-10
-------
year of birth into account. Using duration of exposure, lagged 15 years, provided a slightly better fit
(p = 0.02), while models with cumulative (nontransformed), maximum or average exposure did not fit as
well. In the Cox regression analysis with categorical exposures and a 15-year lag, the top cumulative
exposure quintile had a statistically significant OR for breast cancer incidence of 1.74 (95%
CI= 1.16-2.65).
In the subcohort with interviews, 233 incident breast cancer cases were identified. Information
on various risk factors for breast cancer was also collected in the interviews, but only parity and breast
cancer in a first-degree relative turned out to be important predictors of breast cancer incidence. In
internal analyses with continuous exposure variables, the model with duration of exposure (lagged
15 years) again provided the best fit (p = 0.006). Both the cumulative exposure and log cumulative
exposure models also yielded significant regression coefficients with a 15-year lag (p = 0.02 and
p = 0.03, respectively), taking age, race, year of birth, parity, and breast cancer in a first-degree relative
into account. In the Cox regression analysis with categorical exposures and a 15-year lag, the top
cumulative exposure quintile had a statistically significant OR of 1.87 (95% CI = 1.12-3.10).
Steenland et al. (2003) suggest that their findings are not conclusive of a causal association
between EtO exposure and breast cancer incidence due to inconsistencies in exposure-response trends,
possible biases due to nonresponse, and an incomplete cancer ascertainment. Although that conclusion
seems appropriate, those concerns do not appear to be major limitations. As noted by the authors, it is
not uncommon for positive exposure-response trends to exhibit fluctuations and not be strictly
monotonically increasing, possibly due to random variation or imprecision in exposure estimates.
Furthermore, the consistency of results between the full study cohort, which is less subject to
nonresponse bias, and the subcohort with interviews, which should have full case ascertainment,
alleviates some of the concerns about those potential biases.
In a study of 299 female workers employed in a hospital in Hungary where EtO sterilizers were
used, Kardos et al. (2003) observed 11 cancer deaths, including 3 breast cancer deaths, compared with
slightly more than 4 expected total cancer deaths. Site-specific expected deaths are not available in this
study, so RR estimates cannot be determined. However, the observation of 3 breast cancer deaths, with
at most 4.4 (with Hungarian national rates as the referent) total cancer deaths expected, is indicative of
3-11
-------
an increased risk of breast cancer,9 and this characterization is supported by the reference of Major et al.
(2001) to a cluster of breast cancer cases in female nurses at the same hospital.
3.1.1. Conclusions Regarding the Evidence of Cancer in Humans
Most of the human studies suggest a possible increased risk of lymphohematopoietic cancers and
female breast cancer, but the total weight of the epidemiological evidence does not provide conclusive
proof of causality. Of the seven relevant criteria of causality envisioned by Hill (1965). temporality,
coherence, biological plausibility, and analogy are clearly satisfied. There is also evidence of
consistency in the response and of a dose-response relationship (biological gradient). On the other hand,
most of the relative risk estimates are not large in magnitude, so the evidence of strength is more limited.
See Section 3.5.1 for a more detailed discussion of the Hill criteria as applied to the EtO database.
The large NIOSH cohort of workers at 14 sterilization plants around the country provides the
strongest evidence of carcinogenicity (Steenland et al.. 2004; Stavner et al.. 1993; Steenland et al..
1991). A statistically significant positive trend was observed in the risk of lymphohematopoietic
neoplasms with increasing (log) cumulative exposure to EtO, although the results for this model were
reported only for males (the sex difference is not statistically significant, however, and the trend for both
sexes combined is statistically significant; see Appendix D). Despite limitations in the data, most other
epidemiologic studies have also found elevated risks of lymphohematopoietic cancer from exposure to
EtO (summarized briefly in Section 3.1 and Table 3-1; see also Appendix A for more details, in
particular Table A-5 for a summary of study results and limitations). Furthermore, when the exposure is
relatively pure, such as in sterilization workers, there is an elevated risk of lymphohematopoietic cancer
that cannot be attributed to the presence of confounders such as those that could potentially appear in the
chlorohydrin process. Moreover, the studies that do not report a significant lymphohematopoietic
cancer effect from exposure to EtO have major limitations, such as small numbers of cases and
inadequate exposure information (see Table A-5 in Appendix A).
In addition, the NIOSH studies have reported increases in the risk of both breast cancer mortality
and incidence in women (Steenland et al.. 2004; Steenland et al., 2003). Other studies have also
reported increases in the risk of breast cancer in women exposed to EtO at commercial sterilization
plants (Mikoczv et al.. 201 1; Norman et al.. 1995) as well as in Hungarian hospital workers exposed to
Hungarian age-standardized female cancer mortality rates reported by the International Agency for Research on Cancer
(http://eu-cancer.iarc.fr/EUCAN/Countrv.aspx7ISOCountrv Cd=348) suggest that the ratio of breast cancer deaths to total
cancer deaths in Hungarian females is about 0.14 (23.5/100,000 breast cancer mortality rate versus 163.6/100,000 total cancer
mortality rate). A comparison of this general population ratio with the ratio of 0.68 for breast cancer to total cancer mortality
in the Kardos et al. (2003) study is necessarily crude because the general population ratio is not based on the age-standardized
rates that would correspond to the age distribution of the person-time of the women in the study, which are unknown.
However, the large difference between the ratios (0.68 for the study versus 0.14 for the general population) indicates an
increased risk of breast cancer in the study.
3-12
-------
EtO (Kardos et al.. 2003). In two other studies where exposure to EtO would be expected to have
occurred among female employees, either no elevated risks were seen (Coggon et al.. 2004) or breast
cancer results were not reported (Hogstedt. 1988; Hogstedt et al.. 1986). However, these latter studies
had far fewer cases to analyze than did the NIOSH studies, they did not have individual exposure
estimates, and they relied on external comparisons (see Table 3-2 for a brief summary and Table A-5 in
Appendix A for more details). The Steenland et al. (2004) and Steenland et al. (2003) studies, on the
other hand, used the largest cohort of women potentially exposed to EtO and clearly show significantly
increased risks of breast cancer incidence and mortality based upon internal exposure-response analyses.
In summary, the largest database of evidence pertaining to a cancer risk from human exposure to
EtO is for cancers of the lymphohematopoietic system. Increases in the risk of lymphohematopoietic
cancer are present in most of the studies, manifested as an increase in leukemia and/or cancer of the
lymphoid tissue. The few studies that fail to demonstrate any increased risks of cancer do not have
those strengths of study design that give confidence to the reported lack of an exposure-related effect.
The evidence of lymphohematopoietic cancer is strongest in the one study (the NIOSH study) that
appears to possess the fewest limitations. In this large study, a significant dose-response relationship
was evident with cumulative exposure to EtO. However, this effect was observed primarily in males,
and the magnitude of the effect was not large. Similarly, in most of the other studies, the increased risks
are not great, and other chemicals in some of the workplaces cannot be ruled out as possible
confounders. Thus, the findings of increased risks of lymphohematopoietic cancer in the NIOSH and
other studies cannot conclusively be attributed to exposure to EtO.
There is also strong evidence of an elevated risk of breast cancer from exposure to EtO in a few
studies. The clearest evidence again comes from the large NIOSH studies, which found positive
exposure-response relationships for both breast cancer incidence and mortality (319 incident breast
cancer cases; 103 breast cancer deaths). In addition, a recent follow-up study of a Swedish cohort of
sterilizer workers reported significant increases in the incidence rate ratios in the highest two cumulative
exposure quartiles compared to the workers with cumulative exposures below the median. Of the four
other studies that included females, none approached the size of the NIOSH studies in terms of breast
cancer data—the study with the next largest breast cancer database had only 12 cases. Nonetheless, two
of the four other studies were supportive of an increased risk of breast cancer.
3-13
-------
Table 3-1. Epidemiological studies of ethylene oxide and human cancer—lymphohematopoietic cancer results3
Study/population/
industry
Number of
subjects
Lymphohematopoietic cancer results
Comments
Hoestedt (1988) and
Hoestedt et al. (1986).
Sterilizers, production
workers, Sweden.
709
(539 men, 170
women)
Cancer deaths Observed Expected SMR (95% CI)
leukemia (ICD-8 204-207) 7 0.8 9.2 (3.7, 19)b
lymphohematopoietic 9 2.0 4.6 (2.1,8.7)b
(ICD-8 200-208)
Insufficient follow-up; 12.0% of cohort had
died (85 deaths).
Exposure to other chemicals.
Coeeonetal. (2004).
Update of Gardner et al.
(1989).
Sterilizing workers in eight
hospitals and users in four
companies, Great Britain.
2,876
(1,864 men,
1,012 women)
Cancer deaths Observed Expected SMR (95% CI)
leukemia (ICD-9 204-208) 5 4.6 1.1 (0.35,2.5)
leukemia 5 2.6 1.9 (0.62,4.5)b
(definite or continual exposure)
NHL (ICD-9 200 + 202) 7 4.8 1.5 (0.58, 3.0)b
lymphohematopoietic 17 12.9 1.3 (0.77,2.1)b
(ICD-9 200-208)
Short follow-up; 19.6% of cohort had died
(565 deaths).
Exposure to other chemicals.
Kiesselbachetal. (1990).
Productio n workers
(methods unspecified)
from eight chemical plants
in West Germany.
2,658 men
Cancer deaths Observed Expected SMR (95% CI)
leukemia (ICD-9 204-208) 2 2.35 0.85 (0.10,3.1)
lymphohematopoietic 5 5 1.0 (0.32,2.3)
(ICD-9 200-208)
Insufficient follow-up; 10.1% of cohort had
died (268 deaths).
Exposure to other chemicals.
Benson and Teta (1993).
Follow-up of only the
chlorohydrin-exposed
employees from Greenbere
etal. (1990) cohort.
Production workers at a
chemical plant in West
Virginia.
278 men
Cancer deaths Observed Expected SMR (95% CI)
leukemia and aleukemia 4 1.14 3.5 (0.96,8.9)
lymphosarcoma 1 0.50 2.0(0.05,11)
and reticulosarcoma
lymphohematopoietic 8 2.72 2.9 (1.3,5.8)
(ICD NS)
EtO exposures reported to be low in the
chlorohydrin process.
Exposure to other chemicals.
Very small cohort; thus, small numbers of
specific cancers despite long follow-up
(52.9% had died; 147 deaths).
-------
Table 3-1. Epidemiological studies of ethylene oxide and human cancer—lymphohematopoietic cancer results3 (continued)
Study/population/
industry
Number of
subjects
Lymphohematopoietic cancer results
Comments
Swaenetal. (2009).
Update of Teta et al.
(1993) IGreenberuet al.
(1990) cohort minus all
chlorohydrin-exposed
employees] plus cohort
enumeration extended an
additional 10 years, adding
167 workers.
Production workers and
users at two chemical
plants in West Virginia.
2,063 men
Cancer deaths Observed Expected SMR (95% CI)
leukemia 11 11.8 0.93 (0.47,1.7)
leukemia 9 NR 1.5 (0.69,2.9)
(in workers hired before 1956)
NHL 12 11.5 1.05 (0.54,1.8)
lymphohematopoietic 27 30.4 0.89 (0.59, 1.3)
(ICD NS)
Internal Cox regression analyses:
No statistically significant trends were observed for lymphoid (17 deaths) or
leukemia (11 deaths) cancer categories for continuous cumulative exposure.
Small cohort; long follow-up time (50.8%
had died; 1,048 deaths).
Crude exposure assessment, especially for
the early time periods.
Exposure to other chemicals.
Steealand et al. (2004).
Update of Steenland et al.
(1991). Stavner et al.
(1993).
Sterilizers of medical
equipment and spices; and
manufacturers and testers
of medical sterilization
equipment, in 14 plants in
the United States.
18,254
(45% male,
55% female)
Cancer deaths Observed Expected SMR (95% CI)
leukemia (ICD-9 204-208) 29 NR 0.99 (0.71,1.36)
NHL (ICD-9 200 + 202) 31 NR 1.00 (0.72,1.35)
lymphohematopoietic 79 NR 1.00 (0.79, 1.24)
(ICD-9 200-208)
Internal Cox regression analyses:
"lymphoid" cancers (ICD-9 200, 202, 203, 204): OR= 3.0 (p = 0.046)
in highest cumulative exposure group, with 15-yr lag; significant regression
coefficient for continuous log cumulative exposure (p = 0.02).
lymphohematopoietic cancer (ICD-9 200-208): OR = 2.96 (p = 0.03)
in highest cumulative exposure group, with 15-yr lag; significant regression
coefficient for continuous log cumulative exposure (p = 0.009).
Large cohort; thus, substantial number of
deaths (2,852) despite short follow-up
(15.6% had died).
High-quality exposure assessment.
No evidence of exposure to other
occupational carcinogens.
No increase in lymphohematopoietic cancer
risk with increase in exposure in women
Results from internal Cox regression
analyses for both sexes combined from
Sections D.3 and D.4 of Appendix D.
Bisanti etal. (1993).
Chemical workers licensed
to handle EtO and other
toxic chemicals, Italy.
1,971 men
Cancer deaths Observed Expected SMR (95% CI)
leukemia (ICD-9 204-208) 2 1.0 1.9(0.23,7.0)
lymphosarcoma and 4 0.6 6.8 (1.9,17)
reticulosarcoma (ICD-9 200)
lymphohematopoietic 6 2.4 2.5 (0.91,5.5)
(ICD-9 200-208)
In group only licensed to handle EtO (n = 637):
leukemia 2 0.3 6.5 (0.79,23)
lymphosarcoma and 3 0.2 17 (3.5,50)
reticulosarcoma
lymphohematopoietic 5 0.7 7.0 (2.3, 16)
Insufficient follow-up; 3.9% of cohort had
died (76 deaths).
Exposure to other chemicals.
-------
Table 3-1. Epidemiological studies of ethylene oxide and human cancer—lymphohematopoietic cancer results3 (continued)
Study/population/
industry
Number of
subjects
Lymphohematopoietic cancer results
Comments
Mikoc/v ct al. (2011).
Update of Haemar et al.
(1995) and Haemar et al.
(1991).
Two plants that produced
disposable medical
equipment, Sweden
2,171
(862 men,
1,309 women)
Cancer cases Observed Expected SIR (95% CI)
leukemia (ICD-7 204-205) 5 3.58 1.40 (0.45,3.26)
NHL (ICD-7 200 + 202) 9 6.25 1.44(0.66,2.73)
lymphohematopoietic 18 14.4 1.25 (0.74, 1.98)
(ICD-7 200-209)
In internal analyses of lymphohematopoietic cancer: IRR (95% CI)
0-0.13 ppm-yr (n = 1,039; 7 cases) 1.00
0.14-0.21 ppm-yr (« = 486; 5 cases) 1.17 (0.36, 3.78)
>0.22 ppm-years (n = 495; 5 cases) 0.92 (0.28, 3.05)
Small, young cohort (171 deaths; 203 cancer
cases).
Estimated cumulative exposures were
generally low.
There was no unexposed referent group for
internal analyses.
Norman et al. (1995).
Sterilizers of medical
equipment and supplies
that were assembled at this
plant, New York
1,132
(204 men,
928 women)
Cancer cases Observed Expected SIR (95% CI)
leukemia (ICD NS) 1 0.54 1.85 (0.05, 10)b
Short follow-up period and small cohort
(only 28 cancer cases).
Swaenetal. (1996).
Nested case-control study;
cases and controls from a
large chemical production
plant, Belgium.
10 cases of
Hodgkin
lymphoma
(3 exposed;
7 confirmed)
and
200 controls;
all male
Cancer OR (95% CI)
Hodgkin lymphoma (ICD 201) 8.5 (1.4, 40)
Hypothesis-generating study to investigate a
cluster of Hodgkin lymphomas observed at
a chemical plant.
Exposure to other chemicals.
Olsenetal. (1997).
Four EtO production
plants (chlorohydrin
process) in three states.
1,361 men
Cancer deaths Observed Expected SMR (95% CI)
leukemia (ICD-8 204-207) 2 3.0 0.67(0.08,2.4)
lymphosarcoma and 1 1.1 0.91(0.02,5.1)
reticulosarcoma (ICD-8 200)
lymphohematopoietic 10 7.7 1.3 (0.62,2.4)
(ICD-8 200-209)
Short follow-up and small cohort; 22.0%
had died; 300 deaths.
Exposure to other chemicals.
-------
Table 3-1. Epidemiological studies of ethylene oxide and human cancer—lymphohematopoietic cancer results3 (continued)
Study/population/
industry
Number of
subjects
Lymphohematopoietic cancer results
Comments
Kardos etal. (2003).
Female workers from
pediatric clinic of hospital
inEger, Hungary.
299 women
1 lymphoid leukemia death; expected number not reported.
Short follow-up period and small cohort
(11 cancer deaths).
Possible exposure to natural radium, which
permeates the region
Extracted from Table A-5 of Appendix A, with addition of some summary results (e.g., SMRs); see Table A-5 and Appendix A for more study details.
bCalculated by the EPA assuming Poisson distribution
ICD NS = ICD codes not specified; NR = not reported; IRR = internal incidence ratios.
-------
Table 3-2. Summary of epidemiological results on ethylene oxide and female breast cancer (all sterilizer workers)3
Study
Number of
women
Breast cancer results
Comments
Hoestedt et al.
(1986) and Ho estedt
r 19881
Swedish incidence
and mortality study
170
NR
Eight deaths (seven from cancer) had
occurred among the women; breakdown
by cancer type not reported.
Coeeonetal. (2004)
Great Britain
mortality study
1,011 women
hospital workers
Exposure category Observed Expected SMR (95% CI)
Continual 5 7.2
Intermittent 0 0.7
Unknown 6 5.2
ALL 11 13.1 0.84 (0.42,1.51)
11 breast cancer deaths.
14% of the cohort of 1,405 (including
males) hospital workers had died.
Steenland et al.
(2004)
U.S. mortality study
9,908
SMR in highest quartile of cumulative exposure (with20-yr lag) = 2.07 (p < 0.05).
Significant Cox regression coefficient for log cumulative exposure (20-yr lag)
(p = 0.01).
103 breast cancer deaths.
Steenland et al.
(2003)
U.S. breast cancer
incidence study;
nested within
Steenland et al.
(2004) cohort
7,576 employed
for>l yr; 5,139
with interviews
Full cohort results'.
Cox regression analysis OR = 1.74 (95% CI: 1.16, 2.65) for highest cumulative
exposure quintile (15-yr lag).
p = 0.05 for regression coefficient with log cumulative exposure (15-yr lag).
Subcohort results:
Cox regression analysis OR = 1.87 (95% CI: 1.12,3.10) for highest cumulative
exposure quintile (15-yr lag).
p = 0.02 for regression coefficient with cumulative exposure (15-yr lag); p = 0.03
with log cumulative exposure (15-yr lag).
319 cases in full cohort.
233 cases in subcohort with interviews.
Mikoczv et al.
(2011).
Update of Hasmar et
al. (1995) and
Hasmar etal. (1991).
Swedish cancer
incidence study
1,309
41 cases vs. 50.9 expected SIR = 0.81 (95% CI: 0.58, 1.09).
In internal analyses: IRR(95%CI)
0-0.13 ppm-years (n = 615; 10 cases) 1.00
0.14-0.21 ppm-years (n = 287; 14 cases) 2.76 (1.20, 6.33)
>0.22 ppm-years (n = 295; 17 cases) 3.55 (1.58, 7.93)
41 cases.
-------
Table 3-2. Summary of epidemiological results on ethylene oxide and female breast cancer (all sterilizer workers)3
(continued)
Study
Number of
Women
Breast Cancer Results
Comments
Norman et al. (1995)
928
SIRs ranged from 1.72 (95% CI: 0.99, 3.00) to 2.40 (95% CI: 1.32,4.37)
depending on calendar year of follow-up, assumptions about completeness of
follow-up, and reference rates used.
12 cases.
U.S. cancer
incidence study
Kardos et al. (2003)
299
11 cancer deaths observed compared with 4.38, 4.03, or 4.28 expected (p <0.01),
based on comparison populations of Hungary, Heves County, and city of Eger,
respectively; 3 were breast cancer deaths (i.e., 3 breast cancer deaths vs. ~4.3 total
deaths expected). Although the expected number of breast cancer deaths was not
reported, the number of breast cancer deaths observed for the total deaths expected
is indicative of an increased risk of breast cancer (see footnote 9 in Section 3.1).
Three breast cancer deaths.
Hungarian mortality
study
Extracted from Table A-5 of Appendix A; see Table A-5 and Appendix A for more study details, and also Table 3-1 above.
NR = not reported; OR = odds ratio; SIR = standardized incidence ratio; IRR = internal incidence ratios.
-------
3.2. EVIDENCE OF CANCER IN LABORATORY ANIMALS
The International Agency for Research on Cancer (IARC) monograph (IARC. 1994b) has
summarized the rodent studies of carcinogenicity, and Health Canada (2001) has used this information
to derive levels of concern for human exposure. The EPA concludes that the IARC summary of the key
studies is valid and is not aware of any animal cancer bioassays that have been published since 1994.
The Ethylene Oxide Industry Council (EPIC. 2001) also reviewed the same studies and did not cite
additional studies. The qualitative results are described here and the incidence data are tabulated in the
unit risk derivation section of this document.
One study of oral administration in rats has been published; there are no oral studies in mice.
Dunkelberu (1982) administered EtO in vegetable oil to groups of 50 female Sprague-Dawley rats by
gastric intubation twice weekly for 150 weeks. There were two control groups (untreated and oil
gavage) and two treated groups (7.5 and 30 mg/kg-day). A dose-dependent increase in the incidence of
malignant tumors in the forestomach was observed in the treated groups (8/50 and 31/50 in the low- and
high-dose groups, respectively). Of the 39 tumors, 37 were squamous cell carcinomas, and metastases
to other organs were common in these animals. This study was not evaluated quantitatively because oral
risk estimates are beyond the scope of this document.
One inhalation assay was reported in mice (NTP. 1987). and two inhalation assays were reported
in rats [(Lynch et al., 1984a; Lynch et al., 1984c) in males; (Garman et al., 1986. 1985; Snellings et al.,
1984) in both males and females]. In the National Toxicology Program (NTP) mouse bioassay (NTP,
1987), groups of 50 male and 50 female B6C3Fi mice were exposed to EtO via inhalation at
concentrations of 0, 50, and 100 ppm for 6 hours per day, 5 days per week, for 102 weeks. Mean body
weights were similar for treated and control animals, and there was no decrease in survival associated
with treatment. A concentration-dependent increase in the incidence of tumors at several sites was
observed in both sexes. These data are summarized in Table 3-3. Males had carcinomas and adenomas
in the lung. Females had carcinomas and adenomas in the lung, malignant lymphomas,
adenocarcinomas in the uterus, and adenocarcinomas in the mammary glands. The NTP also reports
that both sexes had dose-related increased incidences of cystadenomas of the Harderian glands, but these
are benign lesions and are not considered further here.
In the Lynch et al. (Lynch et al., 1984a; Lynch et al., 1984c) bioassay in male Fischer 344 (F344)
rats, groups of 80 animals were exposed to EtO via inhalation at concentrations of 0, 50, and 100 ppm
for 7 hours per day, 5 days per week, for 2 years. Mean body weights were statistically significantly
decreased in both treated groups compared with controls (p < 0.05). Increased mortality was observed
in the treated groups, and the increase was statistically significant in the 100-ppm exposure group
(p < 0.01). Lynch et al. (1984a) suggest that survival was affected by a pulmonary infection alone and
in combination with EtO exposure. Concentration-dependent increases in the incidence of mononuclear
cell leukemia in the spleen, peritoneal mesothelioma in the testes, and glioma in the brain were observed
3-20
-------
(see Table 3-4). The fact that the increased incidence of mononuclear cell leukemia was statistically
significant in the low-exposure group, but not in the high-exposure group, is probably attributable to the
increased mortality in the high-exposure group. The increased incidence in just the terminal kill rats in
the 100-ppm group was statistically significant compared with controls.
Table 3-3. Tumor incidence data in National Toxicology Program Study of
B6C3Fi mice (NTT, 1987)a and exposure-response modeling resultsb
Gender/tumor type
EtO concentration
(time-weighted average)0
ECio
(LECio)d,
(mg/m3)
Unit risk
(0.1/LECio)
(per mg/m3)
0 ppm
50 ppm
(16.3 mg/m3)
100 ppm
(32.7 mg/m3)
Males
Lung adenomas plus
carcinomas
11/49
19/49
26/49e
6.94
(4.51)
2.22 x 10"2
Females
Lung adenomas plus
carcinomas
2/44
5/44
22/49f
14.8
(9.12)
1.1 X 10"2
Malignant
lymphoma
9/44
6/44
22/498
21.1
(13.9)
7.18 x 10-3
Uterine
carcinoma
0/44
1/44
5/49h
32.8
(23.1)
4.33 x 10"3
Mammary
carcinoma1
1/44
8/448
6/49
9.69
(5.35)
1.87 x lO-2
aIncidence data were adjusted by the EPA by eliminating the animals that died prior to the occurrence of the first
tumor or prior to 52 wk, whichever was earlier.
Statistical analyses and exposure-response modeling were conducted by the EPA.
0 Adjusted by the EPA to continuous exposure from experimental exposure conditions of 6 hr/d, 5 d/wk;
1 ppm= 1.83 mg/m3.
Calculated by the EPA using Tox Risk program
'-p < 0.01 (pairwise Fisher's exact test).
fp < 0.001 (pairwise Fisher's exact test).
ip < 0.05 (pairwise Fisher's exact test).
hp = 0.058 by pairwise Fisher's exact test compared to concurrent controls; however, uterine carcinomas are rare
tumors in female B6C3Fi mice, andp < 0.0001 by pairwise Fisher's exact test compared to the NTP historical
control incidence of 1/1,077 for inhalation (air) female B6C3Fi mice fed the NIH-07 diet.
'Highest dose was deleted in order to fit a model to the dose-response data.
ECio = effective concentration (modeled) corresponding to a 10% extra risk of tumor incidence; LECio = lower
95% (one-sided) confidence limit on the ECio.
3-21
-------
Table 3-4. Tumor incidence data in Lynch et al. (1984a, c) study of male
F344 rats and exposure-response modeling results
Tumor type
Concentration (time-weighted average)3
ECio
(LECi„)b
(mg/m3)
Unit risk
(0.1/LECio)
(per mg/m3)
0 ppm
50 ppm
(19.1 mg/m3)
100 ppm
(38.1 mg/m3)
Splenic mononuclear
cell leukemia0
24/77
38/79d
30/76
7.11
(3.94)
2.54 x 10"2
Testicular peritoneal
mesothelioma
3/78
9/79
2 l/79e
16.7
(11.8)
8.5 x lO-3
Brain mixed-cell
glioma
0/76
2/77
5/79e
65.7
(37.4)
2.68 x lO"3
aAdjusted by the EPA to continuous exposure from experimental exposure conditions of 7 hr/d, 5 d/wk;
1 ppm= 1.83 mg/m3.
bCalculated by the EPA using Tox Risk program
°Highest dose deleted while fitting the dose-response data.
dp < 0.05 (pairwise Fisher's exact test).
'-p < 0.01 (pairwise Fisher's exact test).
ECio = effective concentration (modeled) corresponding to a 10% extra risk of tumor incidence; LECio = lower
95% (one-sided) confidence limit on the ECio.
In the bioassay conducted by Snellings et al. (1984), 120 male and 120 female F344 rats in each
sex and dose group were exposed to EtO via inhalation at concentrations of 0 (two control groups of
120 rats of each sex were used), 10, 33, and 100 ppm for 6 hours per day, 5 days per week, for 2 years,
with scheduled kills at 6 (10 rats per group), 12 (10 rats per group), and 18 (20 rats per group) months.
Significant decreases in mean body weight were observed in the 100-ppm exposure group in males and
in the 100-ppm and 33-ppm exposure groups in females.
During the 15th month of exposure, an outbreak of viral sialodacryoadenitis occurred, resulting in
the deaths of 1-5 animals per group. Snellings et al. (1984) claim that it is unlikely that the viral
outbreak contributed to the EtO-associated tumor findings. After the outbreak, mortality rates returned
to preoutbreak levels and were similar for all groups until the 20th or 21st month, when cumulative
mortality in the 33-ppm and 100-ppm exposure groups of each sex remained above control values. By
the 22nd or 23rd months, mortality was statistically significantly increased in the 100-ppm exposure
groups of both sexes.
In males, concentration-dependent increases in the incidence of mononuclear cell leukemia in the
spleen and peritoneal mesothelioma in the testes were observed, and in females an increase in
mononuclear cell leukemia in the spleen was seen. These data are summarized in Table 3-5. Note that
these investigators observed the same types of tumors (splenic leukemia and peritoneal mesothelioma)
seen by Lynch et al. (1984a) and Lynch et al. (1984c). Snellings et al. (1984) only report incidences (of
incidental and nonincidental primary tumors for all exposure groups) for the 24-month (terminal) kill.
3-22
-------
However, in their paper they state that significant findings for the mononuclear cell leukemias were also
obtained when all rats were included and that a mortality-adjusted trend analysis yielded positive
findings for the EtO-exposed females (p < 0.005) and males (p < 0.05). Similarly, Snellings et al.
(1984) report that when male rats with unscheduled deaths were included in the analysis of peritoneal
mesotheliomas, the EtO exposure appeared to be associated with earlier tumor occurrence, and a
mortality-adjusted trend analysis yielded a significant positive trend (p < 0.005). In later publications
describing brain tumors (Garman et al.. 1986. 1985). both males and females had a
concentration-dependent increased incidence of brain tumors (see Table 3-5). Garman et al. (1986) and
Garman et al. (1985) report incidences including all rats from the 18- and 24-month kills and all rats
found dead or killed moribund. The earliest brain tumors were observed in rats killed at 18 months.
3.2.1. Conclusions Regarding the Evidence of Cancer in Laboratory Animals
In conclusion, EtO causes cancer in laboratory animals. After inhalation exposure to EtO,
statistically significant increased incidences of cancer have been observed in both rats and mice, in both
males and females, and in multiple tissues (lung, mammary gland, uterus, lymphoid cells, brain, tunica
vaginalis testis). In addition, one oral study in rats has been conducted, and a significant dose-dependent
increase in carcinomas of the forestomach was reported.
3.3. SUPPORTING EVIDENCE
3.3.1. Metabolism and Kinetics
Information on the kinetics and metabolism of EtO has been derived primarily from studies
conducted with laboratory animals exposed via inhalation, although some limited data from humans
have been identified. Details are available in several reviews (Fennel! and Brown, 2001; Csanadv et al..
2000; Brown et al., 1998; Brown et al., 1996).
Following inhalation, EtO is absorbed efficiently into the blood and rapidly distributed to all
organs and tissues. EtO is metabolized primarily by two pathways (see Figure 3-1): (1) hydrolysis to
ethylene glycol (1,2-ethanediol), with subsequent conversion to oxalic acid, formic acid, and carbon
dioxide; and (2) glutathione conjugation and the formation of »Y-(2-hydroxyethyl)cysteine and
N-acetylated derivatives (WHO, 2003). From the available data, the route involving conjugation with
glutathione appears to predominate in mice. In larger species (including humans), the conversion of EtO
is primarily via hydrolysis through ethylene glycol. Because EtO is an epoxide capable of reacting
directly with cellular macromolecules, both pathways are considered to be detoxifying.
3-23
-------
Table 3-5. Tumor incidence data in Snellings et al. (1984) and Garman et al. (1985) reports on F344 ratsa and
exposure-response modeling resultsb
Concentration (time-weighted average)0
ECio
10 ppm
33 ppm
100 ppm
(LECi„)e
Unit risk (0.1/LECio)
Gender/tumor type
0 ppmd
(3.27 mg/m3)
(10.8 mg/m3)
(32.7 mg/m3)
(mg/m3)
(per mg/m3)
Males
Splenic mononuclear cell leukemia
13/97
9/51
12/398
9/308
12.3
1.56 x lO-2
(13%)f
(18%)
(32%)
(30%)
(6.43)
Testicular peritoneal
2/97
2/51
4/39
4/308
22.3
8.66 x 10"3
mesothelioma
(2.1%)
(3.9%)
(10%)
(13%)
(11.6)
Primary brain tumors
1/181
1/92
5/858
7/87h
36.1
4.5 x lO-3
(0.55%)
(1.1%)
(5.9%)
(8.1%)
(22.3)
Females
Splenic mononuclear cell leukemia
11/116
11/548
14/48h
15/261
4.46
3.23 x lO-2
(9.5%)
(21%)
(30%)
(58%)
(3.1)
Primary brain tumors
1/188
1/94
3/92
4/808
63.8
3.07 x lO"3
(0.53%)
(1.1%)
(3.3%)
(5%)
(32.6)
denominators refer to the number of animals for which histopathological diagnosis was performed. For brain tumors, Garman et al. (1985) included animals in the
18-month and the 24-month sacrifice and found dead or euthanized moribund of those alive at the time of the first brain tumor, whereas for the other sites, Snellings et al.
(1984) included animals only at the 24-month sacrifice.
Statistical analyses and exposure-response modeling were conducted by the EPA.
0 Adjusted by the EPA to continuous exposure from experimental exposure conditions of 6 hr/d, 5 d/wk; 1 ppm = 1.83 mg/m3.
dResults for both control groups combined.
Calculated by the EPA using Tox Risk program
fNumbers in parentheses indicate percentage incidence values.
%p < 0.05 (pairwise Fisher's exact test).
hp < 0.01 (pairwise Fisher's exact test).
[p < 0.001 (pairwise Fisher's exact test).
ECio = effective concentration (modeled) corresponding to a 10% extra risk of tumor incidence; LECio = lower 95% (one-sided) confidence limit on the ECio.
-------
CH, CH;
Ethylene oxide
Glutathione / \ Enzymatic and
•/ \
transferase / X
hvdroivsh
GSCH2CH2GH
S-2-(liydroxyethyl-ghitathione)
ir
cys-ch2ch:gh
S-2-(liydroxvetliyl) cvsteine
1
N-acethyl-S-(2-hydroxyethyl)
cysteine
HOCH2CH2OH
1.2-etlianediol
J
HOCHXHO
hvdroxvaceta ldehvde
" I
hqch2co,h
Glvcolic acid
' I
OHCCO;H
Glvoxylic acid
/ \
HCO-H CO:HCO:H
Formic acid Oxalic acid
+
CO,
Figure 3-1. Metabolism of ethylene oxide.
3-25
-------
Among rodent species, there are clear quantitative differences in metabolic rates. The rate of
clearance of EtO from the blood, brain, muscle, and testes was measured by Brown et al. (1998) and
Brown et al. (1996). Clearance rates were nearly identical across blood and other tissues. Following a
4-hour inhalation exposure to 100 ppm EtO in mice and rats, the average blood elimination half-lives
ranged from 2.4 to 3.2 minutes in mice and 11 to 14 minutes in rats. The elimination half-life in humans
is 42 minutes (Filser et al.. 1992). and the half-life in salt water is 4 days (IARC. 1994b).
In a more detailed study in mice, Brown et al. (1998) measured EtO concentrations in mice after
4-hour inhalation exposures at 0, 50, 100, 200, 300, or 400 ppm. They found that blood EtO
concentration increased linearly with inhaled concentrations of less than 200 ppm, but above 200 ppm
the blood concentration increased more rapidly. In addition, glutathione levels in liver, lung, kidney,
and testes decreased as exposures increased above 200 ppm. The investigators interpreted this, along
with other information, to mean that at low concentrations the metabolism and disappearance of EtO is
primarily a result of glutathione conjugation, but at higher concentrations, when tissue glutathione
begins to be depleted, the elimination occurs via a slower nonenzymatic hydrolysis process, leading to a
greater-than-linear increase in blood EtO concentration.
Fennel! and Brown (2001) constructed physiologically based pharmacokinetic (PBPK) models of
uptake and metabolism in mice, rats, and humans, based on previous studies. They reported that the
models adequately predicted blood and tissue EtO concentrations in rats and mice, with the exception of
the testes, and blood EtO concentrations in humans. Modeling 6-hour inhalation exposures yielded
simulated blood peak concentrations and areas under the curve (AUCs) that are similar for mice, rats,
and humans (human levels are within about 15% of rat and mouse levels; see Figure 3-2). In other
words, exposure to a given EtO concentration in air results in similar predicted blood EtO AUCs for
mice, rats, and humans.
These studies show that tissue concentrations in mice, rats, and humans exposed to a particular
air concentration of EtO are approximately equal and that they are linearly related to inhalation
concentration, at least in the range of exposures used in the rodent cancer bioassays (i.e., 100 ppm and
below).
3.3.2. Protein Adducts
EtO forms DNA (see Section 3.3.3.1) and hemoglobin adducts within tissues throughout the
body (Walker et al.. 1992a; Walker et al.. 1992b). Formation of hemoglobin adducts has been used as a
measure of exposure to EtO. The main sites of alkylation are cysteine, histidine, and the N-terminal
valine; however, for analytical reasons, the N-(2-hydroxyethyl)valine adduct is generally preferred for
measurements (Walker et al.. 1990). Walker et al. (1992b) reported measurements of this hemoglobin
adduct and showed how the adduct concentration changes according to the dynamics of red blood cell
turnover. Walker et al. (1992b) measured hemoglobin adduct formation in mice and rats exposed to 0,
3-26
-------
3, 10, 33, 100, and 300 (rats only) ppm of EtO (6 hours/day, 5 days/week, for 4 weeks). Response was
linear in both species up to 33 ppm; at 100 ppm, the slope had increased. The exposure-related decrease
in glutathione concentration in liver, lung, and other tissues observed by Brown et al. (1998) in mice is a
plausible explanation for the increasing rate of hemoglobin adduct formation at higher exposures.
—~-
— mouse
— ¦ -
-rati
—A-
- rat2
—human
EtO exposure concentration (ppm)
Figure 3-2. Simulated blood areas under the curve for EtO following a 6-hour
exposure to EtO from the rat, mouse, and human physiologically based
pharmacokinetic models of Fennell and Brown (2001)
Based on data presented in Fennell and Brown (2001). (Rati and rat2 results use
different values for pulmonary uptake.)
In humans, hemoglobin adducts can be used as biomarkers of recent exposure to EtO (IARC,
2008; Boogaard, 2002; I ARC 1994b), and several studies have reported exposure-response relationships
between hemoglobin adduct levels and EtO exposure levels [e.g., (van Sittert et al., 1993; Schulte et al.,
1992)1. Hemoglobin adducts are good general indicators of exposure because they are stable (DNA
3-27
-------
adducts, on the other hand, may be repaired or fixed as mutations and hence are less reliable measures of
exposure). However, Fost et al. (1991) noted that human erythrocytes showed marked interindividual
differences in the amounts of EtO bound to hemoglobin, and Yon a et al. (2001) reported that levels of
N-(2-hydroxyethyl)valine were approximately twofold greater in persons with a (AY'/"/ /-null genotype
than in those with positive genotypes. Endogenous EtO (see Section 3.3.3.1) also contributes to
hemoglobin adduct levels, making it more difficult to detect the impacts of low levels of exogenous EtO
exposure. In addition, Walker et al. (1993) reported that hemoglobin adducts in mice and rats were lost
at a greater rate than would be predicted by the erythrocyte life span and suggested that EtO exposure
can reduce the lifespan of erythrocytes.
Together, the rodent studies on EtO-hemoglobin adduct levels and on blood EtO concentrations
and tissue glutathione levels (see Section 3.3.1) support the hypothesis that decreasing capacity for rapid
detoxification (i.e., tissue glutathione depletion) results in greater than linear EtO blood content and
protein adduct levels as exposure concentrations exceed 100 ppm.
3.3.3. Genotoxicity
Since the first report of EtO induction of sex-linked recessive lethals in Drosophila (Rapoport.
1948), numerous papers have reported positive genotoxic activity in biological systems spanning a wide
range of assay systems, from bacteriophage to higher plants and animals. Figure 3-3 shows the 203 test
entries in the EPA Genetic Activity Profile database in 2001. In prokaryotes and lower eukaryotes, EtO
induced DNA damage and gene mutations in bacteria, yeast, and fungi and gene conversions in yeast.
In mammalian cells (from in vitro and/or in vivo exposures), EtO-induced effects include unscheduled
DNA synthesis, DNA adducts, gene mutations, sister chromatid exchanges (SCEs), micronuclei, and
chromosomal aberrations. Genotoxicity, in particular increased levels of SCEs and chromosomal
aberrations, has also been observed in blood cells of workers occupationally exposed to EtO. Several
publications contain details of earlier genetic toxicity studies [e.g., (1ARC. 2008; Kolman et al., 2002;
Bolt. 2000; Natarajan et al.. 1995; Preston et al„ 1995; IARC, 1994b; Dellarco et al„ 1990; Ehrenberg
and Hussain. 1981 )1. This review briefly summarizes the evidence of the genotoxic potential of EtO,
focusing primarily on more recent studies that provide information on the mode of action of EtO (see
Appendix C for more details from some individual studies).
3-28
-------
ETHYLENE OXIDE NCEA-01
LED
0.0001
0.001
0.01
0.1
1.0
10
100
1000
10000
10
100
1000
10000
100000
1000000
HID
[This is an updated version of the figure in I ARC (1994b).1
See Appendix B for list of references.
Figure 3-3. Display of 203 data sets, including bacteria, fungi, plants, insects, and
mammals (in vitro and in vivo), measuring the full range of genotoxic endpoints.
NCEA = National Center for Environmental Assessment; LED = lowest effective dose; HID = highest ineffective
dose. Test systems are listed along the horizontal axis in the following order: prokaryotes, lower eukaryotes, plants,
insects, mammalian in vitro, human in vitro, body fluids, mammalian in vivo, human in vivo. Codes for the test
systems (appearing above or below the vertical lines) are described in
https: //mo no graphs .iarc .fr/ENG/Mo no graphs/vol62/mo no62-9 .pdf.
3.3.3.1. DNAAdducts
EtO is a direct-acting Sn2 (substitution-nucleophilic-bimolecular)-type monofunctional
alkylating agent that forms adducts with cellular macromolecules such as proteins (e.g., hemoglobin, see
Section 3.3.2) and DNA (Pauwels and Veulemans, 1998). Alkylating agents may produce a variety of
different DNA alkylation products (Beranek, 1990) in varying proportions, depending primarily on the
electrophilic properties of the agent. Reactivity of an alkylating agent is estimated by its Swain-Scott
75-21 -8
GENETIC ACTIVITY PROFILE
2001
s c
L L
H H
S
A
0
_ S
A
5
d v®
F%
s
z N
F c
R
N-
C
F
q_d
M- M
H H G"+
_L |F
-r HI'
D
L
R
M
-g"R J
D
HU-V
T^H-
H— _
G ^
Ji
ss
AA
79
PROKARY LOW EUK PLNT INS
MAMM VITRO
HUMAN F MAMM VIVO
HU
IARC human carcinogen (group 1: human - limited, animal - sufficient)
3-29
-------
substrate constant (.s-value), which ranges from 0 to 1, and EtO has a high .s-value of 0.96 (Beranek.
1990; Golberu. 1986; Warwick. 1963). Acting by the S\2 mechanism and having a high substrate
constant both favor alkylation at the N7 position of guanine in the DNA (Walker et al.. 1990). The
predominant DNA adduct formed by EtO and other SN2-type alkylating agents is
N7-( 2-hydroxyethyl )guan ine (N7-HEG). After in vitro treatment of DNA with EtO, Segerback (1990)
identified three adducts, N7-HEG, N3-hydroxyethyladenine (N3-HEA), and 06-hydroxyethylguanine
(06-HEG), in the ratios 200:8.8:1 (although in vivo the ratio may be closer to 300:1:1, see below); two
other peaks, suspected of representing other adenine adducts, were also observed at levels well below
that of N7-HEG.
While the N7-HEG adducts are abundant, their mutagenic potential may be minimal, as adducts
in this position are unlikely to interfere with the hydrogen-bonding involved in DNA base-pairing and
can be rapidly depurinated (Boysen et al.. 2009). Imidazole ring-opening of N7-HEG could result in
stable, potentially mutagenic lesions (Solomon. 1999); however, EtO-induced N7-HEG ring-opening
has not specifically evaluated in vivo (IARC. 2008). While less abundant, N3-HEA adducts may inhibit
DNA replication by interacting with the minor groove of the DNA helix, which could lead to strand
scission at the replication fork, and 06-HEG adducts are thought to be highly promutagenic as they can
directly interfere with nucleotide base-pairing, typically resulting in thymine incorporation (Mazon et
al.. 2009). At present, both the identity of the responsible adduct(s) and the mechanism(s) by which
such DNA adducts may induce mutations are unknown. Possible mechanisms include DNA misrepair
and enzymatic or chemical depurination followed by the insertion of the incorrect base [typically an
adenine; see, e.g., IARC (2008); Houle et al. (2006); Tates et al. (1999)1. although the levels of apurinic
sites were not increased in rats following EtO exposure [Rusyn et al. (2005); see also Section 3.4], Such
lesions could also lead to the formation of DNA single-strand breaks and, subsequently, to chromosomal
damage (see Section 3.3.3.3).
In addition to exposures from external sources, EtO is produced endogenously through the
cytochrome P450-mediated conversion of ethylene (Tornqvist, 1996), which itself is produced during
normal physiological processes (see Section C.7 of Appendix C). Such processes reportedly include
oxidation of methionine and hemoglobin, lipid peroxidation of fatty acids, and metabolism of intestinal
bacteria [reviewed in (Bolt. 2000; IARC, 1994a)l. The percentage of endogenous ethylene converted to
EtO is unknown. However, only -3% of exogenous ethylene was converted to EtO in workers exposed
to 0.3 ppm (Tornqvist et al., 1989). Furthermore, a concentration of 3,000 ppm ethylene induced -10%
of the 7-HEVal hemoglobin or N7-HEG DNA adduct levels as did 100 ppm EtO in various rat tissues
(Rusvn et al„ 2005). This evidence suggests that exogenous ethylene exposure is unlikely to contribute
significantly to the effects associated with exposure to exogenous EtO in humans or rodents.
Endogenous production of EtO contributes significantly to background levels of DNA adducts,
making it difficult to detect the impacts of low levels of exogenous EtO exposure on DNA adduct levels.
3-30
-------
For example, in DNA extracted from the lymphocytes of unexposed individuals, mean background
levels of N7-HEG ranged from 2 to 8.5 pmol/mg DNA (Bolt. 1996). Using sensitive detection
techniques and an approach designed to separately quantify both endogenous N7-HEG adducts and
"exogenous" N7-HEG adducts induced by EtO treatment in rats, Marsden et al. (2009) reported
increases in exogenous adducts in DNA of spleen and liver tissue at the lowest dose administered
(0.0001 mg/kg injected intraperitoneal [i.p.] daily for 3 days) and statistically significant linear
dose-response relationships (p < 0.05) for exogenous adducts in all three tissues examined (spleen, liver,
and stomach), although the authors caution that some of the adduct levels induced at low EtO
concentrations are below the limit of accurate quantitation (for further discussion of the dose-response
relationships, see Section 4.5). Note that the whole range of doses studied by Marsden et al. (2009) lies
well below the dose corresponding to the lowestmost lowest-observed-adverse-effect level (LOAEL)
from an EtO cancer bioassay (see Section C.7 of Appendix C). Marsden et al. (2009) also observed
significant increases in endogenous N7-HEG adduct formation in the liver and spleen, but not the
stomach, at the two highest doses (0.05 and 0.1 mg/kg), suggesting that, in addition to direct adduct
formation via alkylation, exogenous EtO can induce endogenous N7-HEG adduct production or inhibit
removal indirectly, and in a tissue-specific manner. Marsden et al. (2009) hypothesized that this indirect
adduct formation by EtO results from the induction of ethylene generation under conditions of oxidative
stress, although tissue ethylene levels were not measured directly.
In experiments with rats and mice exposed via inhalation to EtO at concentrations of 0, 3, 10, 33,
100, or 300 (rats only) ppm for 6 hours per day, 5 days per week, for 4 weeks, Walker et al. (1992a)
measured N7-HEG adducts in the DNA of lung, brain, kidney, spleen, liver, and testes. At 100 ppm, the
adduct levels for all tissues except testes were similar (within a factor of 3), despite the fact that not all
of these tissues are targets for toxicity. The study's data on the persistence of the DNA adducts indicate
that DNA repair rates differ in different tissues. Although Walker et al. (1992a) suggested that N7-HEG
adducts are likely to be removed by depurination forming apurinic/apyrimidinic (AP) sites in DNA, a
later study from the same group showed that EtO-induced DNA damage is repaired without
accumulation of AP sites or involving base excision repair (Rusvn et al., 2005). In rats exposed to
300 ppm, steady-state levels of 06-HEG adducts were detected in both target (brain, spleen) and
nontarget (lung, kidney) tissues, and N3-HEA adducts were observed in the spleen (no other tissues
evaluated); these adducts were not detectable in control animals and were present in the exposed rats at
3-31
-------
levels -250-300 times lower than the N7-HEG levels (Walker et al.. 1992a).10
Two studies provide evidence of N7-HEGDNA adduct formation in human populations
occupationally exposed to EtO, one reporting a modest increase in white blood cells (van Delft et al..
1994) and the other a four- to fivefold increase in granulocytes (Yong et al.. 2007) compared to
unexposed controls. However, these differences were not statistically significant due to high
interindividual variation in adduct levels.
The results from the available studies reporting DNA adduct formation or gene mutation
frequency in humans and/or laboratory animals following exogenous EtO exposure are summarized in
Table 3-6.
10 In a study published after the cutoff date for literature inclusion and described in more detail in Section J.4.1 of Appendix
J, Zhang etal. (2015) exposed male B6C3Fi mice to 0, 100, or 200 ppmEtO for 6 hours/day, 5 days/week, for 12 weeks and
examined the lungs for DNA adducts using more sensitive techniques than those used by Walker etal. (1992a). The Zhang
etal. (2015) study supports the identification of the 06-HEG adduct as a direct product of EtO reactivity and adds coherence
to the available database by observing an exposure-related increase in lung 06-HEG levels at lower concentrations than
previously evaluated (i.e., 100-200 ppmvs. 300 ppm), quantification in another rodent species (i.e., mice vs. rats), and even
detection in the majority of unexposed lung samples (3/5), suggesting that endogenous EtO may be responsible for a low
background level of this potentially mutagenic DNA adduct.
3-32
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
N7-HEG DNA ADDUCTS
Intraperitoneal exposure
Marsden et
al. (2009)
4(4)
Rats/F344
Spleent
3 days; killed
4 hr after last
dose
0, 0.0001,0.0005,0.001*,
0.005*, 0.01*, 0.05*,
0.1*
+
Dose-dependent and linear increases in
exogenous N7-HEG adduct levels in all tissues;
endogenous N7-HEG levels in liver and spleen
increased in the two highest dose groups, but did
not change in the stomach at any dose.
Liver
0, 0.0001,0.0005,0.001,
0.005*, 0.01*, 0.05*,
0.1*
+
Stomach
0, 0.0001, 0.0005*, 0.001*,
0.005*, 0.01*, 0.05*, 0.1*
+
Marsden et
al. (2007)
3(3)
Liver, heart,
colon
1 day; killed
6 hr post-
treatment
0, 0.01,0.1*, 0.5*, 1*
+
Dose-dependent increases inN7-HEG adduct
levels in a tissue-specific manner; adduct levels in
liver significantly higher than those in heart or
colon at 0.1 and 1 mg/kg.
Spleerf
Liver, lung
stomach, heart,
kidney
3 days; killed
2, 4, 8, or 10 hr
after last dose
0, 0.1*, 1*
+
Dose-dependent increases inN7-HEG adduct
levels in a tissue-specific manner; adduct levels
did not accumulate with three doses compared to
a single dose in the same study.
Inhalation exposure
Yons et al.
(2007)
58(6)
Humans
Granulocytes
4.7-6.5 yr
0.003-0.36 (8-hr TWA)
-
Considerable interindividual variation in N7-HEG
adduct levels reported.
van Delft et
al. (1994)
42 (29)
Humans
WBCt
NR
2-5
Increase inN7-HEG adduct levels was observed
inEtO-exposed persons but was not statistically
significant.
Wu et al.
(1999)
2-5 (8-16)
Rats/F344
Brain, spleen^
Liver, lung
4 wk; 6 hr/d,
5 d/wk
0,3*, 10*, 33* or 100*
+
Dose-dependent and linear increases inN7-HEG
adduct levels; lung levels > brain levels > liver
levels > spleen levels at the highest dose.
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals (continued)
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
Walker et
10 (10)
Rats/F344
Braint, spleent
Lung
4 wk; 6 hr/d,
5 d/wk
0, 3, 10*, 33*, 100*
also 300* from Walker et
+
Dose-dependent increases in N7-HEG in all
tissues.
al. (1992a)
al. (1990)
3(3)
tissues from
Walker etal.
(1990)
Brain, spleent
Lung, kidney
1-4 wk; 6 hr/d,
5 d/wk
0, 300*
+
Significant increase in 06-HEG levels observed in
all tissues; levels reached steady state by two
weeks.
3(3)
tissues from
Walker etal.
(1990)
Spleen
4 wk; 6 hr/d,
5 d/wk; killed
0, 1, 3, or
5 days after
exposure
0, 300*
+
N3-HEA levels, evaluated in spleen only, were
elevated immediately and 1 day after exposure;
could not be detected 3 and 5 days after exposure.
van Sittert
1-4 (1-2)
Rat/Lewis
Liver
4 wk; 6 hr/d,
5 d/wk; killed
5,21, 35, or
49 days after
exposure
0, 50*, 100*, 200*
+
Dose-dependent increases inN7-HEG adducts at
5 days postexposure, which persisted up to
35 days postexposure; at 49 days after exposure
(200 ppm group only), adduct levels had
decreased almost to background levels.
etal. (2000)
Rusvnetal.
3-4 (3-4)
Rats/F344
Braint, spleent
Liver
1,3, or 20
days; 6 hr/d,
5 d/wk; killed
2 h after
exposure as
well as 6, 24,
and 72 h after
1-day exposure
0, 100*
+
Duration-dependent increases inN7-HEG adduct
levels; increased rapidly over the first few days,
and then more slowly up to 20 days; brain levels
> spleen levels > liver levels at 20 days; loss of
adducts initially faster in spleen compared to
brain and liver.
(2005)
Walker et
7(7)
Rats/F344
Spleent
4 wk; 6 hr/d,
5 d/wk
0, 200*
+
Significant increase inN7-HEG adducts in
exposed versus control group.
al. (2000)
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals (continued)
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
Walker et
al. (1990)
4-5 (4-5)
Rats/F344
WBCt
Brainf, splcciri"
Liver, lung
kidney, testis
1, 3, 7, 14, 28
days; 6 hr/d,
5 d/wk
0, 300*
+
Duration-dependent increase inN7-HEG adducts
in all tissues; brain levels > lung levels > WB C
levels > spleen levels > kidney levels > liver
levels > testis levels.
WBCt
Brainf, splcciri"
Liver, lung
kidney, testis
4 wk; 6 hr/d,
5 d/wk; killed
1-10 days after
exposure
0, 300*
+
Duration-dependent increase inN7-HEG adduct
in all tissues; loss of N7-HEG initially faster in
WBC and spleen compared to other tissues.
Brain, spleent
Lung, kidney
1, 3, 7, 14 days;
6 hr/d, 5 d/wk
0, 500*
+
Duration-dependent increase inN7-HEG adducts
in all tissues; brain levels > lung levels > spleen
levels > kidney levels.
Wuetal.
(1999)
4-7 (8-9)
Mice/B6C3Fi
Braint, spleent,
lungt
Liver
4 wk; 6 hr/d,
5 d/wk
0,3*, 10*, 33*, or 100*
+
Dose-dependent and linear increases inN7-HEG
adducts in a tissue-specific manner (lung levels >
brain levels > spleen levels > liver levels, at 100
ppm).
Walker et
al. (1992a)
4(4)
Mice/B6C3Fi
Braint, spleent,
lungt
4 wk; 6 hr/d,
5 d/wk
0, 10*, 33*, 100*
+
Dose-dependent increases inN7-HEG adducts
similarly in all tissues; DNA adduct levels 2-3
times lower than identical tissues in concurrently
exposed rats.
Ml 1 A 1 IONS
I nt rape rito lie al c \ posu re
Tates et al.
4(4)
Rats/Lewis
Splenic
lymphocytest
1 day;
evaluated 35 or
42 days
postexposure
0, 10*, 20*, 40*, 80*
+
Hprt mutations: dose-dependent increase in MFs
(1999)
only 35 days postexposure.
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals (continued)
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
Tates et al.
8 (5-6)
1 day;
evaluated 19 or
32 days
postexposure
0, 20*, 40*, 80*
+
Hprt mutations: dose-dependent increase in MFs
(1999)
only 32 days postexposure.
(cont.)
Walker and
4(4)
Mice/B6C3Fi
pups
Splenic T Cells
5 consecutive
days
30*, 60*, 90*, 120*
mg/kg
+
Hprt mutations: dose-dependent increase in MF
Skopek
in exon 3 of hprt gene.
(1993)
Walker and
3(3)
Mice/B6C3Fi
pups
Splenic T Cells
2, 6, or 9 doses
given every
other day
100* mg/kg
+
Hprt mutations: dose-dependent increase in MF
Skooek
in exon 3 of hprt gene; 11 base-pair substitutions
(4 AT and 2 GC transversions, 3 AT and 2 GC
transitions) and seven +1 frameshift mutations in
a run of six consecutive G bases reported.
(1993)
Generoso et
50-75 (25)
Mice/T-stock
Germ cells from
males mated
postexposure to
(SEC x
C57BL)Fi
females
5 wk; once
daily, 5 d/wk
0, 30*, 60* mg/kg
+
Heritable translocation: dose-dependent increase
al. (1980)
in frequency of translocation heterozygotes
Generoso et
12 (12)
Mice/
(101 x C3H)Fi
Germ cells from
males mated
postexposure to
either of
T-stock. (SEC x
C57BL)Fi, (101
x C3H)Fi, or
(C3H x
C57BL)Fi
females
Single injection
0, 150* mg/kg
+
Dominant lethal mutations', all four stocks were
al. (1980)
positive for DLM during 4.5-7.5 days post-
treatment.
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals (continued)
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
In drinking water
Tates et al.
4-5 (4-5)
Rats/Lewis
Splenic
lympltocytcs
30 days;
analyzed 6, 19,
33, 41, and
44 days after
last exposure
88*, 220*, 440*
+
Hprt mutations: dose-dependent increase in MFs.
r 19991
with expression maximum 19, 33, and 41 days
after exposure, for the respective doses.
Inhalation exposure
Tates et al.
7(7)
Humans
PBLst
<5 yr
<0.005-0.02
-
HPRT mutations: cytogenetic effects (SCEs and
r19951
>15 yr
<0.005-0.01
—
MN) also were negative (see Table 3-8); would
have needed 50 subjects/group to detect a 50%
increase inMF.
Major etal.
Budapest
hospital
nurses: 9 (14)
Humans
PBLst
4 yr
2.7 (2.7-10.9)
HPRT mutation and variant frequencv: VF in
(20011
both Eger EtO-exposed group and controls was
higher than both Budapest groups, industrial and
historical controls. Cytogenetic effects (CAs
and/or SCEsI were positive Iscc Table 3-8. Maior
etal. (19961],
Eger hospital
nurses: 27
(10)
15 yr
5.4 (2.7-82)
Tates et al.
Factory
Workers: 15
(15)
Humans
PBLst
3-27 yr
(mean, 12 yr)
17-33
+
HPRT mutations: 60% increase inMF in factory
(19911
workers was statistically significant, but 55%
increase inMF in hospital workers was not.
Hospital
workers: 9
(8)
2-6 yr (mean,
4 yr)
20-25
van Sittert
6-8 (10)
Rat Lewis
Splenic
lympltocytcs
4 wk; 6 hr/d,
5 d/wk
0, 50, 100, 200*
+
Hprt mutations: MF analvzed 21 or 22 davs
etal. (20001
postexposure.
Tates et al.
8 (1-2)
Rats/Lewis
Splenic
lympltocytcs
4 wk; 6 hr/d,
5 d/wk
0, 50*, 100*, 200*
+
Hprt mutations: dose-dependent increases in MF;
(19991
in 200 ppm concentration group, MF 1.4-fold
higher than background levels in controls.
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals (continued)
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
Walker et
7(7)
Rats/F344
Splenic T Cellsf
4 wk; 6 hr/d,
5 d/wk
0, 200*
+
Hprt mutations: -sixfold higher in exposed
al. (2000)
animals compared to unexposed controls.
In drinking water
Embree et
15 (10)
Rats/Long-Evans
Germ cells from
males mated
postexposure to
female rats
4 hi4
1,000*
+
Dominant lethal mutations: mutagenic index
al. f 1977>
positive for first 5 wk of mating.
Recio et al.
4-5 (5)
Mice/B6C3Fi
lacl (TG)
Bone marrow^
12 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100, 200
-
Lacl mutations: at 200 ppm significant increase
(2004)
in AT—>TAtransversion mutations (25.4%) vs.
controls (1.4%).
4-6 (6)
24 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100, 200
-
6(6)
48 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100*, 200*
+
Sisk et al.
4(4)
Mice/B6C3Fi
lacl (TG)
Bone marrow
4 wk; 6 hr/d,
5 d/wk
0, 50, 100, 200
-
Lacl mutations: Negative 2 or 8 wk
(1997)
post-treatment.
Walker et
7-9 (5)
Mice/B6C3Fi
lacl (TG)
Splenic
lvmpltocytcs
4 wk; 6 hr/d,
5 d/wk
0, 50*, 100*, 200*
+
Hprt mutations: dose-dependent increase in MFs.
al. (1997)
Thymic
lymphocytes^
0, 50, 100*, 200*
+
Hprt mutations: dose-dcDcndent increase in MFs:
MF increased at 2 hr after treatment reaching a
peak at 2 wk postexposure in 200 ppm group.
Sisk et al.
4(4)
Mice/B6C3Fi
lacl (TG)
Spleent
4 wk; 6 hr/d,
5 d/wk
0, 50, 100, 200
—
Lacl mutations: Negative 2 or 8 wk
(1997)
post-treatment.
Walker et
7(7)
Mice/B6C3Fi
Splenic T Cells
4 wk; 6 hr/d,
5 d/wk
0, 200*
+
Hprt mutations: -fivefold hi gher than controls:
al. (2000)
necropsied 8 wk postexposure.
Sisk et al.
4(4)
Mice/B6C3Fi
lacl (TG)
Lungt
4 wk; 6 hr/d,
5 d/wk
0, 50, 100, 200*
+
Lacl mutations: MF positive at the highest dose
(1997)
compared to control 8 wk post-treatment.
Recio et al.
5(5)
Mice/B6C3Fi
lacl (TG)
Testes
12 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100, 200
-
Lacl mutations: no increases inanv one specific
(2004)
type of mutation observed at any concentration.
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals (continued)
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
Recio et al.
(2004)
5(6)
24 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100, 200
-
(cont.)
6(6)
48 wk; 6 hr/d,
5 d/wk
0, 25*, 50*, 100*, 200
+
Sisk et al.
4(4)
Mice/B6C3Fi
lacl (TG)
Germ cells
4 wk; 6 hr/d,
5 d/wk
0, 50, 100, 200
-
Lacl mutations: Negative 2 or 8 wk
(1997)
post-treatment.
NTP
50 (50)
Mice/B6C3Fi
108 spontaneous
lung tumors,
and 23 EtO-
induced lung
tumors
102 wk; 6 hr/d,
5 d/wk
0, 50*, 100*
+
Kras mutations (codons 12. 13. and 61): detected
(1987):
Hone et al.
(2007)
in 100% of EtO-induced tumors vs. 25% of
spontaneous tumors; 91% mutations in EtO-
induced tumors were codon 12 Gly —> Val vs.1%
in spontaneous tumors (primarily codon 12 Gly
-> Asp).
27 spontaneous
HG tumors, and
21 EtO-induced
HG tumors
0, 50*, 100*
+
Kras mutations (codons 12. 13. and 61): detected
in 86% of all EtO-induced tumors vs. 7% of
spontaneous tumors; mutations in EtO-induced
tumors were primarily codon 12 Gly —> Cys and
codon 13 Gly —> Argvs. spontaneous tumors
codon61 Gin—> Leu.
6 EtO-induced
uterine
carcinomas t (no
spontaneous
tumors)
0, 50*, 100*
+
Kras mutations (codons 12. 13. and 61): detected
in 83% of all EtO-induced tumors;_mutations in
EtO-induced tumors were primarily codon 13 Gly
—> Gly.
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals (continued)
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
NTP
50 (50)
Mice/B6C3Fi
19 spontaneous
mammary
carcinomas, and
12 EtO-induced
mammary
carcinomas t
102 wk; 6 hr/d,
5 d/wk
0, 50*, 100*
+
v53 mutations (exons 5-8): P53 mutations were
(1987);
Houle et al.
(2006)
induced in a dose-dependent manner, detectable
in 67% of all EtO-induced tumors vs. 58% of
spontaneous tumors; p53 protein expression was
induced in a dose-dependent manner.
+
Hras mutations (codon61): detectable in33% of
EtO-induced tumors vs. 26% of spontaneous
tumors; 75% of EtO-induced tumors bearing I Iras
mutations contained concurrentp53 mutations vs.
40% of spontaneous tumors.
Generoso et
29-44
(29-45)
Mice/
(C3H x ioi)Fi
Germ cells from
males mated
postexposure to
either T-stock
females or (SEC
x C57BL)Fi
8.5 wk; 6 hr/d,
5 d/wk for
6 wk, then
daily for 2.5
wk
0, 165, 204*, 250*, 300*
+ (DLM)
Increases in both DLM and HT levels were
dose-dependent; higher frequencies of
translocations recovered in males mated to
T-stock females.
al. (1990)
0, 165*, 204*, 250*, 300*
+ (HT)
Lewis etal.
(1986)
1,891 (1,348)
progeny
Mice/DBA/2J
Germ cells from
males mated
postexposure to
untreated
C57BL/6J
females
From 7, 24, or
28 wk; 6 hr/d,
5 d/wk
0, 200*
+
(DVEE)
Progeny selected for evaluation based upon
external appearance deviating from expected Fi
phenotype; eight suchvariants were conceived
from exposed mice, while only one was observed
from concurrent controls. Authors estimated
MF = 212/105 progeny.
Generoso et
36-58 (36)
Mice/
(101 x C3H)Fi
Germ cells from
males mated
postexposure to
(C3H x
C57BL)Fi
females
2 or 11 wk;
6 hr/d, 5 d/wk
0, 255*
+
Dominant lethal effects: Duration-dependent
al. (1983)
increase in% dominant lethals.
-------
Table 3-6. Dose-response results for ethylene oxide-induced DNA adducts and mutations in humans and laboratory
animals (continued)
Study3
Number
exposed per
exposure
group
(number of
controls)
Species/Strain
Tissue(s)
assessed
Exposure
duration
Exposure concentration
(ppm) or dose (mg/kg)b
Results
Comments0
Generoso et
16 (16)
Mice/
(101 x C3H)Fi
Germ cells from
males mated
postexposure to
(C3H x
C57BL)Fi
females
4 days; 6 hr/d
0, 300*, 400*, 500*
+
Dominant lethal effects: dose-related nonlinear
al. (1986'l
increase.
Generoso et
16 (16)
Mice/
(101 x C3H)Fi
Germ cells from
males mated
postexposure to
females
6, 3 and 1.5 hr
for the different
exposure
groups,
respectively
0, 300*, 600*, 1,200*
+
Dominant lethal effects: dose-related linear
al. (1986'l
increase.
'+' = positive (statistically significant in exposed vs. controls);= negative (statistically no difference between exposed and controls).
*Concentrations or doses reported by the study authors as associated with effects that were significantly significant atp< 0.05 compared with the corresponding
control group.
^Tissues associated with tumor formation in that species following chronic exposure to EtO.
"Data sorted in the order of endpoint (adduct or mutation), route of exposure, species, tissue (for mutations), dose, and duration.
bThe doses for inhalation are inppm and for all other routes in mg/kgbody weight.
0 Authors' conclusions.
dTreated males mated with untreated females for one week and changed to new set of females every week for 10 wk
hr = hour; wk = week; d = day; N7-HEG = N7-(2-hydroxyethyl)guanine; 06-HEG = 06-(2-hydroxyethyl)guanine; WBC = white blood cells; MF = mutation
frequency; N3-HEA = N3-(2-hydroxyethyl)adenine; VF = variant frequency; hprt = hypo xanthine phosphoribosyl transferase; SCEs = sister chromatid
exchanges; MN = micro nucleus; lacl = lactose-inducible lac operon transcriptional repressor; PBLs = peripheral blood lymphocytes; HG = Harderian gland;
TG = transgenic; DLM = dominant lethal mutation; HT = heritable translocation; DVEE = dominant visible and electrophoretically expressed mutations.
-------
3.3.3.2. Point Mutations
EtO has consistently yielded positive results in in vitro mutation assays from bacteriophage,
bacteria, fungi, yeast, insects, plants, and mammalian cell cultures (including human cells). For
example, EtO induces single base pair deletions and base substitutions in the HPRTgene in human
diploid fibroblasts (Kolman and Chovanec. 2000; Lambert et al., 1994; Bastlova et al., 1993) in vitro.
The results of in vivo studies on the mutagenicity of EtO have also been consistently positive following
ingestion, inhalation, or injection [e.g., Tates et al. (1999)1; these studies are summarized in Table 3-6.
Increases in the frequency of mutations in genes such as Hprt and LacI, which are experimentally
convenient to evaluate as surrogate markers of cancer-associated mutagenesis but unlikely to be directly
involved in cellular transformation themselves (Albertini. 2001), have been observed in T-lymphocytes
(Hprt locus) (Walker et al., 1997) and in bone marrow and testes (LacI locus) (Recio et al.. 2004) from
transgenic mice exposed to EtO via inhalation at concentrations similar to those in carcinogenesis
bioassays with this species (NTP, 1987). At somewhat higher concentrations than those used in the
carcinogenesis bioassays (200 ppm, but for only 4 weeks), increases in the frequency of gene mutations
have also been observed in the lungs of transgenic mice (LacI locus) (Sisk et al., 1997) and in
T-lymphocytes of rats (Hprt locus) (van Sittert et al., 2000; Tates et al., 1999). In in vivo studies with
male mice, EtO also causes heritable mutations and other effects in germ cells (Generoso et al., 1990;
Lewis et al„ 1986).
In a study of 12 mammary gland carcinomas in EtO-exposed B6C3Fi mice from the 1987 NTP
bioassay (NTP. 1987) and 19 mammary gland carcinomas in control B6C3Fi mice from various NTP
bioassays from the same time period, Houle et al. (2006) measured mutation frequencies in exons 5-8 of
the Trp53 tumor suppressor gene (homologous to the human TP53 gene) and in codon 61 of the Hras
proto-oncogene. Mutation frequencies in the mammary carcinomas of EtO-exposed mice were only
slightly increased over frequencies in spontaneous mammary carcinomas (33% of the carcinomas in the
EtO-exposed mice had Hras mutations versus 26% of spontaneous tumors; 67% of the carcinomas in the
EtO-exposed mice had Trp53 mutations versus 58% of spontaneous tumors); however, the tumors in the
EtO-exposed mice exhibited distinctly different mutational spectra in the Trp53 and Hras genes,
compared to the spontaneous tumors, and more commonly displayed concurrent mutations of the two
genes (Houle et al.. 2006). The mutational spectra reported by Houle et al. (2006) in both Trp53 and
Hras indicate that purine bases (i.e., guanine and adenine) were the predominant targets for mutations in
tumors of EtO-exposed mice, while the majority of mutations in spontaneous tumors involved
pyrimidine bases (primarily cytosine) (see Section 3.4.1.3 for further discussion of these mutations).
Furthermore, Houle et al. (2006) detected about sixfold higher levels of p53 protein expression in the
mammary carcinomas of EtO-exposed mice than in spontaneous mammary carcinomas, and there was
an apparent dose-response relationship between EtO exposure level and both p53 protein expression and
Trp53 gene mutation (three of the seven tumors in the 50-ppm exposure group and all five tumors in the
3-42
-------
100-ppm group had increased protein expression; also, three Trp53 gene mutations were found in the
seven tumors in the 50-ppm exposure group and nine were found in the five tumors in the 100-ppm
group).
Some of the same investigators conducted a similar study of Kras mutations (evaluating codons
12, 13 and 61) in lung, Harderian gland, and uterine tumors in EtO-exposed B6C3Fi mice from the NTP
bioassay (Hong et al.. 2007). Substantial increases were observed in Kras mutation frequencies in the
tumors from the EtO-exposed mice. Kras mutations were reported in 100% of the lung tumors from
EtO-exposed mice versus 25% of spontaneous lung tumors (108 NTP control animal tumors, including
8 from the EtO bioassay), in 86% of Harderian gland tumors from EtO-exposed mice versus 7% of
spontaneous Harderian gland tumors (27 NTP control animal tumors, including 2 from the EtO
bioassay), and in 83% of uterine tumors from EtO-exposed mice (there were no uterine tumors in
control mice in the 1986 NTP bioassay and none were examined from other control animals).
Furthermore, a specific Kras mutation, aG->T transversion in codon 12, was nearly universal in lung
tumors from EtO-exposed mice (21/23) but rare in lung tumors from control animals (1/108). Other
specific mutations were also predominant in the Harderian gland and uterine tumors, but too few Kras
mutations were available in spontaneous Harderian gland tumors, and no spontaneous uterine tumors
were examined; thus, meaningful comparisons of mutation incidence between tumors in EtO-exposed
mice and spontaneous tumors could not be made for these sites. While the uterine carcinomas in
EtO-exposed mice contained Kras mutations resulting from C -> T transitions, the primary mutations
reported in the lung and Harderian gland tumors of exposed mice were from G -> C and G -> T
transversions, consistent with purine nucleotides being the predominant target for the Hras and Trp53
mutagenesis in the aforementioned mammary gland carcinomas from EtO-exposed mice (Houle et al..
2006). Interestingly, codon 12 Kras mutations were the most prevalent Kras mutations in lung tumors
of EtO-exposed mice (23/1/1 mutations in codons 12/13/61, respectively), while codon 13 mutations
were more frequent in Harderian gland tumors and uterine carcinomas (9/16/5 and 0/5/not determined,
respectively), suggesting that some tissue-specific factors may regulate the formation and resolution of
EtO-induced genotoxicity or facilitate the neoplastic progression of clones bearing specific mutations
over others. Overall, these data strongly suggest that EtO-induced mutations in proto-oncogenes (Kras,
Hras) and tumor-suppressor genes (Trp53) play a role in EtO-induced carcinogenesis in multiple tissues.
Only a few studies have investigated gene mutations in people occupationally exposed to EtO.
In one study, HPRTmutant frequency in peripheral blood lymphocytes was measured in a group of
9 EtO-exposed hospital workers, a group of 15 EtO-exposed factory workers, and their respective
controls (Tates et al., 1991). EtO exposure scenarios suggest higher exposures in the factory workers,
and this is supported by the measurement of higher hemoglobin adduct levels in those workers. HPRT
mutant frequencies were increased by 55% in the hospital workers, but the increase was not statistically
significant. In the factory workers, a statistically significant increase of 60% was reported. In a study of
3-43
-------
workers in an EtO production facility (Tates et al.. 1995). HPRTmutations were measured in three
exposed groups and one unexposed group (seven workers per group). No significant differences in
mutant frequencies were observed between the groups; however, the authors stated that about
50 subjects per group would have been needed to detect a 50% increase.
Major et al. (2001) measured HPRT mutations in female nurses employed in hospitals in Eger
and Budapest, Hungary. This study and an earlier study measuring effects on chromosomes (see
Section 3.3.3.3) were conducted to examine a possible causal relationship between EtO exposure and a
cluster of cancers (mostly breast) in nurses exposed to EtO in the Eger hospital. There was no apparent
increase in cancer among nurses exposed to EtO in the Budapest hospital. Controls were female
hospital workers in the respective cities, and nurses in Eger with known cancers were excluded. Mean
peak levels of EtO were 5 mg/m3 (2.7 ppm) in Budapest and 10 mg/m3 (5.4 ppm) in Eger. HPRT variant
frequencies in both controls and EtO-exposed workers in the Eger hospital were higher than either group
in the Budapest hospital, but there was no significant increase among the EtO-exposed workers in either
hospital when compared with the respective controls. The authors noted that the HPRT variant
frequencies among smoking EtO-exposed nurses in Eger were significantly higher than among smokers
in the Eger controls; however, the fact that the HPRT variant frequency was almost three times higher in
nonsmokers than in smokers in the Eger hospital control group raises questions about the basis of the
claimed EtO effect.
3.3.3.3. Chromosomal Effects
As discussed by Preston (1999) in an extensive review of the cytogenetic effects of EtO, a
variety of cytogenetic assays can be used to measure induced chromosome damage. However, most of
the assays commonly employed measure events that are detectable only in the first (or in some cases the
second) metaphase after exposure and require DNA synthesis to convert DNA damage into a
chromosomal aberration. In addition, DNA repair is operating in peripheral lymphocytes to repair
induced DNA damage. Thus, for acute exposures, the timing of sampling is of great importance. For
chronic studies, the endpoints measure only the most recent exposures, and if the time between last
exposure and sampling is long, any induced DNA damage not converted to a stable genotoxic alteration
is certain to be missed. The events measured include all types of chromosomal aberrations, micronuclei,
SCE, and numerical chromosomal changes (aneuploidy). While SCEs are evidence that chromosomal
damage has been successfully repaired, and are frequently evaluated as indirect biomarkers of
genotoxicity, SCEs themselves are not chromosomal mutations, unlike chromosomal aberrations and
micronuclei, which can both directly result from misrepaired chromosome damage. Consistent with this
distinction, SCEs are not as strongly or consistently associated with carcinogenesis as chromosomal
aberration or micronuclei induction (Zeiger. 2010). Stable chromosomal aberrations include reciprocal
translocations, inversions, and some fraction of insertions and deletions, as well as some numerical
3-44
-------
changes. However, until the development of fluorescent in situ hybridization, chromosome banding
techniques were needed to detect these types of aberrations.
In in vitro assays, EtO has consistently tested positive in studies for multiple types of
chromosomal effects, including DNA strand breaks, SCEs, micronuclei, and chromosomal aberrations
[e.g., see Table 11 of 1 ARC (2008)1. Of note, Adam et al. (2005) measured the sensitivity of different
human cell types to EtO-induced DNA damage using the comet assay, which measures direct strand
breaks and/or DNA damage converted to strand breaks during alkaline treatment. Adam et al. (2005)
reported dose-dependent increases in DNA damage in the concentration range 0-100 |iM in each of the
cell types examined with no notable cytotoxicity. At the lowest concentration reported (20 (jM),
significant increases in DNA damage were observed in lymphoblasts, lymphocytes, and breast epithelial
cells, but not in keratinocytes or cervical epithelial cells, suggesting that breast epithelial cells may have
increased sensitivity to EtO-induced genotoxicity compared to other nonlymphohematopoietic cell
types. In addition, Godderis et al. (2006) investigated the effects of genetic polymorphisms on DNA
damage induced by EtO in peripheral blood lymphocytes of 20 nonsmoking university students. No
significant increases in micronuclei were observed following EtO treatment; however, dose-related
increases in DNA strand breaks were seen in the comet assay. Glutathione S-transferase (GST)
polymorphisms did not have a significant impact on the EtO-induced effects; however, significant
increases in DNA strand breaks were associated with low-activity alleles of two DNA repair enzymes
compared to wild-type alleles.
In vivo, inhalation studies in laboratory animals have demonstrated that EtO exposure levels in
the range of those used in the rodent bioassays induce SCEs in several species, including rats [see
Table 3-7; see also Table 11 of 1 ARC (2008)1. SCEs and micronuclei in mice have not been well
studied, but the available results are generally positive for these effects [see Table 3-7; see also Table 11
of 1 ARC (2008)1. In inhalation studies of chromosomal aberrations in mice, EtO exposure levels in the
range of those used in the rodent bioassays consistently induce chromosomal aberrations, with lower
exposure levels requiring longer durations of exposure (see Table 3-7). In rats, evidence for micronuclei
and chromosomal aberrations from short-term exposures (<4 weeks) to these same exposure levels is
lacking. In particular, studies by van Sittert et al. (2000) and Lorenti Garcia et al. (2001) observed
increases in micronuclei and chromosomal aberrations in splenic lymphocytes of rats exposed to 50,
100, or 200 ppm EtO for 6 hours/day, 5 days/week, for 4 weeks compared to levels from control rats,
but the increases were not statistically significant. IARC (2008) noted, however, that "strong
conclusions cannot be drawn" from these two studies because the cytogenetic analyses "were initiated
5 days after the final day of exposure, a suboptimal time, and the power of the fluorescent in-situ
hybridization studies were limited by analysis of only a single chromosome and the small numbers of
rats per group examined," which was 3 per exposure group in both of the studies, although numerous
cells/rat were examined. In addition, the more recent study of chromosomal aberrations in mice by
3-45
-------
Donner et al. (2010) showed a clear duration effect, with no significant increases at 6 weeks of exposure
to those same EtO concentrations but with statistically significant increases in the 100 and 200 ppm
groups starting at 12 weeks of exposure and a statistically significant increase at the lowest exposure
level tested (25 ppm) at 48 weeks of exposure.
In humans, various studies of occupationally exposed workers have reported SCEs and other
chromosomal effects associated with EtO exposure, including micronuclei and chromosomal
aberrations. The genotoxicity of EtO was demonstrated in humans as early as 1979. Table 3-8
summarizes the cytogenetic effects of EtO on human exposures (see also Sections C.3-C.5 of
Appendix C for more details on some of the studies).
As illustrated in Table 3-8, numerous studies observed increased SCEs in occupationally
exposed workers, especially for workers with the highest exposures [e.g., (Major et al.. 1996; Sarto et
al.. 1991; Tates et al.. 1991; Sarto et al.. 1987)1. Several studies of occupationally exposed workers have
also reported increased micronucleus (MN) formation in lymphocytes (Ribeiro et al.. 1994; Tates et al..
1991). in nasal mucosal cells (Sarto et al.. 1990). and in bone marrow cells (Hogstedt et al.. 1983).
although this endpoint seems to be less sensitive than SCEs and data for workers with the higher
exposures are limited. An association between increased micronucleus frequency and cancer risk has
been reported in at least one large prospective general population study (Bonassi et al.. 2007). In
addition, chromosomal aberrations have been reported in multiple studies of workers occupationally
exposed to EtO (Ribeiro et al.. 1994; Tates et al.. 1991; Sarto et al.. 1987). especially for workers with
the highest exposures. Chromosomal aberrations have been linked to an increased risk of cancer in
several large prospective general population studies [e.g., Boffetta et al. (2007); Rossner et al. (2005);
Hagmar et al. (2004); Liou et al. (1999)1.
3-46
-------
Table 3-7. Ethylene oxide-induced cytogenetic effects in laboratory animals
Study3
Number
exposed
per
exposure
group
(number
of
controls)
Species/strain
Tissue assessed
Exposure
duration
Exposure
concentration (ppm)
or dose (mg/kg)b
Cytogenetic
observations
Comments0
CA
SCE
MN
Intraperitoneal injection
Farooai et al.
(1993)
4(4)
Mice/Swiss
albino
Bone marrow
cells
Single injection;
killed 24 hr after
dosing
0, 30, 60*, 120*, 150*
+
+
+
Significant positive association between
eachendpoint (CAs, SCEs, and MN) and
exposure concentration.
Jenssenand
Ramel (1980)
3(3)
Mice/CBA
Bone marrow
cells
Single injection;
killed 24 hr after
dosing
0, 50, 100, 125*,
150*, 175*
ND
ND
+
No dose-response relationship observed.
Intravenous injeetion
Aooelgren et
al. ("19781
4-8 (10)
Rats/Sprague-
Dawley
Bone marrow
cells
Two injections,
given 30 and 6 hr
before killing
0, 100*, 150, 200 (T)
ND
ND
+
EtO toxic at the highest dose.
Aooelgren et
al. (1978)
4-8(11)
Mice/NMRI
Bone marrow
cells
Two injections,
given 30 and 6 hr
before killing
0, 50, 100*, 150*,
200*, 300 (T)
ND
ND
+
Dose-dependent increase inMN induction
in bone marrow cells; EtO toxic at the
highest dose.
Inhalation exposure
Lvnchet al.
(1984b)
9-12 (12)
Monkeys/
Cynomolgus
PBLs
2 yr; 7 hr/d, 5 d/wk
0, 50*, 100*
+
+
ND
Significant positive associations between
exposure concentration and SCEs,
chromosome-type CAs, and chromatid-
type CAs.
Yaeer and
Benz (1982)
3(3)
Rabbits/New
Zealand white
PBLs
12 wk; 6 hr/d,
5 d/wk
0, 10, 50*, 250*
ND
+
ND
Significant positive association between
SCE level and exposure concentration.
Analyzed 1, 7, 12, 15 wk
postexposure—elevated SCE levels
persisted above base-line at 15 wk
postexposure.
-------
Table 3-7. Ethylene oxide-induced cytogenetic effects in laboratory animals
Study3
Number
exposed
per
exposure
group
(number
of
controls)
Species/strain
Tissue assessed
Exposure
duration
Exposure
concentration (ppm)
or dose (mg/kg)b
Cytogenetic
observations
Comments0
CA
SCE
MN
Yaeer M987>
3(3)
Rabbits/New
Zealand white
PBLs
40 d (6 hr/d), 20 d
(6 hr/d), or 64 d
(0.25 hr twice a
day), 5 d/wk, for
different exposure
groups,
respectively
0, 200*, 400*, 1,500*
ND
+
ND
All treated groups significantly different in
SCEs compared to controls. Cumulative
exposure for all groups is same
(48,000 ppm x hours)—no difference in
SCEs within exposure groups.
Klieerman et
al. M983>
3-4 (4)
Rats/F344
PBLs
1 day; 6 hr/d
0, 50, 150, 450*
+
ND
Dose-dependent trend observed for SCEs.
3-4 (4)
3 day; 6 hr/d
0, 50*, 150*, 450*
+
ND
Significant positive association between
SCE level and exposure concentration
Preston and
Abcriicthv
(1993)
6(6)
Rats/F344
PBLs
1, 2, 3, 4 wk;
6 hr/d, 5 d/wk
0, 150*
+
ND
SCE levels significantly higher than
controls at all time points postexposure.
Duration-dependent (1-4 wk) increase in
SCEs.
van Sittert et
al. (2000)
3(3)
Rats/Lewis
Splenic
lymphocytes
4 wk; 6 hr/d,
5 d/wk
+
Significant positive association between
SCE level and exposure concentration;
analyses were initiated 5 days after the
final day of exposure.
Lorenti
Garcia et al.
(2001)
3(5)
Rats/Lewis
Splenic
lymphocytes
4 wk; 6 hr/d,
5 d/wk
0, 50*, 100*, 200*
+
Significant positive association between
SCE level and exposure concentration,
persisting up to 3 wk postexposure;
analyses were initiated 5 days after the
final day of exposure.
-------
Table 3-7. Ethylene oxide-induced cytogenetic effects in laboratory animals (continued)
Study3
Number
exposed
per
exposure
group
(number
of
controls)
Species/strain
Tissue assessed
Exposure
duration
Exposure
concentration (ppm)
or dose (mg/kg)b
Cytogenetic
observations
Comments0
CA
SCE
MN
One et al.
(1993)
18 (18)
Rats/F344/CR/
BR
Spleen cells
3, 6, or 9 mon6,
2, and 1 hr/d for
the different
exposure groups,
respectively (i.e.,
600 ppm-hr
cumulative
exposure per day),
5 d/wk
0, 100*, 300*, 600*
ND
+
ND
SCE levels significantly higher than
controls for all dose-rate and duration
groups and a trend of duration-dependent
increase in SCE levels.
Bone marrow
cells
0, 100*, 300*, 600*
ND
+
ND
SCE levels significantly higher than
controls for all dose-rates for 6- and
9-month time points and only for the low
dose-rate group at 3-month time point. No
duration-dependent increase in SCEs.
Verenes and
Pritts (1994)
10 (10)
Rats/F344
Bone marrow
cells
4 wk; 6 hr/d,
5 d/wk
0, 200*
ND
ND
+
Ethylene exposures up to 3,000 ppminthe
same study were negative for MN
induction
Donner et al.
(2010)
4-8 (4-7)
Mice/B6C3Fi
PBLs
6 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100, 200
-
ND
ND
Significant positive association between CA
level and exposure concentration, and
significant positive association with
duration after >12 wk at 200 ppm;
reciprocal translocations significantly
increased after >12 wk at >100 ppm.
12 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100*, 200*
+
ND
ND
24 wk; 6 hr/d,
5 d/wk
0, 25, 50*, 100*, 200*
+
ND
ND
48 wk; 6 hr/d,
5 d/wk
0, 25*, 50*, 100*,
200*
+
ND
ND
Ribeiro etal.
(1987)
10 (10)
Mice/Swiss
Webster
Bone marrow
cells
1 day; 6 hr/d
0, 200, 400*, 600*
+
ND
ND
Significant positive association between CA
level and exposure concentration
10 (10)
2 wk; 6 hr/d,
5 d/wk
0, 200*, 400*
+
ND
ND
Significant positive association between CA
level and exposure concentration
Verenes and
Pritts (1994)
10 (10)
Mice/B6C3Fi
Bone marrow
cells
4 wk; 6 hr/d,
5 d/wk
0, 200*
ND
ND
+
Ethylene exposures up to 3,000 ppminthe
same study were negative for MN
induction
-------
Table 3-7. Ethylene oxide-induced cytogenetic effects in laboratory animals (continued)
Study3
Number
exposed
per
exposure
group
(number
of
controls)
Species/strain
Tissue assessed
Exposure
duration
Exposure
concentration (ppm)
or dose (mg/kg)b
Cytogenetic
observations
Comments0
CA
SCE
MN
Donner et al.
(2010)
3-8 (4-8)
Mice/B6C3Fi
Spermatogonia
cells
6 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100, 200
—
ND
ND
Only reciprocal translocations reported;
total aberrations not presented. Positive
trend test result at 12 wk considered
equivocal.
12 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100, 200*
(+)
ND
ND
24 wk; 6 hr/d,
5 d/wk
0, 25, 50, 100*, 200
+
ND
ND
48 wk; 6 hr/d,
5 d/wk
0, 25*, 50*, 100*,
200*
+
ND
ND
Ribeiro etal.
(1987)
10 (10)
Mice/Swiss
Webster
Primary
spermatocytes
1 day; 6 hr/d
0, 200, 400*, 600*
+
ND
ND
Significant positive association between CA
level and exposure concentration
2 wk 6 hr/d,
5 d/wk
0, 200*, 400*
+
ND
ND
Significant positive association between CA
level and exposure concentration
'+' = positive (statistically significant in exposed vs. controls),= nonpositive (statistically no difference between exposed and controls) as reported by authors, typically
via pair-wise comparisons;'(+)' = equivocal or weakly positive result; ND, not determined.
*Doses at which incidence of reported effects were statistically significantly higher (p < 0.05) than the corresponding control groups.
aStudies arranged by intraperitoneal and other routes followed by inhalation, cancer target tissues followed by nontarget tissues, largest to smallest species, lowest dose to
highest dose, and durationby lowest to highest when doses are similar.
bppm for inhalation route and mg/kg for all other routes.
0 Authors' interpretations.
CA = chromosomal aberration; SCE = sister chromatid exchange; MN = micro nucleus; ND = not determined, yr = year; hr = hour; d = day; wk = week; T = toxic;
PBLs = peripheral blood lymphocytes.
-------
Table 3-8. Ethylene oxide-induced cytogenetic effects in humans
Study3
Number exposed
(number of controls)
Exposure duration
(yr)
Ethylene oxide level in
air (ppm)
Cytogenetic
observations
Comments'3 0
Range
Mean
Range
Mean
(TWA)
CA
SCE
MN
Karelova et al.
Sterilization unit: 22(10)
1-8
NR
0-2.6d
NR
+
ND
ND
Significantly higher CAs in exposed vs. control
groups. Significant difference in CAs between
smokers and nonsmokers of control group, but not
exposed group. However, smoking appears to
increase CA levels in exposed group.
(1987)
Factory workers: 21(20)
2-17
NR
0-4.5d
NR
+
ND
ND
Laboratory workers: 25
(20)
1-15
NR
0-4.8d
NR
+
ND
ND
Tates et al.
7(7)
<5
NR
<0.005-0.02
NR
ND
-
-
(1995)
7(7)
>15
NR
<0.05-0.01
NR
ND
-
-
7(7)
Accidental exposure
28-429d
NR
ND
-
-
Clare et al.
(1985)
33 (32)
1-14
NR
0.05-8
0.01e
(+)
ND
ND
Positive correlation between total number of
aberrations and duration of exposure.
Schulte et al.
LEG: 9(1)
NR
NR
0.02-0.02d
0.02
ND
-
-
(1992)
HEG: 12(1)
NR
NR
0.27-1.36d
0.54
ND
-
-
LEG: 32(8)
NR
5.1
0-0.3d
0.04
ND
+
Levels of HEVal adducts and SCEs significantly
positively associated with cumulative EtO
exposure after controlling for smoking exposure.
HEG: 11 (8)
NR
9.5
0.13-0.3d
0.16
ND
+
-
Sarto et al.
Preparation area: 5(10)
0.1^1
2
<1-4.4
0.025f
ND
-
-
Observations in the total exposed group
(sterilization + preparation workers combined)
were not significantly different compared to
referent group.
(1991)
Sterilization area 5(10):
4-12
8.6
NR
0.38f
ND
+
Preparation area: 5(10)
0.1^1
2
<1-4.4
0.025f
ND
ND
-g
Sterilization area: 5(10)
4-12
8.6
NR
0.38f
ND
ND
_g
Sarto et al.
9 (27)
0.5-12
5
0.025-0.38d
NR
ND
ND
-
(1990)
3 (27)
Accidental exposure
NR
>0.38h
ND
ND
+
Van Sittert et al.
19(35)
1-5
NR
<0.05-8
<0.05
(+)
ND
ND
Positive correlation between number of
chromosome breaks and duration of employment.
(1985)
17(35)
6-14
NR
<0.05-8
<0.05
(+)
ND
ND
Hansen et al.
(1984)
14(14)
NR
NR
<0.07-4.3d
NR
ND
-
ND
Smoking history not associated with SCE levels.
-------
Table 3-8. Ethylene oxide-induced cytogenetic effects in humans (continued)
Study3
Number exposed
(number of controls)
Exposure duration
(yr)
Ethylene oxide level in
air (ppm)
Cytogenetic
observations
Comments'3 c
Range
Mean
Range
Mean
(TWA)
CA
SCE
MN
Maver et al.
<19911
34 (23)
NR
8h
<0.1-2.4d
<0.3
+
Levels of HEVal adducts and SCEs significantly
positively associated withEtO exposure and DNA
repair capacity suppressed inEtO-exposed
individuals, independent of smoking history.
Sarto et al.
<1984)
LEG: 22(22)
0.6—4
3
0.2-0.5d
0.35
(+)
+
ND
Weak positive correlation between total CA level
and duration of exposure.
HEG: 19(19)
1.5-15
6.8
3.7-20d
10.7
+
+
ND
Significant positive correlation between SCE level
and EtO exposure concentration.
Stollev et al.
Site I: 13(12)
NR
NR
0.5d
NR
-
-
ND
(1984);
Site II: 22(19)
NR
NR
5-10d
NR
-
+
ND
(1986)
Site III: 25-26(22)
NR
NR
5-20d
NR
+
+
ND
Significant positive correlation between SCE level
and EtO exposure concentration; SCE level weakly
positively associated with CAs.
Pero etal. (19811
LEG (packers): 12 (5)
1-8
4
0.5-1
NR
-
ND
ND
Observations in the total exposed group
(sterilization + packing workers combined) were
significantly higher compared to referent group.
HEG (sterilizers): 5 (5)
0.8-3
1.6
5-10
NR
+
ND
ND
Podd et al.
Smokers: 11
NR
NR
0.5—417k
NR
ND
-
ND
High urinary concentrations of hydroxyethyl
mercapturic acid present in exposed compared to
controls, but not correlated withEtO exposure.
(1994)
Nonsmokers: 14
NR
NR
0.5-208k
NR
ND
-
ND
Schulte et al.
(1995)
Hospital workers (LEG):
28 (8)
NR
5.5
0-0.30
0.08
ND
+
-
MN levels significantly different between LEG
and HEG, but not compared to control subjects.
Hospital workers (HEG):
10 (NR)
NR
10.0
0.13-0.30
0.17
ND
+
-
Tomkins et al.
<1993)
47 (47)
NR
NR
NR
<1
ND
Exposed and control groups were matched for age,
sex, and smoking. Levels of SCE and somatic cell
mutation were not associated withEtO exposure.
However, SCE levels were associated with current
smokers but not with former or never smokers.
Hoestedt et al.
Factory I: 18(11)
0.5-8
3.2
NR
<1
+
-
+
Significant positive correlations between different
cytogenetic endpoints in peripheral blood
lymphocytes and bone marrow cells.
<1983)
Factory I: 18 (10)
0.5-8
3.2
NR
<1
ND
ND
+J
Factory II: 10 (9)
0.5-8
1.7
NR
<1
+
-
ND
-------
Table 3-8. Ethylene oxide-induced cytogenetic effects in humans (continued)
Study3
Number exposed
(number of controls)
Exposure duration
(yr)
Ethylene oxide level in
air (ppm)
Cytogenetic
observations
Comments'3 c
Range
Mean
Range
Mean
(TWA)
CA
SCE
MN
Richmond et al.
LEG: 79(141)
1-10
NR
l-40d
NR
-
+
ND
Significant positive correlation between exposure
duration and CA level in HEG but not LEG.
r19851
HEG: 50(141)
1-10
NR
l-40d
NR
+
+
ND
Sarto et al.
(1987)
10(10)
NR
NR
0-9.3d
1.84
ND
+
ND
Significant positive correlation between SCE level
and EtO exposure concentration.
Ribeiro etal.
75 (22)
3-14
7
2-5d
NR
+
ND
+
(1994)
75 (22)
3-14
7
2-5d
NR
ND
ND
-g
Mai or etal.
LEG: 9(14)
NR
4
2.7-10.9
2.7
+
-
ND
CA levels in HEG significantly higher than LEG.
(1996)
HEG: 27(10)
NR
15
2.7-82
5.5
+
+
ND
Yaeer et al.
LEG: 9(13)
NR
0.5
NR
131
ND
-
ND
(1983)
HEG: 5(13)
NR
0.5
NR
5011
ND
+
ND
Tates et al.
(1991)
Factory workers: 15(15)
3-27
12
17-33
NR
+
+
+
Factory workers showed 17 times higher SCE
frequencies compared with hospital workers.
Hospital workers: 9(8)
2-6
4
20-25
NR
+
+
-
Laurent et al.
Smokers: 15 (7)
0.5-10
4.5
20-123
NR
ND
+
ND
Higher levels of SCEs reported in workers with
greater duration of exposure. Effects from
smoking were not found to be additive in EtO
exposed smoking workers compared with
nonsmokers.
(1984)
Nonsmokers: 10 (15)
0.5-10
5.7
20-123
NR
ND
+
ND
Garry et al.
(1979)
12(12)
NR
0.42
NR
36d
ND
+
ND
SCE levels remained elevated 8 wk postexposure.
-------
Table 3-8. Ethylene oxide-induced cytogenetic effects in humans (continued)
Study3
Number exposed
(number of controls)
Exposure duration
(yr)
Ethylene oxide level in
air (ppm)
Cytogenetic
observations
Comments'3 c
Range
Mean
Range
Mean
(TWA)
CA
SCE
MN
Lerda and Rizzi
10(10)
NR
3
60-69d
NR
+
+
ND
(1992)
'+' = positive (statistically significant in exposed vs. controls),= nonpositive (statistically no difference between exposed and controls) as reported by authors, typically
via pairwise comparisons;'(+)' = some indication of a response associated with exposure, as reported by authors, typically statistically significant by one analytical
methods but not others (i.e., correlation or trend test vs. pair-wise comparison), or associated with duration vs. concentration.
aStudies are arranged in order of increasing mean exposure concentration, when available, or by the lower end of the range when a mean is not available.
bAuthors' interpretations.
°A11 studies analyzed inPBLs except where indicated.
dTWA (8-hr).
"Geometric mean.
fTWA (6.5 hr).
gBuccal epithelial cells.
hExposed acutely from sterilizer leakage in addition to chronic exposure.
'Maximum years exposed.
JErythroblast cells and polychromatic erythrocytes.
kPeak concentrations.
'Average 6-mo nth cumulative exposure (mg).
yr = year; CA = chromosomal aberration; SCE = sister chromatid exchange; MN = micro nucleus; HEVal adduct = hydroxyethylvaline adduct; ND = not determined;
NR = not reported; PBLs = peripheral blood lymphocytes; LEG = low-exposure group; HEG = high exposure group; wk = week.
-------
3.3.3.4. Summary
As presented above (see Sections 3.3.3.1 and 3.3.3.2) and summarized in Table 3-9, inhalation
exposure to >3 ppm EtO for <4 weeks generally increased N7-HEG DNA adduct levels in all tissues
examined in a tissue-specific concentration-dependent manner in both mice and rats, with no apparent
difference in sensitivity between cancer target and nontarget tissues in either rodent species. No studies
measured DNA adducts levels following >4 weeks of EtO exposure. Tissue N7-HEG DNA adduct
levels in control or exposed mice were 2-3 times lower than levels reported in analogous tissues from
similarly exposed rats, and adduct levels in unexposed rats were approximately 10 times lower than
levels in the peripheral blood cells of unexposed humans. In rats, <3 daily i.p. injections of doses as low
as 0.0005-0.001 mg/kg (comparable to <0.1 ppm inhalation exposure) increased the incidence of
exogenous N7-DNA adducts in a tissue-specific concentration-dependent manner. Administration of
0.1 mg/kg (comparable to ~1 ppm inhalation exposure) induced rat tissue levels of exogenous N7-HEG
adducts comparable to the amount of endogenous adducts in concurrent controls, while >10 mg/kg
induced a concentration-dependent increase in Hprt mutations in rat lymphocytes. Mutation frequency
in the reporter genes Hprt and LacI was increased in a concentration-dependent manner primarily in
lymphocytes from rats and mice exposed to concentrations associated with significant tumor induction
in cancer bioassays (i.e., >50 ppm) for up to 48 weeks, with dominant lethal mutations and heritable
translocations reported in germ cells from male mice following exposure to >165 ppm. In mice exposed
to >50 ppm for 102 weeks, the frequency of novel Trp53 and Hras mutations increased in mammary
tumors, with greater increases reported in novel Kras mutations in lung, Harderian gland, and uterine
tumors. Furthermore, the vast majority of the mutations in tumors from EtO-exposed mice occurred at
purine nucleotides (i.e., adenine and guanine), which is consistent with the formation of EtO DNA
adducts on guanine (N7-HEG and 06-HEG) and adenine (N3-HEA) bases in vivo. Human data on DNA
adducts and mutations are more limited. In humans, N7-HEG adducts were not significantly increased
in peripheral blood samples from two occupational exposure studies, while HPRT mutations were only
observed in peripheral blood lymphocytes from a cohort of factory workers after 12 years of exposure to
17-33 ppm, and not in other workers after a shorter duration of exposure to similar concentrations, or a
longer duration of exposure to much lower concentrations.
As presented above (see Section 3.3.3.3) and summarized in Table 3-10, cytogenetic effects
(chromosomal aberrations, micronuclei and/or SCEs) were increased in a concentration-dependent
manner after 3-730 days of exposure to >50 ppm EtO, frequently the lowest concentration evaluated, in
various mammalian species, including nonhuman primates, rabbits, rats, and mice. In studies evaluating
multiple time points, cytogenetic effects were observed at lower concentrations following longer
exposure durations in both mice and rats. Neither chromosomal aberrations nor micronuclei were
induced in rats exposed to <450 ppm for <3 days or to <200 ppm for <4 weeks (there were no studies of
these endpoints in rats with durations longer than 4 weeks), while positive associations were consistently
3-55
-------
reported between exposure concentration and increased lymphocyte SCE levels. In mice, chromosomal
aberration levels exposed to <600 ppm for <48 weeks were positively associated with exposure
concentration, and significant responses were observed at lower exposure concentrations (>25 ppm)
with longer exposure durations (48 weeks). SCE and micronucleus induction in mice likely follow a
similar pattern, but insufficient exposure-response data were available for any definitive interpretation.
In humans, few positive associations were reported in cohorts exposed to <0.04 ppm (mean
time-weighted average), while predominantly positive associations with increased SCE formation and
exposure concentration, or chromosomal aberrations and exposure duration, were reported in cohorts
exposed to 0.04-0.4 ppm. Exposures >1 ppm were positively associated with increased incidence of
micronuclei formation, as well as SCEs and chromosomal aberrations. Overall, in the majority of
cohorts, the induction of chromosomal aberrations and SCEs appeared to be positively associated with
both exposure concentration and duration; this association is not as clear with micronuclei formation,
which was rarely evaluated in the more highly exposed populations.
In conclusion, the available data from in vitro studies, laboratory animal models, and
epidemiological studies establish that EtO is a mutagenic and genotoxic agent that causes various types
of genetic damage in a manner positively associated with both exposure concentration and duration,
including chromosome mutations (chromosomal aberrations, micronuclei) as well as genetic mutations
in proto-oncogenes (Kras, Hras) and the tumor suppressor Trp53. To the extent that they have been
evaluated, similar outcomes have been reported in both exposed humans and laboratory animals. In
rodents, genotoxicity is induced in germ cells, as well as in cancer target and nontarget tissues,
frequently at the lowest concentrations evaluated. While N7-HEG DNA adducts appear to be the most
sensitive outcome following EtO exposure, this may be an artifact of experimental design, as few rodent
studies evaluated clastogenic endpoints following exposures to <50 ppm, and none evaluated exposures
<25 ppm. Together, the concentration- and duration-dependent positive associations consistently
observed in available laboratory animal and human studies strongly support a causal relationship
between EtO exposure and genotoxicity or mutagenesis in numerous tissues.
3-56
-------
Table 3-9. Summary of exposure and duration patterns for ethylene oxide -induced DNA adducts and mutations in
humans and laboratory animals (inhalation studies)
DNA adducts3
Mutations'3
Posi(i\c ix'spoiiscs in humans with occupational exposuix'''
Exposure (ppm)
lime ^yr;
Total positive
responses (%)d
Exposure (ppm)
lime ^yr;
Total positive
responses (%)d
0-5
>5-10
>10
0-5
>5-10
>10
0-5
ND
0/2
ND
0/2 (0)
0-5
0/2
ND
0/1
0/3 (0)
>5-25
ND
ND
ND
ND
>5-25
0/1
ND
1/2
1/3 (33)
>25
ND
ND
ND
ND
>25
ND
ND
ND
ND
I'osilixo responses in nils''
Exposure (ppm)
Time (wk)
Total positive
responses (%)d
Exposure (ppm)
Time (wk)
Total positive
responses (%)d
0-0.15
>0.15-0.5
>0.5-4
0-12
>12-48
>48-102
0-1
ND
ND
ND
ND
0-50
1/2
ND
ND
1/2 (50)
>1-50
ND
ND
19/19
19/19 (100)
>50-200
4/5
ND
ND
4/5 (80)
>50-500
14/14
14/14
41/49
69/77 (90)
>200-1,200
1/1
ND
ND
1/1 (100)
Ptisili\c ix'sponscs in mice*'
Exposure (ppm)
Time (wk)
Total positive
responses (%)d
Exposure (ppm)
Time (wk)
Total positive
responses (%)d
0-0.15
>0.15-0.5
>0.5-4
0-12
>12-48
>48-102
0-1
ND
ND
ND
ND
0-50
1/10
2/8
4/4e
7/22 (32)
>1-50
ND
ND
18/21
18/21 (86)
>50-200
8/19
5/10
4/4e
17/33 (52)
>50-100
ND
ND
7/7
7/7 (100)
>200-1,200
11/11
ND
ND
11/11 (100)
aIncludes primarily N7-hydroxyethylguanine (N7-HEG) adducts; other adducts (e.g., Os-hydroxyethylaguanine, 3-hydroxyethyladenine) were rarely evaluated, and when
reported, were induced only in conditions also significantly increasing N7-HEG levels.
bIncludes mutations at Hprt (rodents), HPRT (humans), Lac I, p53, dominant lethal mutations and reciprocal translocations in various tissues.
cThe number of individual exposure groups with positive, exposure-associated results for DNA adducts or mutation endpoints, for each tissue evaluated, are expressed as
a fraction of the total number of exposed groups evaluated (i.e., number of exposure groups with a positive response/total number of groups exposed) within each
concentration (row) and duration (column) category, by species.
dThe sum of the row combining groups across each exposure duration range category, by exposure concentration group, for eachendpoint category, expressed as a
fraction (and percentage).
eTumor tissue evaluated for the presence of mutations in oncogenes IIras or Kras, and/or the tumor suppressor gene p53.
yr = year; ND = not determined; wk = week.
-------
Table 3-10. Summary of exposure and duration patterns for ethylene oxide-induced cytogenetic effects in humans and
laboratory animals (inhalation studies)
Chromosomal aberrations (CA)
Sister chromatid exchanges (SCE)
Micronuclei (MN)
Posili\e responses in humans willi occupational exposure1'
Exposure
(ppm)
Time (yr)
Total positive
responses (%)b
Time (yr)
Total positive
responses (%)b
Time (yr)
Total positive
responses (%)b
0-5
>5-10
>10
0-5
>5-10
>10
0-5
>5-10
>10
0-5
8/13
1/3
ND
9/16 (56)
3/8
6/6
0/1
9/15 (60)
2/6
1/9
0/1
3/16(19)
>5-25
1/1
1/1
2/2
4/4 (100)
2/3
2/2
2/2
6/7 (86)
0/1
ND
1/2
1/3 (33)
>25
1/1
ND
ND
1/1 (100)
3/4
ND
ND
3/4 (75)
0/1
ND
ND
0/1 (0)
Posili\e responses in rals'1
Exposure
(ppm)
Time (wk)
Total positive
responses (%)b
Time (wk)
Total positive
responses (%)b
Time (wk)
Total positive
responses (%)b
0-4
>4-24
>24-48
0-4
>4-24
>24-48
0-4
>4-24
>24-48
0-50
0/4
ND
ND
0/4 (0)
3/4
ND
ND
3/4 (75)
0/2
ND
ND
0/2 (0)
>50-150
0/8
ND
ND
0/8 (0)
7/8
4/4
2/2
13/14 (93)
0/2
ND
ND
0/2 (0)
>150-450
0/4
ND
ND
0/4 (0)
4/4
8/8
4/4
16/16 (100)
1/3
ND
ND
1/3 (33)
Posiliu* responses in mice"
Exposure
(ppm)
Time (wk)
Total positive
responses (%)b
Time (wk)
Total positive
responses (%)b
Time (wk)
Total positive
responses (%)b
0-4
>4-24
>24-48
0-4
>4-24
>24-48
0-4
>4-24
>24-48
0-100
ND
4/18
6/6
10/24 (42)
ND
ND
ND
ND
ND
ND
ND
ND
>100-200
1/1
2/6
2/2
6/12 (50)
ND
ND
ND
ND
1/1
ND
ND
1/1 (100)
>200-600
6/6
ND
ND
6/6 (100)
ND
ND
ND
ND
ND
ND
ND
ND
aThe number of individual exposure groups with positive, exposure-associated results for chromosomal aberrations, sister chromatid exchanges, or micronuclei endpoints,
for each tissue evaluated, are expressed as a fraction of the total number of exposed groups evaluated (i.e., number of exposure groups with a positive response/total
number of groups exposed) within each concentration (row) and duration (column) category, by species. Human studies that did not report exposure duration were
omitted from the counts.
bThe sum of the row combining groups across each exposure duration range category, by exposure concentration group, for eachendpoint category, expressed as a
fraction (and percentage). Endpoints for which results were equivocal or weakly positive ['(+)'] were considered negative for the counts,
yr = year; wk = week; ND = not determined.
-------
3.4. MODE OF ACTION
EtO is an alkylating agent that has consistently been found to produce numerous genotoxic
effects in a variety of biological systems ranging from bacteriophages to occupationally exposed
humans. It is carcinogenic in mice and rats, inducing tumors of the lymphohematopoietic system, brain,
lung, connective tissues, uterus, and mammary gland. In addition, epidemiological studies have shown
an increased risk of various types of human cancers (see Table A-5 in Appendix A), in particular
lymphohematopoietic and breast cancers. Target tissues for EtO carcinogenicity in laboratory animals
are varied, and the cancers are not clearly attributable to any specific type of genetic alteration, which
appear following similar exposure durations and concentrations in both cancer target and nontarget
tissues alike. Although the precise mechanisms by which the multisite carcinogenicity in mice, rats, and
humans occurs are unknown, EtO is clearly a mutagenic and genotoxic agent, as discussed in
Section 3.3.3, and mutagenicity and genotoxicity are well established as playing a key role in
carcinogenicity. Section 3.4.1 discusses possible mechanisms by which a mutagenic mode of action
might be instrumental in EtO carcinogenesis, Section 3.4.2 briefly summarizes the limited evidence for
alternative or additional modes of action, and Section 3.4.3 presents an analysis of the evidence for a
mutagenic mode of action for EtO carcinogenicity under the EPA's mode-of-action framework [(U.S.
EPA. 2005a). Section 2.4.3],
3.4.1. Possible Mechanisms for Mutagenic Mode of Action
3.4.1.1. General Mechanisms
Exposure of cells to DNA-reactive agents results in the formation of carcinogen-DNA adducts.
The formation of DNA adducts results from a sequence of events involving absorption of the agent,
distribution to different tissues, and accessibility of the molecular target (Swenberget al., 1990).
Alkylating agents may induce several different DNA alkylation products (Beranek. 1990) with varying
proportions, depending primarily on the electrophilic properties of the agent. The predominant DNA
adduct formed by EtO is N7-HEG, although other adducts, such as N3-HEA and 06-HEG, have also
been observed, in much lesser amounts (Walker et al., 1992a; Segerback, 1990). In addition to direct
DNA adduct formation via alkylation, Marsden et al. (2009) observed an indirect effect of EtO exposure
on endogenous N7-HEG adduct formation at the highest doses tested and hypothesized that exogenous
EtO could also indirectly increase endogenous EtO-DNA adduct formation via oxidative stress (see also
Section 3.3.3.1 and Appendix C). An alternative possibility is that the responsible DNA adduct removal
and repair mechanisms were overwhelmed, or inhibited, although the effect of EtO exposure on DNA
repair pathway activity has not been evaluated.
The various adducts are processed by different repair pathways, and the subsequent genotoxic
responses elicited by unrepaired DNA adducts are dependent on a wide range of variables. While the
specific adduct(s) responsible for EtO-induced mutagenesis and genotoxicity and the mechanism(s) by
3-59
-------
which this adduct(s) induces heritable genotoxic damage are unknown, the similar induction of N7-HEG
adducts in both cancer target and nontarget tissues in vivo suggests that the formation of DNA adducts
may not be the limiting factor in regulating EtO-induced carcinogenesis (see Sections 3.3.3.1 and
3.3.3.4).
It had been postulated that the predominant EtO-DNA adduct, N7-HEG, although unlikely to be
directly promutagenic, could be subject to depurination, resulting in an apurinic site which could be
vulnerable to miscoding during cell replication [e.g., Walker and Skopek (1993); see also
Section 3.3.3.1], However, in a study designed to test this hypothesis, Rusvn et al. (2005) failed to
detect an accumulation of abasic sites in brain, spleen, and liver tissues of rats exposed to EtO. Rusvn et
al. (2005) conclude that the accumulation of abasic sites is unlikely to be a primary mechanism for EtO
mutagenicity, although they note that it is also possible that their assay was not sufficiently sensitive to
detect small increases in abasic sites or that abasic sites are only mutagenic under conditions of rapid
cell turnover, when cell replication may occur before repair of the abasic site (the tissues examined in
their study were relatively quiescent). Another potential mechanism for EtO-induced mutagenicity is
the direct mutagenicity of the promutagenic adducts such as 06-HEG, although these adducts are
generally considered to occur at levels too low to explain all of the observed mutagenicity (1ARC.
2008). In an in vitro study, Tompkins et al. (2009) exposed plasmid DNA to a range of EtO
concentrations in water and reported that only the N7-HEG adduct was detectable after exposure to EtO
concentrations up to 2,000 |iM; at higher EtO concentrations (>10 mM), N1-hydroxyethyladenine and
06-HEG adducts were also quantifiable but at much lower levels than the N7-HEG adducts. Tompkins
et al. (2009) then examined the mutagenicity of these adducts in a supF forward mutation assay and
reported that the relative mutation frequencies were statistically significantly elevated only for plasmids
exposed to these higher EtO concentrations. Note, however, that increases in relative mutation
frequency were observed for N7-HEG adduct levels corresponding to lower EtO concentrations, and
biologically relevant EtO-related increases in mutation frequency at these lower concentrations cannot
be ruled out given the variability of the data and the limitations of the study (see Appendix C,
Sections C.l.1.2 and C.2.2, for a more detailed discussion of this study). An additional mechanism
suggested for EtO-induced mutagenicity is the imidazole ring-opening of N7-HEG, which can result in
stable, potentially mutagenic lesions; however, EtO-induced N7-HEG ring-opening has not been
corroborated in vivo (1ARC. 2008).
The events involved in the formation of chromosomal damage by EtO are similarly unknown.
N-alklylated bases are removed from DNA by base excision repair pathways. A review by Memisoglu
and Samson (2000) notes that the action of DNA glycosylase and apurinic endonuclease creates a DNA
single-strand break, which can in turn lead to DNA double-strand breaks (DSBs). DSBs can also be
produced by normal cellular functions, such as during V(D)J recombination in the development of
lymphoid cells or topoisomerase Il-mediated cleavage at defined sites, or they can result from
3-60
-------
interference in normal cellular functions, such as the inhibition of DNA replication, or chromosome
missegregation during mitosis. Furthermore, genotoxic damage in cells undergoing mitosis may induce
chromosome segregation errors via selective stabilization of kinetochore-microtubule assemblies by
DNA damage response proteins, providing one possible link between gene- and chromosome-level
damage (Bakhoum et al.. 2014). Reviews of mechanisms of DSB repair indicate that the molecular
mechanisms are not fully understood (Lieber. 2010; Pfeiffer et al.. 2000). These reviews provide a
thorough discussion of both sources (endogenous and exogenous) of DSBs and the variety of repair
pathways that have evolved to process the breaks. Although homology-directed repair generally restores
the original sequence, during nonhomologous end-joining, the ends of the breaks are frequently
modified by addition or deletion of nucleotides. The lack of accumulation of abasic sites observed in the
Rusvn et al. (2005) study discussed above argues against a mechanism involving abasic sites as hot
spots for strand breaks, although it is possible that abasic sites accumulate more readily in replicating
lymphocytes, which were not examined by Rusvn et al. (2005). Another postulated mechanism for
EtO-induced strand breaks is via the formation of hydroxy ethyl adducts on the phosphate backbone of
the DNA, but this mechanism requires further study (1ARC. 2008).
3.4.1.2. Mechanisms Specific to Lymphohematopoietic Cancers
Lymphohematopoietic malignancies, like all other cancers, are considered to be a consequence
of an accumulation of genetic and epigenetic changes involving multiple genes and chromosomal
alterations. Although it is clear that chromosome translocations are common features of some
hematopoietic cancers, there is evidence that mutations in TP 5 3 or NRAS are involved in certain types of
leukemia (U.S. EPA. 1997). It should also be noted that therapy-related leukemias exhibiting reciprocal
translocations are generally seen only in patients who have previously been treated with
chemotherapeutic agents that act as topoisomerase II inhibitors (U.S. EPA. 1997). In NHL, the BCL6
gene is frequently activated by translocations (Chaganti et al., 1998) as well as by mutations within the
gene-coding sequence (Lossos and Levy. 2000). Preudhotntne et al. (2000) observed point mutations in
the AML1 gene in 9 of 22 patients with the M0 type (minimally differentiated acute myeloblasts
leukemia) of acute myeloid leukemia (AML), and Harada et al. (2003) identified AM LI point mutations
in cases of radiation-associated and therapy-related myelodysplastic syndrome (MDS)/AML. In both
reports, point mutations within the coding sequence were found in patients with normal karyotypes as
well as some with translocations or other chromosomal abnormalities. Zharlyganova et al. (2008)
identified AML1 mutations in 7 of 18 radiation-exposed MDS/AML patients but in none of
13 unexposed MDS/AML cases. Other point mutations have also been identified in therapy-related
MDS/AML patients, including TP53 gene mutations after exposure to alkylating agents (Christiansen et
al.. 2001) and mutations in RAS family members and other genes in the receptor tyrosine kinase signal
transduction pathway (Christiansen et al.. 2005). Several models have been developed to integrate these
3-61
-------
various types of genetic alterations. One recent model suggests that the pathogenesis of MDS/AML can
be subdivided into at least eight genetic pathways that have different etiologies and different biologic
characteristics (Pedersen-Biergaard et al.. 2006).
A mode-of-action-motivated modeling approach based solely on chromosome translocations has
been proposed by Kirman et al. (2004). The authors suggest a nonlinear dose-response relationship for
EtO and leukemia, postulating that "chromosomal aberrations are the characteristic initiating events in
chemically induced acute leukemia and gene mutations are not characteristic initiating events." They
propose that EtO must be responsible for two nearly simultaneous DNA adducts, yielding a
dose-squared (quadratic) relationship between EtO exposure and leukemia risk. However, as discussed
above, there is evidence that does not support the assumption that chromosomal aberrations represent
the sole initiating event. In fact, these aberrations or translocations could be a downstream event
resulting from genomic instability, which itself could result from genetic damage in cells undergoing
mitosis (see Section 3.4.1.1.). In addition, it is not clear that acute leukemia is the lymphohematopoietic
cancer subtype associated with EtO exposure; in the large NIOSH study, increases in
lymphohematopoietic cancer risk were driven by increases in lymphoid cancer subtypes. Furthermore,
even if two reactions with DNA resulting in chromosomal aberrations or translocations are
early-occurring events in some EtO-induced lymphohematopoietic cancers, it is not necessary that both
events be associated with EtO exposure (e.g., background error repair rates or exposure to other
alkylating agents may be the cause). Moreover, EtO could also produce translocations indirectly by
forming DNA or protein adducts that affect the normally occurring recombination activities of
lymphocytes or the repair of spontaneous DSBs. Evidence suggests that human leukocytes are more
sensitive to increased DNA fragmentation than are human epithelial cells following EtO exposure in
vitro (Adam et al., 2005), and increased levels of genotoxicity have been noted in human leukocytes
bearing polymorphisms in the XRCC3 DSB-repair pathway component [Godderis et al. (2006); see
Section C.6 in Appendix C], Thus, broader mode-of-action considerations are not regarded as
supportive of the hypothesis that the exposure-response relationship is purely quadratic.
3.4.1.3. Mechanisms Specific to Breast Cancer
Breast cancer is similarly considered to be a consequence of an accumulation of genetic and
epigenetic changes involving multiple genes and chromosomal alterations (Ingyarsson, 1999). Again,
the precise mechanisms by which EtO induces breast cancer are unknown. As discussed in
Section 3.3.3.2, Houle et al. (2006) noted that the mammary gland carcinomas in EtO-exposed mice
exhibited a distinct shift in the mutational spectra of the Trp53 and Hras genes, compared to
spontaneous tumors, and more commonly displayed concurrent mutations of the two genes. The
mutational spectra reported by Houle et al. (2006) in both Trp53 and Hras indicate that purine bases
(i.e., guanine and adenine) were the predominant targets for mutations in tumors of EtO-exposed mice,
3-62
-------
while the majority of mutations in spontaneous tumors involved pyrimidine bases (primarily cytosine).
While HRAS mutations in human breast cancer are rare (0.4% samples in COSMIC),11 and tumor
mutations in RAS genes are more frequently reported at codon 12 than 61 [codon 12 mutations were not
evaluated by Houle et al. (2006)1. half of the Hras mutations in mammary gland tumors in the
EtO-exposed mice resulted from an A —~ T transversion, which was the only codon 61 HRAS mutation
reported in human breast cancer. While murine Trp53 and human TP 53 genes are highly homologous,
they exhibit sequence differences in post-translational modification and intragenic suppressor sites, as
well as specific critical codons, which prevent facile comparison of site-specific mutational events
between species. In human breast cancers, TP53 is the second most frequently mutated gene (23% of
samples in COSMIC); at the exon level, the majority of both TP 5 3 mutations in human breast cancers
and Trp53 mutations in mammary gland tumors in EtO-exposed mice are similarly distributed across
exons 5-8 in the DNA-binding domain, suggesting that EtO-induced tumors bear Trp53 mutations
affecting p53 function in a manner similar to the TP53 mutations reported in human breast cancers.
As noted above, polymorphisms in a XRCC family gene associated with DSB repair were
associated with increased DNA fragmentation in human leukocytes (see Section C.6 in Appendix C),
and other DSB repair genes (e.g., BRCA1, BRCA2, XRCC1) are known to regulate breast cancer
susceptibility (Shi et al.. 2004; Hu et al.. 2002). but the role of any of these pathways in mediating
EtO-induced murine mammary gland or human breast cancers remains unknown. In addition, the comet
assay results of Adam et al. (2005) suggest that human breast epithelial cells may have increased
sensitivity to EtO-induced genotoxicity compared to cervical and epidermal epithelial cells (see
Section C.6 in Appendix C); however, the basis for any increased sensitivity of breast epithelial cells is
similarly unknown.
3.4.1.4. Summary on Mutagenic Mode of Action
In summary, EtO induces a variety of types of genetic damage. It directly interacts with DNA,
causing concentration- and duration-dependent increases in DNA adducts, genetic mutations, including
mutations in proto-oncogenes and tumor suppressor genes, and chromosome damage in various rodent
tissues and human peripheral blood cells. EtO-induced genotoxicity is observed after shorter exposure
durations and at lower exposure concentrations than those associated with tumor induction in both
rodents and occupationally exposed humans (see Section 3.3.3.4). Depending on a number of variables,
EtO-induced DNA adducts may (1) be repaired, (2) result in a base-pair mutation during replication, or
(3) be converted to a DSB, which also may be repaired or result in unstable (micronuclei) or stable
(translocation) cytogenetic damage. The available data are strongly supportive of a mutagenic mode of
"Catalogue of Somatic Mutations in Cancer (COSMIC), accessed 08 June, 2016 Chttp://cancer.saneer.ac.uk/was/ browse/:
tissue = breast, all subtissue and histology selections set to "include all"). Out of 3,087 breast cancer tissue samples,
13 contained mutations at HRAS; out of 12,318 breast cancer tissue samples, 2,877 had mutations in TP53.
3-63
-------
action involving gene mutations and chromosomal aberrations (translocations, deletions, or inversions)
that critically alter the function of oncogenes or tumor suppressor genes. Although it is clear that
chromosome translocations are common features of many hematopoietic cancers, there is evidence that
mutations in TP53, AML1, or NRAS are also involved in some leukemias. The above scientific evidence
along with the summarized genotoxicity evidence in Section 3.3.3.4 provide support for a mutagenic
mode of action.
3.4.2. Evidence and Possible Mechanisms for Alternative Modes of Action
There are no compelling alternative or additional hypothesized modes of action for EtO
carcinogenicity. For example, there is no cytotoxicity or other toxicity indicative of regenerative
proliferation or some other toxicity-related mode of action. Oxidative stress has been hypothesized as a
mode of action, but there is little evidentiary support for this hypothesis and the role of oxidative stress
in EtO-induced carcinogenicity is speculative at this time (see Section 3.3.2 and Sections J.3.2 and J.4.1
of Appendix J, as well as the response to Comment 6 in Appendix K).
3.4.3. Analysis of the Mode of Action for Ethylene Oxide Carcinogenicity under the EPA's
Mode-of-Action Framework
In this section, the evidence for a mutagenic mode of action for EtO carcinogenicity is analyzed
under the mode of action framework in the EPA's 2005 Guidelines for Carcinogen Risk Assessment
rU.S. EPA (2005a). Section 2.4.3],
The hypothesis is that EtO carcinogenicity has a mutagenic mode of action. This hypothesized
mode of action is presumed to apply to all of the tumor types.
The key events in the hypothesized mutagenic mode of action are (1) DNA adduct formation by
EtO, which is a direct-acting alkylating agent; (2) the resulting heritable genetic damage, including DNA
mutations, particularly in oncogenes and tumor suppressor genes, as well as chromosomal alterations;
and (3) clonal expansion of mutated cells during later stages of cancer development; eventually resulting
in (4) tumor formation. Mutagenicity is a well-established cause of carcinogenicity.
1. Is the hypothesized mode of action sufficiently supported in the test animals?
Consistent with the EPA's 2005 Guidelines for Carcinogen Risk Assessment [U.S. EPA (2005a).
Section 2.4.3], this mode-of-action analysis for a mutagenic mode of action is organized around the Hill
"criteria" (or considerations) developed for the analysis of epidemiological studies (Hill. 1965). These
considerations are denoted in italics in the discussion below.
Numerous studies have demonstrated that EtO forms protein and DNA adducts, in mice and rats
(see Sections 3.3.2, 3.3.3.1, and 3.3.3.4), and there is incontrovertible evidence that EtO is mutagenic
and genotoxic (see Sections 3.3.3.2, 3.3.3.3, and 3.3.3.4). The evidence for causal associations between
3-64
-------
the key events and tumor formation has strength and consistency. Increases in the frequency of gene
mutations in reporter genes have been observed in the lung, T-lymphocytes, bone marrow, and testes of
transgenic mice and in T-lymphocytes of rats exposed to EtO via inhalation at concentrations similar to
those inducing tumors in the rodent carcinogenesis bioassays. In addition, in the lung, uterine,
mammary gland and Harderian gland tumors from EtO-exposed mice in those bioassays, dramatic shifts
toward guanine and adenine mutations have been observed in the mutational spectra of the
proto-oncogenes Hras and Kras, as well as the tumor suppressor Trp53, consistent with the propensity
of EtO to form DNA adducts on purine bases (see Section 3.3.3.2).
Inhalation studies in laboratory animals have also demonstrated that EtO exposure levels in the
range of those used in the rodent bioassays induce SCEs in several species and consistently induce
chromosome aberrations in mice (see Sections 3.3.3.3 and 3.3.3.4). No inhalation studies of SCEs or
micronuclei are available for mice. In rats, although SCEs are consistently observed in the available
studies, the results for micronuclei formation and chromosomal aberrations following subchronic (up to
4-week) inhalation exposures to the same exposure levels as those used in the rodent bioassays have
been nonpositive (see Sections 3.3.3.3 and 3.3.3.4); however, I ARC (2008) has noted analytical
limitations with some of these analyses (see Section 3.3.3.3). In addition, Donner et al. (2010)
demonstrated a clear duration effect in mice, with chromosomal aberrations being induced at those same
exposure levels only following longer exposure durations (>12 weeks).
Specificity is not expected for a multisite mutagen and carcinogen such as EtO (U.S. EPA.
2005a).
A temporal relationship is clearly evident, with DNA adducts, point mutations, and
chromosomal effects observed in acute and subchronic assays (see Sections 3.3.3.1 - 3.3.3.3).
Dose-response relationships have been observed between EtO exposure in vivo and DNA
adducts, SCEs, and Hprt and Trp53 mutations (see Section 3.3.3). A mutagenic mode of action for EtO
carcinogenicity also clearly comports with notions of biological plausibility and coherence because EtO
is a direct-acting alkylating agent. Such agents are generally capable of forming DNA adducts, which in
turn have the potential to cause genetic damage, including mutations; and mutagenicity, in its turn, is a
well-established cause of carcinogenicity. This chain of key events is consistent with current
understanding of the biology of cancer.
In addition to the clear evidence supporting a mutagenic mode of action in test animals, there are
no other compelling hypothesized modes of action for EtO carcinogenicity. For example, there is no
evidence of cytotoxicity or other cellular dysfunction indicative of regenerative proliferation, and
little-to-no evidence supporting some other toxicity-related mode of action, such as oxidative stress (see
Section 3.4.2).
3-65
-------
2. Is the hypothesized mode of action relevant to humans?
The evidence discussed above demonstrates that EtO is a systemic mutagen in test animals; thus,
there is the presumption that it would also be a mutagen in humans. Moreover, human evidence directly
supports a mutagenic mode of action for EtO carcinogenicity. Several studies of humans have reported
exposure-response relationships between hemoglobin adduct levels and EtO exposure levels [e.g., van
Sittert et al. (1993); Schulte et al. (1992); see Section 3.3.2], demonstrating the ability of EtO to bind
covalently in systemic human cells, as it does in rodent cells. DNA adducts in EtO-exposed humans
have not been well studied, and the evidence of increased DNA adducts is limited (see Sections 3.3.3.1
and 3.3.3.4). EtO has yielded positive results in in vitro mutagenicity studies of human cells (see
Figure 3-3). Although the studies of point mutations in EtO-exposed humans are few and insensitive
and the evidence for mutations is limited, there is clear evidence from a number of human studies that
EtO causes chromosomal aberrations, SCEs, and micronucleus formation in peripheral blood
lymphocytes, with some evidence of positive relationships with exposure concentration and duration
(see Section 3.3.3.3 and Table 3-8).
Finally, there is strong evidence that EtO causes cancer in humans, including cancer types
observed in rodent studies (i.e., lymphohematopoietic cancers and breast cancer), providing further
weight to the relevance of the aforementioned genotoxic effects to the development of cancer in humans
(see Sections 3.1 and 3.5.1).
In conclusion, the weight of evidence supports a mutagenic mode of action for EtO
carcinogenicity. Although oxidative stress or other processes might contribute to the development of
EtO-induced cancers, the available evidence best supports a mutagenic mode of action as the primary
process in EtO carcinogenicity.
3. Which populations or life stages can be particularly susceptible to the hypothesized mode of action?
The mutagenic mode of action is considered relevant to all populations and life stages.
According to the EPA's Supplemental Guidance (U. S. EPA. 2005b), there may be increased
susceptibility to early-life exposures to carcinogens with a mutagenic mode of action. Therefore,
because the weight of evidence supports a mutagenic mode of action for EtO carcinogenicity, and in the
absence of chemical-specific data to evaluate differences in susceptibility, increased early-life
susceptibility should be assumed and, if there is early-life exposure, the age-dependent adjustment
factors should be applied, in accordance with the Supplemental Guidance (see Section 4.4).
In addition, as discussed in Section 3.5.2 below, people with DNA repair deficiencies or genetic
polymorphisms conveying a decreased efficiency in detoxifying enzymes may have increased
susceptibility to EtO-induced carcinogenicity (see Sections 3.4.1.2 and 3.4.1.3 and Section C.6 of
Appendix C).
3-66
-------
3.5. HAZARD CHARACTERIZATION
3.5.1. Characterization of Cancer Hazard
In studies of humans there is substantial evidence that EtO exposure is causally associated with
lymphohematopoietic cancers and female breast cancer, but the evidence is not strong enough to be
conclusive. Of the seven relevant12 Hill "criteria" (or considerations) for causality (Hill. 1965).
temporality, coherence, biological plausibility, and analogy are readily satisfied, and the other three
criteria {consistency, biological gradient, and strength of association) are satisfied to varying degrees, as
discussed below.
Temporality, the sole necessary criterion, is satisfied because the subjects of all the epidemiology
studies of EtO were workers who were exposed to EtO before the cancers of interest were observed (i.e.,
exposure preceded the development of the disease).
The related criteria of coherence, biological plausibility, and analogy are fulfilled by the
well-established knowledge that (1) EtO is mutagenic and genotoxic, which are common mechanistic
features of many carcinogens; (2) EtO is carcinogenic in rodents, with lymphohematopoietic cancers
being observed in both rats and mice and mammary carcinomas being observed in female mice; and (3)
EtO is an epoxide, and epoxides are capable of directly interacting with DNA and are the active
metabolites of many carcinogens.
There is evidence of consistency among studies with respect to cancer of the
lymphohematopoietic system as a whole. Most of the studies focus on examining risks associated with
subcategories of the lymphohematopoietic system. These cancers include leukemia, Hodgkin
lymphoma, NHL, reticulosarcoma, and myeloma. [Note that, with the exception of the Steenland et al.
(2004) study, which includes lymphocytic leukemia in a lymphoid cancer category, the studies do not
subcategorize leukemia into its distinct myeloid and lymphocytic subtypes.] In most of the studies, an
enhanced risk of cancer of the lymphohematopoietic system is evident, and in some studies, it is
statistically significant. The studies that do not report a significant lymphohematopoietic cancer effect
have major limitations, such as small numbers of cases (from small study size and/or insufficient
follow-up time), inadequate exposure information, and/or reliance on external analyses (see Table 3-1
and Table A-5 in Appendix A). Overall, about 9 of 11 studies (including only the last follow-up of
independent cohorts) with adequate information to determine RR estimates reported an increased risk of
lymphohematopoietic cancers or a subgroup thereof, although not all were statistically significant,
possibly due to the limitations noted above (see Table 3-1 and Table A-5 in Appendix A). The large,
l2Specificity is not expected for an agent like EtO, which is widely distributed across tissues and is a direct-acting, multisite
mutagen CU.S. EPA. 2005a'). and experimental evidence is seldom available for human populations and is not available in the
case of human exposures to EtO.
3-67
-------
high-quality13 NIOSH study shows statistically significant exposure-response trends for lymphoid
cancers and all lymphohematopoietic cancers [Steenland et al. (2004); see Sections D.3 and D.4 of
Appendix D for results for both sexes combined]. Four other studies reported statistically significant
increases in risk (Svvaen et al.. 1996; Benson and Teta. 1993; Bisanti et al.. 1993; Hogstedt et al.. 1986).
although EtO exposures were reportedly low in the Benson and Teta (1993) study and the increased
risks may be due to other chemical exposures. Nonsignificant increases in lymphohematopoietic cancer
risk were observed in four other studies, based on small numbers of cases (Coggon et al.. 2004; 01 sen et
al.. 1997; Hagmar et al.. 1995; Norman et al.. 1995; Hagmar et al.. 1991) [e.g., Norman et al. (1995) had
only 1 case]. Only 2 of the 11 studies showed no evidence of an increase in lymphohematopoietic
cancer risk (Svvaen et al.. 2009; Kiesselbach et al.. 1990).
Regarding consistency in the breast cancer studies, the large, high-quality NIOSH study shows
statistically significant increased risks for both breast cancer mortality [n = 103 deaths; Steenland et al.
(2004)1 and breast cancer incidence [// = 319 cases; Steenland et al. (2003)1. In addition, a recent
follow-up study of a Swedish cohort of sterilizer workers reported significant increases in the incidence
rate ratios for breast cancer in internal analyses [// = 41 cases; Mikoczv et al. (201 1 )1. Two other studies
suggest an increased risk of breast cancer despite their small size [Norman et al. (1995). n = 12 cases;
Kardos et al. (2003). n = 3 deaths]. No elevated risks were seen in the only other study reporting breast
cancer results; however, that study had few cases [Coggon et al. (2004). n= 11 deaths] (see Table 3-2
and Table A-5 in Appendix A).
There is also some evidence of dose-response relationships {biological gradient). In the large,
high-quality NIOSH study, a statistically significant positive trend was observed in the risk of
lymphohematopoietic cancers with increasing (log) cumulative exposure to EtO, although results for this
model were reported only for males (Steenland et al„ 2004) (the sex difference is not statistically
significant, however, and the trend for both sexes combined is also statistically significant; see
Tables D-31 and D-48 in Appendix D). The results for exposure-response analyses were reported for
only two other cohorts, probably because most cohorts had too few cases and/or lacked adequate
exposure information. In the Swaen et al. (2009) study of the UCC cohort, no statistically significant
trends were observed for leukemia or lymphoid cancer using a Cox proportional hazards model with
cumulative exposure, a model which notably did not yield statistically significant trends in the NIOSH
13The NIOSH study (Steenland et al., 2003, 2004) was judged to be a "high-quality" study based onthe attributes discussed
(see Section 3.1 and Section A.2.8 of Appendix A), including availability of individual worker exposure estimates from a
high-quality exposure assessment (which used a comprehensive job-exposure matrix with exposure-level estimates from a
regression model that incorporated a variety of plant and production variables and that explained 85% of the variability in an
independent data set of EtO measurements), cohort study design, large size, inclusion of males and females, adequate
follow-up, absence of any known confounding exposures, and use of internal comparisons. The breast cancer incidence
study using the subcohort with interviews had the additional attribute of investigating and controlling for several breast
cancer risk factors (Steenland et al., 2003).
3-68
-------
study either. Similarly, no exposure-response relationship was observed for lymphohematopoietic
cancers in internal analyses in the Mikoczy et al. (201 1) study, but this study was limited by a small
number of cases (10 exposed cases of all lymphohematopoietic cancers) and the lack of a nonexposed
referent group. For breast cancer, exposure-response analyses were reported only for the NIOSH cohort
and the Swedish sterilizer worker study of Mikoczy et al. (201 1). again presumably because most
cohorts with female workers had too few cases and/or lacked adequate exposure information. For the
NIOSH cohort, these analyses yielded clear, statistically significant trends for both breast cancer
mortality (Steenland et al.. 2004) and breast cancer incidence (Steenland et al.. 2003) for a variety of
models. The Mikoczy et al. (201 1) study reported significant increases in the incidence rate ratios in the
highest two cumulative exposure quartiles compared to the workers with cumulative exposures below
the median, with the highest RR estimate for the highest exposure quartile.
Whereas most of the considerations are largely satisfied, as discussed above, there is little
strength in the associations, as reflected by the modest magnitude of most of the RR estimates. For
example, in the large NIOSH study, the RR estimate for lymphoid cancer mortality in the highest
exposure quartile is about 3.0, and the RR estimate for breast cancer incidence in the highest exposure
quintile in the subcohort with interviews is on the order of 1.9. While large RR estimates increase the
confidence that an observed association is not likely due to chance, bias, or confounding, modest RR
estimates, such as those observed with EtO, do not preclude a causal association (U.S. EPA. 2005a).
With EtO, the modest RR estimates may, in part, reflect the relatively high background rates of these
cancers, particularly of breast cancer incidence.
In addition to the Hill criteria, other factors such as chance, bias, and confounding are considered
in analyzing the weight of epidemiological evidence. Given the consistency of the findings across
studies and the exposure-response relationships observed in the largest study, none of these factors is
likely to explain the associations between these cancers and EtO exposure. Coexposures to other
chemicals are expected to have occurred for workers in the chemical industry cohorts but would have
been much less likely in the sterilizer worker cohorts, such as the NIOSH cohort, which reported no
evidence of confounding exposures to other occupational carcinogens (Steenland et al„ 1991). For
breast cancer in the NIOSH subcohort with interviews (Steenland et al., 2003), other risk factors for
breast cancer were assessed, and statistically significant factors were included in the exposure-response
models.
In conclusion, the overall epidemiological evidence for causal associations between EtO
exposure and lymphohematopoietic cancer as well as female breast cancer was judged to be strong but
less than conclusive.
There is inadequate evidence for other cancer types (e.g., stomach cancer and pancreatic cancer)
in the epidemiology studies.
3-69
-------
The laboratory animal evidence for carcinogenicity is concluded to be "sufficient" based on
findings of tumors at multiple sites, by both oral and inhalation routes of exposure, and in both sexes of
both rats and mice. Tumor types resulting from inhalation exposure included mononuclear cell
leukemia in male and female rats and malignant lymphoma and mammary carcinoma in female mice,
suggesting some site concordance with the lymphohematopoietic and breast cancers observed in
humans, also exposed by inhalation.
The evidence of EtO genotoxicitv and mutagenicity is unequivocal. EtO is a direct-acting
alkylating agent and has invariably tested positive in in vitro mutation assays from bacteriophage,
bacteria, fungi, yeast, insects, plants, and mammalian cell cultures (including human cells). In
mammalian cells (including human cells), EtO-induced genotoxic effects include unscheduled DNA
synthesis, gene mutations, SCEs, and chromosomal aberrations. The results of in vivo genotoxicity
studies of EtO have also been largely positive, following ingestion, inhalation, or injection. Increases in
frequencies of gene mutations have been reported in the lung, T-lymphocytes, bone marrow, and testes
of EtO-exposed mice. In particular, increases in frequencies of proto-oncogene mutations have been
observed in several tumor types from EtO-exposed mice compared to spontaneous mouse tumors of the
same types. Inhalation studies in laboratory animals have demonstrated that EtO exposure levels in the
range of those used in the rodent bioassays induce SCEs in several species, including rats, and
consistently induce chromosomal aberrations in mice (no inhalation studies of SCEs or micronuclei are
available for mice). Evidence for micronuclei and chromosomal aberrations in rats from these same
exposure levels in short-term studies (4 weeks or less) is lacking, although concerns have been raised
about some of the negative studies. A recent mouse study showed clear, statistically significant
increases in chromosomal aberrations with longer durations of exposure (>12 weeks) to the
concentration levels used in the rodent bioassays. The studies of point mutations in EtO-exposed
humans are few and insensitive and the evidence for mutations is limited; however, there is clear
evidence from a number of human studies that EtO causes chromosomal aberrations, SCEs, and
micronucleus formation in peripheral blood lymphocytes, and one study has reported increased levels of
micronuclei in bone marrow cells in EtO-exposed workers.
Under the EPA's 2005 Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a), the
conclusion can be made that EtO is "carcinogenic to humans." In general, the descriptor "carcinogenic
to humans" is appropriate when there is convincing epidemiologic evidence of a causal association
between human exposure and cancer. This descriptor is also appropriate when there is a lesser weight of
epidemiologic evidence that is strengthened by specific lines of evidence set forth in the guidelines (U.S.
EPA. 2005a). which are satisfied for EtO. The lines of evidence supporting the characterization of
"carcinogenic to humans" include the following: (1) there is strong, although less than conclusive on its
own, evidence of cancer in humans associated with EtO exposure via inhalation, specifically, evidence
of lymphohematopoietic cancers and female breast cancer in EtO-exposed workers; (2) there is
3-70
-------
extensive evidence of EtO-induced carcinogenicity in laboratory animals, including
lymphohematopoietic cancers in rats and mice and mammary carcinomas in mice following inhalation
exposure; (3) EtO is a direct-acting alkylating agent whose mutagenic and genotoxic capabilities have
been well established in a variety of experimental systems, and a mutagenic mode of carcinogenic action
has been identified in animals involving the key precursor events of DNA adduct formation and
subsequent DNA damage, including point mutations and chromosomal effects; and (4) there is strong
evidence that the key precursor events are anticipated to occur in humans and progress to tumors,
including evidence of chromosome damage, such as chromosomal aberrations, SCEs, and micronuclei in
EtO-exposed workers.
3.5.2. Susceptible Life Stages and Populations
There are no data on the relative susceptibility of children and the elderly when compared with
adult workers, in whom the evidence of hazard has been gathered, but because EtO does not have to be
metabolized before binding to DNA and proteins, the maturing of enzyme systems in very young
children is thought not to be a predominant factor in its hazard, at least for activation. However, the
immaturity of detoxifying enzymes in very young children may increase children's susceptibility
because children may clear EtO at a slower rate than adults. As discussed in Section 3.3.1, EtO is
metabolized (i.e., detoxified) primarily by hydrolysis in humans but also by glutathione conjugation.
Both hydrolytic activity and glutathione-S-transferase activity apparently develop after birth (Clewell et
al.. 2002); thus, very young children might have a decreased capacity to detoxify EtO compared to
adults. In the absence of data on the relative susceptibility associated with EtO exposure in early life,
increased early-life susceptibility is assumed, in accordance with the EPA's Supplemental Guidance
(U.S. EPA. 2005b), because the weight of evidence supports the conclusion of a mutagenic mode of
action for EtO carcinogenicity (see Section 3.4).
Other than the occurrence of sex-specific cancers (e.g., breast cancer in human females,
mammary and uterine carcinomas in female mice, and testicular peritoneal mesotheliomas in male rats;
see Section 3.2), there is no clear sex difference in EtO-induced carcinogenicity. With the exception of
the sex-specific cancers and the observation of malignant lymphomas in female but not male mice, there
is no sex difference in EtO-induced cancer types in the rat and mouse bioassays. Cancer potency
estimates for females are roughly 50% higher than those for males for both mice and rats (see
Table 4-20 in Section 4.2.5). In humans, in the large NIOSH study (Steenland et al.. 2004). the
association between lymphoid cancers and EtO exposure was seen primarily in males, but the sex
difference was not statistically significant (see Section D.3.3 in Appendix D).
Brown et al. (1996) reported that sex differences in EtO toxicokinetics were observed in mice but
not in rats; female mice had a significantly higher steady-state blood EtO concentration after 4 hours of
exposure to either 100 or 330 ppm than male mice. As noted above and discussed in Section 3.3.1, EtO
3-71
-------
is metabolized primarily by hydrolysis in humans. Mertes et al. (1985) reported no sex difference in
microsomal or cytosolic epoxide hydrolase activities in human liver in vitro using benzo[a]pyrene
4,5-oxide or /ra//.s-stilbene oxide, respectively, as substrates. Using EtO as a substrate, but with far
fewer subjects, Fennel! and Brown (2001) reported similar values for males and females for epoxide
hydrolase activity in human liver microsomes and for glutathione transferase in human liver cytosolic
fractions.
Because EtO is detoxified by glutathione conjugation or hydrolysis, people with genotypes
conveying deficiencies in glutathione-S-transferase or epoxide hydrolase activities may be at increased
risk of cancer from EtO exposure. Yong et al. (2001) measured approximately twofold greater
EtO-hemoglobin adduct levels in occupationally exposed persons with a null GSTT1 genotype than in
those with positive genotypes. Similarly, in a study of hospital workers, Haufroid et al. (2007) reported
increased urinary excretion of a glutathione conjugate of EtO, reflecting increased detoxification of EtO,
associated with a nonnull GSTT1 genotype, although the increase was not statistically significant in all
the regression models tested; associations were less clear for other glutathione-S-transferase or epoxide
hydrolase polymorphisms.
In addition, people with DNA repair deficiencies such as xeroderma pigmentosum, Bloom's
syndrome, Fanconi anemia, and ataxia telangiectasia (Gelehrter et al.. 1990) are expected to be
especially sensitive to the damaging effects of EtO exposure. Paz-v-Mino et al. (2002) have recently
identified a specific polymorphism in the excision repair pathway gene hMSH2. The polymorphism was
present in 7.5% of normal individuals and in 22.7% of NHL patients, suggesting that this polymorphism
may be associated with an increased risk of developing NHL.
3-72
-------
4. CANCER DOSE-RESPONSE ASSESSMENT FOR INHALATION EXPOSURE
Chapter 4 presents the derivation of cancer unit risk estimates from human and rodent data
and discusses the sources of uncertainty in these estimates.
Major findings of Chapter 4:
1. Full lifetime cancer incidence unit risk (upper bound; adjusted with age-dependent
adjustment factors, see Section 4.4) estimate based on human data (lymphoid
cancer and breast cancer in females): 5.0 x lo3 per [j,g/m3 (9.1 per ppm).
2. Unadjusted adult-based unit risk (upper-bound) estimate for use with
age-dependent adjustment factors (see Section 4.4): 3.0 x 10-3 per [j,g/m3 (5.5 per
ppm).
3. Upper-bound estimates of the extra risk of lymphoid cancer and breast cancer
incidence combined for the range of occupational exposure scenarios considered
(i.e., 0.1 to 1 ppm 8-hr TWA for 35 years) (see Section 4.7): 0.081-0.22.
This chapter presents the derivation of cancer unit risk estimates from human and rodent data.
Section 4.1 discusses the derivation of unit risk estimates for lymphohematopoietic cancers, breast
cancer, and total cancer from human data, as well as sources of uncertainty in these estimates. (Note
that the estimates in Section 4.1 were derived under the common assumption that relative risk is
independent of age. This assumption is later superseded by an assumption of increased early-life
susceptibility, and it is the unit risk estimates derived under this latter assumption, which are developed
in Section 4.4, that are the ultimate estimates proposed in this assessment.) Section 4.2 presents the
derivation of unit risk estimates from rodent data. Section 4.3 summarizes the unit risk estimates
derived from the different data sets. Section 4.4 discusses adjustments for assumed increased early-life
susceptibility, based on recommendations from the EPA's Supplemental Guidance (U. S. EPA. 2005b),
because the weight of evidence supports the conclusion of a mutagenic mode of action for EtO
carcinogenicity (see Section 3.4). Section 4.5 presents conclusions about the unit risk estimates. Figure
4-1 presents a flowchart of key steps in the derivation of the unit risk estimate. Section 4.6 compares the
unit risk estimates derived in this EPA assessment to those derived in other assessments. Finally,
Section 4.7 provides risk estimates derived for some general occupational exposure scenarios.
4-1
-------
Human Data (S ection 4.1)
Selection of NIOSH study datasets
Lymphoid cancer
mortality
(S ection 4.1.1)
- All
lymphohematopoietic
canc er mortality
-(Section 4.1.1)
) (
2-piece linear spline
model with knot at 1.600
ppmxdays
i
to
Lymphoid
cancer )
mortality
Lymphoid
cancer
incidence
No final
selection
LHC
incidence
mortality
Breast cancer
mortality )
(Section 4.0.2)
2-piece log-linear
spline model with knot
at 700 ppmxdays
Breast cancer incidence
(Section 4.1.2.3)
2-piece linear spline
model with knot at 5,750
ppm^days
Default* total cancer unit risk estimate
(lymphoid plus breast cancer)
Section (4.1.3)
Breast cancer
mortality.
Breast cancer
incidence
MODEL
SELECTION
DERIVATION
OF DEFAULT*
ENDPOENT-
SPECIHC
UNIT RISK
ESTIMATES
USING LIFE
TABLES
COMBINATION OF
ENDPOINT -SPE CIFIC O-TT
RISK ESTIMATES
Final unit risk estimates
(Section 4.5)
ADJUSTMENTS FOR
POTENTIAL INCREASED
EARLY-LIFE SUSCEPTIBILITY
(SECTION 4.4)
* Under assumption that relative risk is independent of age for all ages.
Figure 4-1. Flowchart of key steps in the derivation of the unit risk estimate from human data. Rectangles depict
preferred options; ovals depict comparison analyses.
-------
4.1. INHALATION UNIT RISK ESTIMATES DERIVED FROM HUMAN DATA
The NIOSH retrospective cohort study of 17,530 workers in 13 sterilizing facilities with
sufficient exposure information [most recent follow-up by Steenland et al. (2004) and Steenland et al.
(2003)1 provides the most appropriate data sets for deriving quantitative cancer risk estimates in humans
for several reasons: (1) exposure estimates were derived for the individual workers using a
comprehensive exposure assessment, (2) the cohort was large and diverse (e.g., 55% female), and
(3) there was little reported exposure to chemicals other than EtO. Exposure estimates, including
estimates for early exposures for which no measurements were available, were determined using a
regression model that estimated exposures to each individual as a function of facility, exposure category,
and time period. The regression model was based on extensive personal monitoring data from
18 facilities from 1976 to 1985 as well as information on factors influencing exposure, such as
engineering controls [Hornung et al. (1994); see also Section A.2.8 in Appendix A], When evaluated
against independent test data from the same set of monitoring data, the model accounted for 85% of the
variation in average EtO exposure levels. The investigators were then able to estimate the cumulative
exposure (ppm x days) for each individual worker by multiplying the estimated exposure for each job
(exposure category) held by the worker by the number of days spent in that job and summing over all
the jobs held by the worker. Steenland et al. (2004) present follow-up results for the cohort mortality
study previously discussed by Steenland et al. (1991) and Stayner et al. (1993). Positive findings in the
current follow-up include increased rates of (lympho)hematopoietic cancer mortality and of breast
cancer mortality in females. Steenland et al. (2003) present results of a breast cancer incidence study of
a subcohort of 7,576 women from the NIOSH cohort.
The other major occupational study [most recent follow-up by Swaen et al. (2009)1 described
risks to workers exposed to EtO at two Union Carbide Corporation (UCC) chemical plants in West
Virginia, but this study is less useful for estimating quantitative cancer risks for a number of reasons.
First, the exposure assessment is much less extensive than that used for the NIOSH cohort, with greater
likelihood for exposure misclassification, especially in the earlier time periods when no measurements
were available (1925-1973). Exposure estimation for the individual workers was based on a relatively
crude exposure matrix that cross-classified three levels of exposure intensity with four time periods.
The exposure estimates for 1974-1988 were based on measurements from air sampling at the West
Virginia plants since 1976. The exposure estimates for 1957-1973 were based on measurements in a
similar plant in Texas. The exposure estimates for 1940-1956 were based loosely on a "rough" estimate
reported for chlorohydrin-based EtO production in a Swedish facility in the 1940s (Hogstedt et al..
1979). The exposure estimates for 1925-1939 were further conjectures based on the Swedish 1940s
estimate. Thus, for the two earliest time periods (1925-1939 and 1940-1956) at least, the exposure
estimates are highly uncertain. (See Section A.2.20 of Appendix A for a more detailed discussion of the
exposure assessment for the UCC cohort.) This is in contrast to the NIOSH exposure assessment in
4-3
-------
which exposure estimates were based on extensive sampling data and regression modeling. In addition,
the sterilization processes used by the NIOSH cohort workers were fairly constant historically, unlike
chemical production processes, which likely involved much higher and more variable exposure levels in
the past. Furthermore, the UCC cohort is of much smaller size and has far fewer deaths than the NIOSH
cohort, it is restricted to males and so cannot be used to investigate breast cancer risk in females, and
there are coexposures to other chemicals.
A third study (Hagmar et al.. 1995; Hagmar et al.. 1991) estimated cumulative exposures for
individual sterilizer workers; however, insufficient exposure-response data are presented for the
derivation of unit risk estimates. A more recent follow-up of this cohort (Mikoczv et al.. 201 1) provides
exposure-response results based on a greater number of cases; however, in the internal analyses,
incidence rate ratios were calculated by comparing the incidence rates for the two highest cumulative
exposure quartiles with that for the 50% of workers with cumulative exposures below the median. Such
results are not well suited to the derivation of unit risk estimates, and the EPA does not have the
individual data to model. Obtaining the data was not pursued because the NIOSH sterilizer worker
study is much larger and has many more cases.
Table 4-1 provides a summary of the considerations taken into account in selecting the NIOSH
study as the basis for the derivation of unit risk estimates. The NIOSH EtO cohort mortality data can be
obtained from the Industrywide Studies Branch of NIOSH.14
"Industrywide Studies Branch; Division of Surveillance, Hazard Evaluations and Field Studies: NIOSH; Centers for Disease
Control and Prevention, 4676 Columbia Parkway MS R-13, Cincinnati, OH 45226, telephone: 513-841-4203.
4-4
-------
Table 4-1. Considerations used in this assessment for selecting epidemiology
studies for quantitative risk estimation
Consideration
Selected studies
Comments
Availability of
quantitative exposure
estimates
1. Mikoc/v et al. (2011) llatest
follow-up of Swedish sterilizer
cohort studies bv Haemar et al.
(1995) and Haemar et al. (1991)1
2. Swaenet al. (2009) llatest follow-
up of UCC cohort]
3. Steenland et al. (2004) and
Steenland et al. (2003) (latest
follow-up of NIOSH cohort)
These are the only three studies with quantitative
exposure estimates, which is an essential
criterion for quantitative risk estimation.
Availability of exposure-
response information
1. Swaenet al. (2009)
2. Steenland et al. (2004) and
Steenland et al. (2003)
The grouped exposure-response results reported
bv Mikoczv et al. (2011) are not well suited to
derivation of unit risk estimates.
Other factors affecting the
utility of epidemiology
studies for quantitative
risk estimation
Steenland et al. (2004) and Steenland
The NIOSH studv [Steenland et al. (2004) and
et al. (2003)
Steenland et al. (2003)1 alone was selected for
quantitative risk estimation, as it was judged to
be substantially superior to the UCC study
(Swaenet al.. 2009) withresoectto a number of
key considerations [in particular, in order of
importance: (1) quality of the exposure
estimates, (2) cohort size, and (3) the absence of
coexposures and the inclusion of women].
The following subsections present the derivation of unit risk estimates, defined as the lifetime
risk of cancer from chronic inhalation of EtO per unit of air concentration, for lymphohematopoietic
cancer mortality and incidence and for breast cancer mortality and incidence in females, based on results
of the recent analyses of the NIOSH cohort.
The exposure-response models used to fit the epidemiological data are empirical "curve-fitting"
models. Model selection for each cancer data set was primarily based on a preference for models of the
individual-level continuous exposure data, prioritization of models that are more tuned to local behavior
in the low-exposure data, and a weighing of statistical and biological considerations. All of the
exposure-response models considered in this assessment are fitted treating exposure as a continuous
variable except for the categorical models and the linear regressions of categorical results, which are
explicitly described as such. The continuous exposure models used are members of the set of general
relative hazard models, which have the form h(7) = ho(7) x cp, or h(7)/ho(7) = (p, where Hi) is the hazard
function, ho(7) is the baseline hazard, and (pis a function of fi, the parameters to be estimated, and z,
a vector of the explanatory data from the epidemiology study [see, e.g., Lanuholz and Richardson
(2010)1. Relative hazard models with two different functional forms were considered in this assessment,
and recognizing that the hazard function ratio expresses relative risk, these are also relative risk (or,
more specifically for this assessment, relative rate [RR]) models. One is an exponential model obtained
4-5
-------
when cp = ePz (Cox proportional hazards model); in this assessment, models of this form will be referred
to as Cox proportional hazards or Cox regression models or log-linear models because they are linear in
(natural) log RR. The second is the linear model, or cp = 1 + fiz [referred to as the linear excess relative
risk model (ERR) by Langholz and Richardson (2010)1; in this assessment, models of this form will be
referred to as linear models.
4.1.1. Risk Estimates for Lymphohematopoietic Cancer
4.1.1.1. Lymphohematopoietic Cancer Results from the NIOSH Study
Steenland et al. (2004) investigated the relationship between (any) EtO exposure and mortality
from cancer at a number of sites using life-table analyses with the U. S. population as the comparison
population. Categorical SMR analyses were also done by quartiles of cumulative exposure. Then, to
further investigate apparent exposure-response relationships observed for (lympho)hematopoietic cancer
and breast cancer, internal exposure-response analyses were conducted using Cox proportional hazards
models, which, when exposure is the only variable in the model, have the form
Relative rate (RR) = d!X">, (4-1)
where represents the regression coefficient and X(/) is the time-dependent exposure (or some function
of exposure, e.g., the natural log of exposure). Internal analyses were done two ways—with exposure as
a categorical variable and with exposure as a continuous variable. A nested case-control approach was
used, with age as the time variable used to form the risk sets. Risk sets were constructed with
100 controls randomly selected for each case from the pool of those surviving to at least the age of the
index case. According to the authors, use of 100 controls per case has been shown to result in ORs
virtually identical to the RR estimates obtained with full cohorts. Cases and controls were matched on
race (white/nonwhite), sex, and date of birth (within 5 years). Exposure was the only variable in the
model, so thep-walue for the model also serves as ap-walue for the regression coefficient, P, as well as
for a test of exposure-response trend.
For lymphohematopoietic cancer mortality, Steenland et al. (2004) analyzed both all
lymphohematopoietic cancers combined and a subcategory of lymphohematopoietic cancers that they
called "lymphoid" cancers; these included NHL, myeloma, and lymphocytic leukemia. Their
exposure-response analyses focused on cumulative exposure and (natural) log cumulative exposure, with
various lag periods. Other EtO exposure metrics (duration of exposure, average exposure, and peak
exposure) were also examined, but models using these metrics did not generally predict
lymphohematopoietic cancer as well as models using cumulative exposure. A lag period defines an
interval before death, or end of follow-up, during which any exposure is disregarded because these
4-6
-------
exposures have likely occurred after the onset of disease. For lymphohematopoietic (and lymphoid)
cancer mortality, a 15-year lag provided the best fit to the data, based on the likelihood ratio test. As is
commonly done, 1 ppm x day was added to cumulative exposures in analyses using the log of
cumulative exposure with a lag, to avoid taking the log of 0. For both all lymphohematopoietic and
lymphoid cancers, Steenland et al. (2004) found stronger positive exposure-response trends in males and
so presented the results for some of the regression models separately by sex. The apparent sex
difference was not statistically significant (see Appendix D), however, and results for both sexes
combined were subsequently obtained from Steenland (see Appendix D; Section 3 for lymphoid cancer,
Section 4 for all lymphohematopoietic cancer). These results are presented in Table 4-2. For additional
details and discussion of the Steenland et al. (2004) study, see Appendix A, and for more details about
the exposure and other characteristics of the full cohort and the lymphoid cancer risk sets, see Section
D.5 of Appendix D.
Table 4-2. Cox regression results for all lymphohematopoietic cancer and
lymphoid cancer mortality in both sexes in the National Institute for
Occupational Safety and Health cohort, for the models presented by
Steenland et al. (2004)
Exposure variable"
7?-valueb
Coefficient (SE)
(per ppm x day)
ORs by category0 (95% CI)
All lymphohematopoietic cancer1'
Cumulative exposure, 15-yr lag
0.40
3.26 x 10"6
(3.49 x 10"6)
Log cumulative exposure, 15-yr
lag
0.009
0.107 (0.0418)
Categorical cumulative
exposure, 15-yr lag
0.10
1.00, 2.33 (0.93-5.86), 3.46 (1.33-8.95),
3.02 (1.16-7.89), 2.96 (1.12-7.81)
Lymphoid cancer1
Cumulative exposure, 15-yr lag
0.22
4.74 x 10"6
(3.35 x 10"6)
Log cumulative exposure, 15-yr
lag
0.02
0.112(0.0486)
Categorical cumulative
exposure, 15-yr lag
0.21
1.00, 1.75 (0.59-5.25), 3.15 (1.04-9.49),
2.44 (0.80-7.50), 3.00 (1.02-8.45)
4-7
-------
Table 4-2. Cox regression results for all lymphohematopoietic cancer and
lymphoid cancer mortality in both sexes in the National Institute for
Occupational Safety and Health cohort, for the models presented by
Steenland et al. (2004) (continued)
Cumulative exposure is in ppm x days.
'/^-values from likelihood ratio test.
cExposure categories are 0; >0-1,199; 1,200-3,679; 3,680-13,499; >13,500 ppm x days.
d9th revision ICD codes 200-208; results based on 74 cases.
eNHL, myeloma, and lymphocytic leukemia (9th revision ICD codes 200, 202, 203, 204); results based on 53 cases.
Source: Additional analyses performed by Steenland (see Sections D.3 and D.4 of Appendix D).
SE = standard error; yr = year.
4.1.1.2. Prediction of Lifetime Extra Risk of Lymphohematopoietic Cancer Mortality
The exposure-response trends for lymphohematopoietic cancers observed by Steenland et al.
(2004) appear to be driven largely by the lymphoid cancers; therefore, the primary risk analyses for
lymphohematopoietic cancer are based on the lymphoid cancer results. Lymphohematopoietic cancers
are a diverse group of diseases with diverse etiologies, and myeloid and lymphoid cells develop from
different progenitor cells; thus, there is stronger support for an etiologic role of EtO in the development
of lymphoid cancers than in the development of the cancers in the aggregate all lymphohematopoietic
cancer category. The consideration of NHL, (plasma cell) myeloma, and lymphocytic leukemia together
as "lymphoid" cancers is consistent with the current World Health Organization classification of such
cancers based on their derivation from B-cells, T-cells, and NK-cells rather than previous distinctions
(Harris et al., 1999).
Nonetheless, for comprehensiveness and for the reasons listed below, risk estimates based on the
all lymphohematopoietic cancer results are presented for comparison. Judging roughly from the
/(-values, the model fits do not appear notably better for lymphoid cancers than for all
lymphohematopoietic cancers (see Table 4-2, /^-values for log cumulative exposure models), and the
"lymphoid" category did not include Hodgkin lymphoma, which also exhibited evidence of
exposure-response trends, although based on few cases (Steenland et al., 2004). In addition,
misclassification or nonclassification of tumor type is more likely to occur for subcategories of
lymphohematopoietic cancer (e.g., 4 of the 25 leukemias in the analyses were classified as "not
specified" and so could not be considered for the lymphoid cancer analysis).
For lymphoid cancer, the results of internal exposure-response analyses of lymphoid cancer in
the NIOSH cohort (Cox regression analyses, summarized in Table 4-2) were used for predicting the
extra risks of lymphoid cancer mortality from continuous environmental exposure to EtO. Extra risk is
defined as
4-8
-------
Extra risk = (Rx - R0)/(l ~ R0),
(4-2)
where Rx is the lifetime risk in the exposed population and R0 is the lifetime risk in an unexposed
population (i.e., the background risk). These risk estimates were calculated using the P regression
coefficients and an actuarial program (life-table analysis) that accounts for competing causes of death.15
An inherent assumption in the Cox regression model and its application in the life-table analyses is that
RR is independent of age. [An alternate assumption of increased susceptibility from early-life exposure
to EtO, as recommended in the EPA's Supplemental Guidance (U. S. EPA. 2005b) for chemicals such as
EtO with a mutagenic mode of action (see Section 3.4), is considered in Section 4.4. This alternate
assumption is the prevailing assumption in this assessment, based on the recommendations in the
Supplemental Guidance. However, risk estimates are first developed under the assumption of age
independence because that is the standard approach in the absence of evidence to the contrary or of
sufficient evidence of a mutagenic mode of action to invoke the divergent assumption of increased
early-life susceptibility.]
For the life-table analysis, U.S. age-specific mortality rates (for both sexes of all races combined)
for all causes and for the relevant subcategories of lymphohematopoietic cancer (NHL [C82-C85 of 10th
revision of the International Classification of Diseases (ICD)], multiple myeloma [C88, C90], and
lymphoid leukemia [C91]) were obtained from the Centers for Disease Control and Prevention
WONDER Online Database for 2008-2012 (CDC. 2015). The risks were computed up to age 85 years
for continuous exposures to EtO beginning at birth.16 Conversions between occupational EtO exposures
and continuous environmental exposures were made to account for differences in the number of days
exposed per year (240 vs. 365 days) and in the amount of EtO-contaminated air inhaled per day [10 vs.
20 m3; U.S. EPA (1994)1. An adjustment was also made for the lag period. The reported standard
errors for the regression coefficients from Table 4-2 were used to compute the 95% upper confidence
limits (UCLs) for the relative rates, based on a normal approximation.
The only statistically significant Cox regression model presented by Steenland et al. (2004) for
lymphoid cancer mortality was for log cumulative exposure with a 15-year lag in males (p = 0.02). This
was similarly true for the analyses of lymphoid cancer using the data for both sexes (see Table 4-2).
15This program is an adaptation of the approach previously used by the Committee on the Biological Effects of Ionizing
Radiation (BEIR. 1988). A spreadsheet illustrating the extra risk calculation for the derivation of the LECoi for lymphoid
cancer incidence (see Section4.1.1.3) is presented in AppendixE.
16Rates above age 85 years are not included because cause-specific disease rates are less stable for those ages. Note that
85 years is not employed here as an average lifespanbut, rather, as a cutoff point for the life-table analysis, which uses actual
age-specific mortality rates. The average lifespan for males and females combined in a life-table analysis truncated at age
85 years is about 75 years.
4-9
-------
However, using the log cumulative exposure model to estimate the risks from low environmental
exposures is problematic because this model, which is intended to fit the full range of occupational
exposures in the study, is inherently supralinear (i.e., risk increases steeply with increasing exposures in
the low exposure range and then plateaus), with the slope approaching infinity as exposures decrease
towards zero, and results can be unstable for low exposures (i.e., small changes in exposure correspond
to large changes in risk; see Figure 4-2). Some consideration was thus given to the cumulative exposure
model, which is typically used and which is generally stable at low exposures (i.e., small changes in
exposure do not correspond to large changes in risk), although the fit to these data was not statistically
significant (p = 0.22). However, the Cox regression model with cumulative exposure is inherently
sublinear (i.e., risk increases gradually in the low exposure range and then with increasing steepness as
exposure increases) and does not reflect the apparent supralinearity of the data demonstrated by the
categorical results and the superior fit of the log cumulative exposure model; thus, this model was not
considered further.
4-10
-------
3.5
3.0
2.5
2.0
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
— • — eA(p*exp)
eA(p*logexp)
• categorical
linear
splinel 00
cumulative exposure (ppm * days)
eA(P*exp): Cox regression results for RR = e(P exposure); eA(P*logexp): Cox regression results for RR = e(P ^exposure)),
categorical: Cox regression results for RR = e(P exP°sure) with categorical exposures, plotted at the mean cumulative
exposure; linear: weighted linear regression of categorical results, excluding highest exposure group (see text);
splinelOO: two-piece log-linear spline model with knot at 100 ppm x days (see text). (Note that, with the exception of
the categorical results and the linear regression of the categorical results, the different models have different implicitly
estimated baseline risks; thus, they are not strictly comparable to each other in terms of RR values, i.e., along the
j'-axis. They are, however, comparable in terms of general shape.)
Source: Steenland reanalyses for males and females combined; see Appendix D (except for linear regression of
categorical results, which was done by the EPA).
Figure 4-2. RR estimate for lymphoid cancer vs. occupational cumulative exposure (with 15-year lag).
-------
In a 2006 external review draft of this assessment (U.S. EPA. 2006a). which relied on the
original published results of Steenland et al. (2004). the EPA proposed that the best way to represent the
exposure-response relationship in the lower exposure region, which is the region of interest for
low-exposure extrapolation, was through the use of a weighted linear regression of the results from the
Cox regression model with categorical cumulative exposure and a 15-year lag [for males only, as this
was the significant finding in the published paper of Steenland et al. (2004)1. In addition, the highest
exposure group was not included in the regression to alleviate some of the "plateauing" in the
exposure-response relationship at higher exposure levels and to provide a better fit to the lower exposure
data. Linear modeling of categorical (i.e., grouped) epidemiologic data and elimination of the highest
exposure group(s) under certain circumstances to obtain a better fit of low-exposure data are both
standard techniques used in EPA dose-response assessments (U.S. EPA. 2012. 2005a). An established
methodology was employed for the weighted linear regression of the categorical epidemiologic data, as
described by Rothman (1986) and used by others [e.g., van Wijngaarden and Hertz-Picciotto (2004)1.
For the subsequent draft assessment, the EPA pursued modeling the individual continuous exposure data
as an alternative to modeling the published grouped data (U.S. EPA. 2014a. b). In addition, both males
and females were included in the modeling of lymphohematopoietic cancer mortality. In consultation
with Steenland, one of the investigators from the NIOSH cohort studies, the EPA determined that an
alternative way to address the supralinearity of the data when using the full continuous exposure data set
(while avoiding the extreme low-exposure curvature obtained with the log cumulative exposure model)
might be to use a two-piece log-linear spline model.
Spline models have been used previously for exposure-response analyses of epidemiological data
(Steenland and Deddens. 2004; Steenland et al., 2001). These models are generally useful for
exposure-response data such as the EtO lymphoid cancer data, for which RR initially increases with
increasing exposure but then tends to plateau, or attenuate, at higher exposures. Such plateauing
exposure-response relationships have been seen with other occupational carcinogens and may occur for
various reasons, including the depletion of susceptible subpopulations at high exposures,
mismeasurement of high exposures, or a healthy worker survivor effect (Stayner et al., 2003). No other
traditional exposure-response models for continuous exposure data that might suitably fit the observed
exposure-response pattern were apparent. Steenland was contracted to do the spline analyses using the
full data set with cumulative exposure as a continuous variable, and his findings are included in
Appendix D (see Section D.3 for lymphoid cancer, Section D.4 for all lymphohematopoietic cancer).
The results of the spline analyses are presented below.
4-12
-------
For the two-piece log-linear spline modeling approach, the Cox regression model (eq 4-1) was
the underlying basis for the splines which were fit to the lymphoid cancer exposure-response data.17
Taking the log of both sides of eq 4-1, log RR is a linear function of exposure (cumulative exposure is
used here), and with the two-piece log-linear spline approach, log RR is a function of two lines which
join at a single point of inflection, called a "knot." The use of the two-piece spline model form is not
intended to imply that an abrupt change in biological response occurs at the knot but, rather, to allow
description of an exposure-response relationship in which the slope of the relationship differs notably in
the low-exposure versus high-exposure regions. The shape of the two-piece log-linear spline model, in
particular the slope in the low-exposure region, depends on the location of the knot. For this assessment,
the knot was generally selected by evaluating different knots in increments of 100 ppm x days over
some range of cumulative exposures starting at 0 and then choosing the one that resulted in the best
(largest) model likelihood. The model likelihood did not change much across the different trial knots for
any of the data sets, but it did change slightly, and the largest calculated likelihood was used as the basis
for knot selection. For more discussion of the two-piece spline approach, see Appendix D.
For the lymphoid cancer data, the range examined for knot selection was from 0 to
15,000 ppm x days, and the largest model likelihood was observed with the knot at 100 ppm x days,
although, as noted above, the model likelihood did not actually change much across the different trial
knots (see Figure D-14 of Appendix D). This model yielded a very steep slope in the exposure range
below the knot of 100 ppm x days (see Figure 4-2),18 and there was low confidence in the slope, given
the limited data in the low-exposure region.
A two-piece linear spline model (with a linear model, i.e., RR = 1 + P x exposure, as the
underlying basis for the spline pieces) was also attempted using the then just-published approach of
Langholz and Richardson (2010) to model the individual data with cumulative exposure as a continuous
variable; however, this model did not alleviate the problem of the excessively steep low-exposure spline
segment (see Figure D-18 in Appendix D) and was not pursued further at that time. The Langholz and
Richardson (2010) approach was also employed to model the lymphoid cancer data using linear RR
models with cumulative exposure and log cumulative exposure as continuous variables; however, these
linear models similarly did not alleviate the problems of the corresponding log-linear RR models (see
Figure D-18 in Appendix D).
Therefore, after examining the modeling analyses, it was proposed in the subsequent SAB
review draft (U.S. EPA. 2014a, b) that the weighted linear regression of the categorical results still
provided the best available approach for deriving the risk estimates for lymphoid cancer. For the
weighted linear regression, the Cox regression results from the model with categorical cumulative
17As parameterized in Appendix D, for cumulative exposures less than the value of the knot, RR = e(|31 x e^os"rci; for
cumulative exposures greater than the value of the knot, RR = e(|31 xejp°s™e+P2x(exPosure knot))
18Althoughthe log-linear spline segments appear fairly linear in the plotted range, they are not strictly linear.
4-13
-------
exposure and a 15-year lag (see Table 4-2) was used, excluding the highest exposure group, as discussed
above.19 The weights used for the ORs were the inverses of the variances, which were calculated from
the confidence intervals.20 Mean and median exposures for the cumulative exposure groups were
provided by Steenland (see Table D-26 of Appendix D).21 The mean values were used for the weighted
regression analysis because, under this model, the cancer response is presumed to be a linear function of
cumulative exposure, which is expected to be best represented by mean exposures. If the median values
had been used, a slightly larger regression coefficient would have been obtained, resulting in slightly
larger risk estimates.22 See Table 4-3 for the results obtained from the weighted linear regression and
Figure 4-2 for a depiction of the resulting model.
Table 4-3. Linear regression of categorical results—modeling results for all
lymphohematopoietic cancer and lymphoid cancer mortality in both sexes in
the National Institute for Occupational Safety and Health cohort3
Cancer endpoint
/>-valueb
Coefficient (SE)
(per ppm x day)
All lymphohematopoietic cancer0
0.08
3.459 x 10-4 (1.944 x 10"4)
Lymphoid cancerd
0.18
2.47 x 10"4 (1.85 x 10"4)
"With cumulative exposure inppm x days as the exposure variable, witha 15-yr lag; excluding the highest exposure
category.
bWald /^-values.
c9lh revision ICD codes 200-208; results based on 74 cases.
dNHL, myeloma, and lymphocytic leukemia (9th revision ICD codes 200, 202, 203, 204); results based on 53 cases.
SE = standard error.
For this final assessment, the EPA reassessed the modeling options. First, however, the EPA
revisited the issue of lag selection for the lymphoid cancer mortality data. After considering model fit
for cumulative exposure with different lag periods across a larger number of models than was previously
evaluated with different lags, the EPA again selected 15 years as the lag period to use for the
exposure-response analyses (see Section D.3.2 of Appendix D). Sensitivity of the results to choice of
19Concerns have been raised that this approach of dropping high-dose data appears arbitrary. It should be noted, however,
that only the highest exposure group was omitted from the linear regression, and the exposure groupings were derived a priori
by the NIOSH investigators and not by the EPA in the course of its analyses.
20Equations for this weighted linear regression approach are presented in Rothman (1986) and summarized in Appendix F.
21Mean exposures for both sexes combined with a 15-year lag for the categorical exposure quartiles in Table 4-2 were 446,
2,143, 7,335, and 39,927 ppm x days. Median values were 374, 1,985, 6,755, and 26,373 ppm x days. These values are for
the full cohort, not just the risk sets.
228% greater regression coefficients for both lymphoid cancer and all lymphohematopoietic cancer.
4-14
-------
lag period is examined in Sections D.3.4 and D.3.5 of Appendix D and summarized at the end of
Section 4.1.1.3).
For the final model selection, the EPA had the following objectives:
1. Use the individual-level continuous exposure data.
2. Prioritize models that are more tuned to local behavior in the low-exposure data (e.g., spline
models) over more global models.
3. Consider the principle of parsimony.
4. Use the same model for both environmental and occupational exposures.
5. Rely less on Akaike information criterion (AIC).23
6. Weigh models based on both biological plausibility and statistical considerations.
In light of these objectives, the linear and log-linear two-piece spline models (with a 15-year lag)
were reconsidered. Two-piece spline models use the individual-level continuous exposure data,
consistent with Objective 1; their use for these data allows for better local fit to the low-exposure data
(Objective 2); the method used to preselect the knot is consistent with principles of parsimony
(Objective 3)24; and a two-piece spline model can be used for both environmental and occupational
exposures (Objective 4).
The linear and log-linear two-piece spline models with the maximum likelihood (and lowest
AIC), both had knots at 100 ppm x days. The AICs for these linear and log-linear two-piece spline
models were 461.4 and 461.8, respectively, which indicate an essentially identical global fit between the
two because the SAS procedure used for the linear models consistently reported -2 log likelihoods and
AICs about 0.4 units lower than the SAS procedure used to fit the log-linear models.25 Both model
forms also had a local maximum of the likelihood (and local minimum of the AIC) at 1,600 ppm x days
r,Thc AIC is a measure of information loss from a dose-response model that can be used to compare a specified set of
models. The AIC is defined as 2p - 21n(L), where p is the number of estimated parameters included in the model and L is the
maximized value of the likelihood function. Among a set of specified models, the model with the lowest AIC is the preferred
model.
24See SAB (2015V page 12.
2SFor the lymphoid cancer data, SAS proc NLP consistently yielded -2 log likelihoods and AICs about 0.4 units lower than
proc PHREG for the same models, including the null model, presumably for computational processing reasons. Proc NLP
was used for the linear RR models, and proc PHREG was used for the log-linear RR models.
4-15
-------
(see Figure D-14 in Appendix D), and these two models also had AICs indicating virtually identical
fits.26
Of the two-piece spline models considered, the two-piece linear spline model with the knot at
1,600 ppm x days was selected. Although this is not the model with the lowest AIC, its AIC differs by a
negligible 0.7 AIC units from the model with the lowest AIC (the two-piece linear spline model with the
knot at 100 ppm x days) and its selection is consistent with Objective 5 above to rely less on AIC and is
supported by additional considerations. One consideration is that the selected model has more exposed
cases in the exposure range of the estimate than do the two-piece spline models with knots at
100 ppm x days. With the knot at 100 ppm x days, there are no exposed cases below the knot, and the
low-exposure slope is entirely determined by the high-exposure segment and its termination at the knot.
In contrast, the low-exposure slope in the two-piece linear spline model with the knot at
1,600 ppm x days has 13 cases below the knot. Because the EPA is interested in the low-exposure slope
for unit risk estimation, the low-exposure slope provided by the selected model, which is based on more
cases in the exposure range of the estimate, is preferred. This is also consistent with Objective 2 above
to prioritize models with more local fit in the low-exposure range.
A second consideration is that the selected model reflects a better weighing of biological and
statistical considerations, consistent with Objective 6 above. The two-piece spline models with the knot
at 1,600 ppm x days have a more gradual rise in the low-exposure region and a more plausible rise at
higher exposures. This exposure-response relationship is considered more biologically realistic than the
two-piece spline models with the knot at 100 ppm x days, which more closely resemble a step-function,
which is not biologically probable for a complex, multistep process such as carcinogenicity in a variable
human population.
Neither of the two-piece spline models with the knot at 1,600 ppm x days had ap-walue <0.05;
however, both were close to 0.05 (p = 0.07 for each model), and the significant exposure-response
relationship has already been established with similar models (e.g., the two-piece spline models with the
knot at 100 ppm x days and the linear and log-linear log cumulative exposure models [see Figure 4-3]),
although the higher-exposure spline segments for the models with the knot at 1,600 ppm x days appear
to underpredict the RRs. Of the two-piece spline models with the knot at 1,600 ppm x days, the linear
spline model was preferred to the log-linear spline model because linearity is a desirable property to
have in risk assessment models. For example, linear low-dose extrapolation can occur without a
discontinuity between the model in the observable range and low-dose extrapolation from the point of
departure, and the unit risk estimate is not dependent on the risk level chosen for determination of the
26An AIC of 462.1 for the linear two-piece spline model and 462.6 for the log-linear; however, for the lymphoid cancer data,
SAS proc NLP (used for the linear RR models) consistently yielded -2 log likelihoods and AICs about 0.4 units lower than
proc PHREG (used for the log-linear RR models) for the same models, including the null model, presumably for
computational processing reasons.
4-16
-------
point of departure, at least within the exposure range of the first spline segment for a spline model. In
addition, with an overall exposure-response relationship that is supralinear, it seems contradictory to use
sublinear model forms for the increments represented by the spline pieces.
Results for the various two-piece spline models considered above are presented in Table 4-4.
The additional two-piece linear spline modeling, along with further sensitivity analyses, was also
conducted by Steenland, under a subsequent contract to the EPA, and results for the two-piece spline
models are provided in more detail in Section D.3 of Appendix D. Steenland also provided results for
linear and log-linear square-root transformation of cumulative exposure models, and these are also
included in the current assessment for completeness (see Table 4-5). These square-root transformation
models were not given much consideration because, of the single-parameter models, the
log-transformation models had better fits, as indicated by lower AICs (460.4 versus 462.8 for the
log-linear models and 460.2 versus 461.8 for the linear models; see Section D.3 of Appendix D), and
because the EPA adhered to the objective to prioritize models that have a greater ability to provide a
good local fit to the low-exposure range, such as two-piece spline models.
Table 4-4. Two-piece spline modeling results for lymphoid cancer and all
lymphohematopoietic cancer mortality in both sexes in the National Institute
for Occupational Safety and Health cohort3
Two-piece spline model
form
Knot
(ppm x days)
/>-valueb
Coefficient0 (SEd)
(per ppm x day)
Lymphoid cancer0
log-linear
100
0.047
(31 = 0.0101 (4.93 x 10"3)
(32 = -0.0101 (4.93 x lO-3)
log-linear
1,600
0.07
(31 = 4.89 x 10-4 (2.55 x 10"4)
(32 = -4.86 x 10"4 (2.6 x 10"4)
linear
100
0.046
(31 = 0.0152 (UB1 =0.0590)
(32 = -0.0152
linear
1,600
0.07
(31 = 7.58 x 10-4 (UB1 = 2.98 x 10"3)
(32 = -7.48 x 10"4
All lymphohematopoietic cancer1
log-linear
500
0.02
Low-exposure spline segment:
(31 = 2.01 x 10-3 (7.73 x 10-4)
4-17
-------
Table 4-4. Two-piece spline modeling results for lymphoid cancer and all
lymphohematopoietic cancer mortality in both sexes in the National Institute
for Occupational Safety and Health cohorta(continued)
aWithcumulative exposure inppm x days as the exposure variable, witha 15-yr lag.
'/^-values from likelihood ratio test.
Tor the two-piece spline models, for exposures below the knot, RR = 1 + ((31 x exp); for exposures above the knot,
RR = 1 + ((31 x exp + (32 x (exp - knot)).
dOr, for linear models of continuous exposures, the profile likelihood 95% (one-sided) upper bounds (UB).
eNHL, myeloma, and lymphocytic leukemia (9th revision ICD codes 200, 202, 203, 204); results based on 53 cases.
f9th revision ICD codes 200-208; results based on 74 cases.
Source: Additional analyses performed by Steenland (see Sections D.3 and D.4 of Appendix D).
SE = standard error.
Table 4-5. Exposure-response modeling results for lymphoid cancer3
mortality in both sexes in the National Institute for Occupational Safety and
Health cohort for models with square-root transformations of exposure
RR Modelb
/j-value'
Coefficient (SE)
(per ppm x day)
Linear square-root transformation model
0.053
6.14 x lO-3 (~d)
Log-linear square-root transformation model
0.08
2.83 x 10 3 (1.5 x 10"3)
aNHL, myeloma, and lymphocytic leukemia (9th revision ICD codes 200, 202, 203, 204); results based on 53 cases.
bAll with square-root transformation of cumulative exposure inppm x days, witha 15-yr lag, as the exposure
variable
7?-valucs from likelihood ratio test.
dStandard errors calculated assuming a Wald approximation are likely unrealistic for the linear RR ("ERR") models
(Langholz and Richardson. 20101. and so have not been provided.
Source: Additional analyses performed by Steenland (see SectionD.3 of AppendixD).
SE = standard error.
As the lymphoid cancer data set is the primary data set used for the derivation of unit risk
estimates for lymphohematopoietic cancers, a summary of all the models considered for modeling the
lymphoid cancer exposure-response data and the judgments made about model selection is provided in
Table 4-6. See Figure 4-3 for a visual representation of the models. See Tables 4-2, 4-3, 4-4, and 4-5
and Section D.3 of Appendix D for other information about the models. As discussed above and
summarized in Table 4-6, the two-piece linear spline model with the knot at 1,600 ppm x days was the
preferred model for the derivation of unit risk estimates of the models considered and was, thus, the
selected model.
4-18
-------
The proportional hazards assumption for the selected model (two-piece linear spline model with
knot at 1,600 ppm x days) was tested by evaluating the significance of an age-interaction term for each
spline regression coefficient, and neither interaction term was statistically significant (see Section D.3.7
of Appendix D).
Table 4-6. Models considered for modeling the exposure-response data for
lymphoid cancer mortality in both sexes in the National Institute for
Occupational Safety and Health cohort for the derivation of unit risk estimates
Model"
/>-valueb
Alt'
Comments
Two-piccc spline models
Linear spline model with
knot at 1,600 ppm x days
0.07
462.1
SELECTED. Adequate statistical and visual fit, including local
fit to low-exposure range; linear model; AIC within two units of
lowest AIC of models considered.
Linear spline model with
knot at 100 ppm x days
0.046
461.4
Good overall statistical fit and lowest AIC of two-piece spline
models, but poor local fit to the low-exposure region, with no
cases below the knot.
Log-linear spline model with
knot at 1,600 ppm x days
0.07
462.6
Linear model preferred to log-linear (see text above).
Log-linear spline model with
knot at 100 ppm x days
0.047
461.8
Good overall statistical fit and tied for lowest AIC° of two-piece
spline models, but poor local fit to the low-exposure region, with
no cases below the knot.
Linear(ERR) models (RR = 1 + |i x exposure)
Linear model
0.13
463.2
Not statistically significant overall fit and poor visual fit.
Linear model with log
cumulative exposure
0.02
460.2
Good overall statistical fit, but poor local fit to the low-exposure
region
Linear model with
square-root transformation
of cumulative exposure
0.053
461.8
Borderline statistical fit, but poor local fit to the low-exposure
region
Log-linear (Cox regression) models (RR = c|i><'xp(>sl"T)
Log-linear model (standard
Cox regression model)
0.22
464.4
Not statistically significant overall fit and poor visual fit.
Log-linear model with log
cumulative exposure
0.02
460.4
Good overall statistical fit; lowest AIC° of models considered;
low-exposure slope becomes increasingly steep as exposures
decrease, and large unit risk estimates can result; preference
given to the two-piece spline models because they have a better
ability to provide a good local fit to the low-exposure range.
Log-linear model with
square-root transformation
of cumulative exposure
0.08
462.8
Not statistically significant overall fit and poor visual fit.
4-19
-------
Table 4-6. Models considered for modeling the exposure-response data for
lymphoid cancer mortality in both sexes in the National Institute for
Occupational Safety and Health cohort for the derivation of unit risk estimates
(continued)
Model"
/>-valueb
AIC
Comments
Linear regression of categorical results
Linear regression of
categorical results, excluding
the highest exposure quartile
0.18
Not statistically significant, probably because the approach,
which is based on categorical data, has low statistical power;
preference given to models that treated exposure as a continuous
variable and that also provided reasonable representations of the
low-exposure region
"All with cumulative exposure as the exposure variable, except where noted, and with a 15-yr lag.
'/^-values from likelihood ratio test, except for linear regression of categorical results, where Wald /^-values are
reported, p < 0.05 considered "good" statistical fit; 0.05
-------
3.5000
|r 3.0000
"S
2.5000
2.0000
(u
Sc 1.5000
1.0000
10000
15000 20000 25000
cumulative exposure (ppm x days)
30000
35000
40000
eA(P*exp)
eA(P*logexp)
eA(P*sqrtexp)
• categorical
linear reg
1+P*exp
— — — l+p*logexp
1+P*sqrtexp
splinelOO
splinel600
— — — linsplinelOO
— — — Iinsplinel600
eA((3*exp): RR = e(|3 exposurel; eA((3*logexp): RR = e(|3 ln(exposure"; eA((3*sqrtexp): RR = e(|3 sqrt(exposure"; categorical: RR = e(|3'exposurel with
categorical exposures, plotted at the mean cumulative exposure; linear reg: weighted linear regression of categorical results, excluding highest
exposure group (see text); 1 + (3*exp: RR = 1 + (3 x exposure; 1 + (3*logexp: RR = 1 + (3 x ln(exposure); 1 + (3*sqrtexp: RR = 1 + (3 x
sqrt(exposure); splinel00(l,600): Two-piece log-linear spline model with knot at 100 (1,600) ppm x days (see text); linsplinel00(l,600):
Two-piece linear spline model with knot at 100 (1,600) ppm x days (see text). (Note that, with the exception of the categorical results and the
linear regression of the categorical results, the different models have different implicitly estimated baseline risks; thus, they are not strictly
comparable to each other in terms of RR values, i.e., along the v-axis. They are, however, comparable in terms of general shape.)
Source: Steenland reanalyses for males and females combined; see Appendix D (except for linear regression of categorical results, which was
done by EPA).
Figure 4-3. Exposure-response models for lymphoid cancer mortality vs. occupational cumulative exposure
(with 15-year lag).
-------
The modeling results for the selected two-piece linear spline model with the knot at
1,600 ppm x days for lymphoid cancer mortality in males and females combined were used with the
actuarial program (life-table analysis) to estimate the exposure level (ECX; "effective concentration")
and the associated 95% lower confidence limit (LECX) corresponding to an extra risk of 1% (x = 0.01).
A 1% extra risk level is commonly used for the determination of the point of departure (POD) for
low-exposure extrapolation from epidemiological cancer data (except for rare cancers); higher extra risk
levels, such as 10%, would be an upward extrapolation for these data. Thus, 1% extra risk was selected
for determination of the POD, and, consistent with the EPA's Guidelines for Carcinogen Risk
Assessment (U.S. EPA. 2005a). the LEC value corresponding to that risk level was used as the POD to
derive the cancer unit risk estimates.
Because EtO is DNA reactive and has direct mutagenic activity (see Section 3.3.3), which is one
of the cases cited by the EPA's Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a) for the
use of linear low-dose extrapolation, a linear low-exposure extrapolation was performed. (Linear
low-exposure extrapolation is also the default approach used in the absence of sufficient evidence for a
nonlinear mode of action, which is also the case for EtO [see Section 3.4]). The ECoi, LECoi, and
inhalation unit risk estimate calculated for lymphoid cancer mortality from the two-piece linear spline
model with the knot at 1,600 ppm x days are presented in Table 4-7 (the incidence results also presented
in Table 4-7 are discussed in Section 4.1.1.3 below). The resulting unit risk estimate for lymphoid
cancer mortality based on the two-piece linear spline model with the knot at 1,600 ppm x days for both
sexes using cumulative exposure with a 15-year lag is 1.99 per ppm (1.99 x 10"3 per ppb). This unit risk
estimate is 2.3 times the unit risk estimate from the most similar alternative model, the two-piece
log-linear spline model with the knot at 1,600 ppm x days (note that the ECois are similar between the
two two-piece spline models with the knot at 1,600 ppm x days, with the one from the log-linear spline
model just 8% greater than that from the linear spline model; however, the LECois are more divergent
[2.3 times]), and about 5.4 times the estimate from the linear regression of categorical results. ECoi,
LECoi, and unit risk estimates from the other models considered are presented for comparison only, to
illustrate the differences in model behavior at the low end of the exposure-response range. As discussed
above, these models were deemed unsuitable for the derivation of risks from (low) environmental
exposure levels. The log-linear log cumulative exposure model, with its marked supralinear curvature in
the lower exposure region, and the two-piece linear and log-linear spline models with the knot at
100 ppm x days, with their very steep low-exposure slopes, yield substantially higher unit risk estimates
(17.5 to 39.4 per ppm). On the other hand, the linear model and the more gradually supralinear
log-linear model with a square-root transformation, neither of which provide a statistically significant fit
to the overall data, yield substantially lower unit risk estimates (0.0314 and 0.0397 per ppm,
respectively).
4-22
-------
Table 4-7. ECoi, LECoi, and unit risk estimates for lymphoid cancer from
various modelsa'b
Model0
Mortalityd
Incidence"1'6
ECoi (ppm)
LECoi (ppm)
Unit risk
(per ppm)
ECoi (ppm)
LECoi (ppm)
Unit risk
(per ppm)
Two-piece spline models
Low-exposure
linear spline from
linear spline model
with knot at
1,600 ppm x days'
0.0198
5.03 x 10 3
1.99§
7.48 x 103
1.90 x 10 3
5.26s
Low-exposure linear
spline from linear
spline model with
knot at
100 ppm x daysh
9.87 x 10"4
2.54 x 10"4
39.4s4
3.73 x 10"4
9.61 x 10"5
104®4
Low-exposure log-
linear spline from
log-linear spline
model with knot at
1,600 ppm x daysf
0.0213
0.0114
0.877
9.92 x lO"3
5.29 x lO"3
1.89g
Low-exposure log-
linear spline from
log-linear spline
model with knot at
100 ppm x daysh
1.03 x lO-3
5.73 x 10-4
17.58.'
4.80 x 10-4
2.66 x 10-4
37.6S-1
1.incur models (KK = 1 + |i x exposure)
Linear model
1.22
0.318
0.0314
0.462
0.120
0.0833
Linear model with
log cumulative
exposure
1.55 x lO-3
Linear model with
square-root
transformation of
cumulative exposure
0.383
k
k
0.0511
k
k
4-23
-------
Table 4-7. ECoi, LECoi, and unit risk estimates for lymphoid cancer from
various modelsa'b (continued)
Model0
Mortality"1
Incidence"1'6
ECoi (ppm)
LECoi (ppm)
Unit risk (per
ppm)
ECoi (ppm)
LECoi (ppm)
Unit risk
(per ppm)
l.n;i-liik';ir models (RR = e1' )
Log-linear model
with log cumulative
exposure
6.33 x 10-3
5.24 x 10-4
19.1s.'
Log-linear model
with square-root
transformation of
cumulative exposure
0.883
0.252
0.0397
0.178
0.0509
0.196
Linear regression ol° categorical results
Linear regression of
categorical results1
0.0607
0.0272
0.368
0.0229
0.0103
0.971
aFrom lifetime continuous exposure. Unit risk = 0.01/LECoi.
bModels from Table 4-6 for whichp < 0.20 for the continuous-exposure models. This criterion resulted in the
omission of the log-linear model with untransformed exposure (standard Cox regression model; p = 0.22).
°A11 with cumulative exposure as the exposure variable, except where noted, and with a 15-yr lag.
dUsingbackground incidence and mortality rates from 2008-2012.
eIncidence estimates are presented here to facilitate comparison; they are derived in Section 4.1.1.3.
fUsing regression coefficient from low-exposure segment of two-piece linear or log-linear spline model with knot at
1,600 ppm x days (see text and Appendix D). Each of the EC0i values is at or below the value of 0.021 ppm
roughly corresponding to the knot of 1,600 ppm x days [(1,600 ppm x days) x (10 m3/20 m3) x (240 d/365 d)/(365
d/yr x 70 yr) = 0.021 ppm], and thus, appropriately in the range of the low-exposure segment.
gTo obtain unit risk estimates less than 1, convert to risk per ppb (e.g., 1.99 per ppm = 1.99 x 10"3 per ppb).
hUsing regression coefficient from low-exposure segment of two-piece linear or log-linear spline model with knot at
100 ppm x days (see text and Appendix D). Each of the EC0i values is below the value of 0.0013 ppmroughly
corresponding to the knot of 100 ppm x days [(100 ppm x days) x (10 m3/20 m3) x (240 d/365 d)/(365 d/yr x 70
yr) = 0.0013 ppm], and thus, appropriately in the range of the low-exposure segment.
'This unit risk estimate is not considered a good estimate of risks from (low) environmental exposure levels (see
text).
JEstimated exposure levels were so low (<3.6 x 10 4 ppm) that the cumulative exposures in some of the age
intervals in the lifet-able analysis were <1 ppm x day, resulting in In cumulative exposures of <0 for those intervals
and unreliable results for this model. 3.6 x 104 ppm would yield a unit risk estimate of about 28 per ppm, which
provides a lower bound on the unit risk estimate for this model.
kProfile likelihood confidence bounds were not calculated for this model.
'Regression coefficient derived from linear regression of categorical Cox regression results from Table 4-2,
dropping the highest exposure category, as described in Section4.1.1.2. Eachofthe ECoi values is appropriately
below the value of 0.090 ppm roughly corresponding to the value of about 7,000 ppm x days (see footnote f for
calculation) above which the linear regression model of the categorical results does not apply (see Figure 4-2).
Results fromthe selected model appear in bold.
4-24
-------
Converting the units, the resulting unit risk estimate of 1.99 per ppm from the selected two-piece
linear spline model with the knot at 1,600 ppm x days corresponds to a unit risk estimate of 1.09 x 10"3
per [j.g/m3 for lymphoid cancer mortality.27
The life-table analysis takes into account competing risks and the occurrence of different
cumulative exposures and different cause-specific background risks at different ages. A crude
approximation of the general approach for obtaining ECoi and LECoi estimates without the use of the
life-table component of the analysis is presented here for illustration. In this crude approach, an estimate
of the lifetime background risk of dying of lymphoid cancer is used rather than age-specific rates. For
lymphoid cancer, a life-table analysis was used to obtain this lifetime background risk estimate because
another source of such an estimate was not readily available. The resulting lifetime background
mortality risk estimate (Ro) is 1.06%. From this and eq 4-2, an estimate of the RR associated with a
1% extra risk can be calculated as RR = (0.99 xRo + 0.01)/Ro = 1.93. Then, a maximum likelihood
estimate (MLE) of the cumulative exposure associated with this RR can be calculated from the
low-exposure spline segment from the selected two-piece linear spline model as exposure = RR - 1)/Pi.
This quantity is an occupational cumulative exposure in ppm x days. To convert to environmental ppm
x years, multiply by (10 m3 breathed at work/day)/(20 rnVday) and (240 days
worked/year)/(365 days/year), as discussed earlier in Section 4.1.1.2, and then divide by 365 days/year.
Because the life-table analysis is based on actual demographic rates, this crude approximation uses an
average U. S. life expectancy of 80 years28 rather than the EPA default average lifespan of 70 years, for a
more appropriate comparison. With a 15-year lag, this means dividing the cumulative exposure by
65 years to get the continuous lifetime exposure level associated with a 1% extra risk (ECoi). The
LECoi is obtained using the same calculations but with the profile likelihood upper bound on betal in
place of betal. With betal of 7.58 x 10"4 per ppm x day and a profile likelihood upper bound on betal
of 2.98 x 10"3 per ppm x day from the selected two-piece linear spline model, these calculations yield an
ECoi of 0.0171 ppm and an LECoi of 0.00434 ppm. In comparison to the estimates presented for the
selected two-piece linear spline model in Table 4-7, this crude approach yields ECoi and LECoi
estimates that are both about 14% lower, which would correspond to a unit risk estimate about 16%
higher.
As discussed above, risk estimates based on the all lymphohematopoietic cancer results are also
derived for comparison. The same methodology presented above for the lymphoid cancer results was
used for the all lymphohematopoietic cancer risk estimates, except that because the risk estimates for
lymphoid cancers are the preferred estimates and the estimates for all lymphohematopoietic cancers are
presented merely for comparison, the background mortality rates used to calculate the estimates for all
27Conversion equation: 1 ppm= 1,830 ng/m3.
2xThc overall U.S. life expectancy in 2014 was 78.8 years (Murphy et al.. 20151.
4-25
-------
lymphohematopoietic cancer were not updated in the current assessment and no additional
exposure-response models were investigated. U.S. age-specific all-cause mortality rates for 2004 for
both sexes of all race groups combined were obtained from NCHS (Arias. 2007) and used to specify the
all-cause background mortality rates in the actuarial program. Age-specific background mortality rates
for all subcategories of lymphohematopoietic cancer (ICD-10 C81-C96) for the year 2004 were obtained
from the NCHS Data Warehouse website (https://www.cdc.gov/nchs/nvss/mortalitv tables.htm).
The results of Steenland's reanalyses using the Cox regression models presented in the Steenland
et al. (2004) paper with data for males and females combined are presented in Table 4-2. As for
lymphoid cancer and for all hematopoietic cancer in males presented in the Steenland et al. (2004)
paper, the only statistically significant Cox regression model was for log cumulative exposure with a
15-year lag (p = 0.01). The cumulative exposure model did not provide an adequate fit to the data and is
not considered further here (p = 0.35).
Because of the problems with the supralinear log cumulative exposure model which are
discussed for the lymphoid cancers above, the EPA again investigated the use of a two-piece log-linear
spline model to attempt to address the supralinearity of the data while avoiding the extreme
low-exposure curvature obtained with the log cumulative exposure model. For the all
lymphohematopoietic cancer mortality data, the range examined for knot selection was from 0 to
7,000 ppm x days, and the largest model likelihood was obtained with the knot at 500 ppm x days (see
Figure D-19 of Appendix D). See Table 4-4 and Section D.4 of Appendix D for parameter estimates
and fit statistics for the two-piece spline model.29 The linear regression model of the categorical results
was also used to derive a cancer unit risk estimate for this data set.
For the weighted linear regression, the results from the Cox regression model with categorical
cumulative exposure and a 15-year lag (see Table 4-2), excluding the highest exposure group, and the
approach discussed above for lymphoid cancer mortality were used. See Table 4-3 for the results
obtained from the weighted linear regression and Figure 4-4 for a graphical presentation of the resulting
linear regression model.
29When the 2014 draft assessment was largely complete (U.S. EPA, 2014a,b), a few linear RR models were attempted, using
the then just-published approach of Langholz and Richardson (2010) to model the individual data with cumulative exposure
as a continuous variable; however, these linear models did not alleviate the problems of the corresponding log-linear RR
models (see Figure D-21 in Appendix D) and have not been pursued further for the all lymphohematopoietic cancer data.
4-26
-------
4.0
3.5
3.0
2.5
2.0
0
5000
10000
15000
20000
25000
30000
35000
40000
cumulative exposure (pprrfdays)
eA(P*exp)
— — eA(P*logexp)
• categorical
¦ 2-piece spline
linear
eA(|3*exp): RR = c1' exposure); eA(|3*logexp): RR = c1' h(exposure)); categorical: RR = c1' exposure) with categorical exposures, plotted
at the mean cumulative exposure; linear: weighted linear regression of categorical results, excluding highest exposure group (see
text); two-piece spline: two-piece log-linear spline model with knot at 500 ppm * days (see text). [Note that, with the exception
of the categorical results and the linear regression of the categorical results, the different models have different implicitly
estimated baseline risks; thus, they are not strictly comparable to each other in terms of RR values (i.e., along the v-axis). They
are, however, comparable in terms of general shape.]
Source: Steenland reanalyses for males and females combined; see Appendix D (except for linear regression of the categorical
results, which was done by the EPA).
Figure 4-4. RR estimate for all lymphohematopoietic cancer vs. occupational cumulative exposure (with 15-year
lag)-
-------
The ECoi, LECoi, and inhalation unit risk estimates calculated for all lymphohematopoietic
cancer mortality from the various models examined are presented in Table 4-8 (the incidence results also
presented in Table 4-8 are discussed in Section 4.1.1.3 below). The unit risk estimate for all
lymphohematopoietic cancer mortality based on the linear regression of the categorical results for both
sexes using cumulative exposure with a 15-year lag is 0.680 per ppm and that based on the log-linear
spline model (knot at 500 ppm x days) is 4.33 per ppm. For comparing with the lymphoid cancer
estimates, the most comparable model is the linear regression of categorical results—the unit risk
estimate for all lymphohematopoietic cancer mortality from the linear regression model of the
categorical results is 85% higher than that for lymphoid cancer mortality from the linear regression
model (see Table 4-7). The unit risk estimate for all lymphohematopoietic cancer mortality from the
log-linear spline model (knot at 500 ppm x days) is 120% higher than (i.e., 2.2 times) that for lymphoid
cancer mortality from the selected linear spline model (knot at 1,600 ppm x days) (see Table 4-7).
4-28
-------
Table 4-8. ECoi, LECoi, and unit risk estimates for all lymphohematopoietic
cancer3
Mortality0
Incidence0'"1
Modelb
ECoi (ppm)
LECoi (ppm)
Unit risk
(per ppm)
ECoi (ppm)
LECoi (ppm)
Unit risk
(per ppm)
Log cumulative
exposure, 15-yr
lag
1.40 x 10"3
__e
__e
__e
Low-exposure
log-linear spline;
cumulative
3.77 x 10-3
2.31 x 10-3
4.33b
2.16 x 10-3
1.32 x 10-3
7.58s
exposure, 15-yr
lag, knot at
500 ppm x daysf
Linear regression
of categorical
results,
cumulative
0.0283
0.0147
0.680
0.0144
7.46 x 10-3
1.34^
exposure, 15-yr
lag11
aFrom lifetime continuous exposure. Unit risk = 0.01/LECoi.
bFrom Steenland's analyses for males and females combined (see Appendix D), log-linear Cox regression models.
°Usingbackground incidence and mortality rates from 2004.
dIncidence estimates presented here to facilitate comparison; they are derived in Section 4.1.1.3.
"Estimated exposure levels were so low (<3.6 x 10 4 ppm) that the cumulative exposures in some of the age
intervals in the life-table analysis were <1 ppm x day, resulting in In cumulative exposures of <0 for those intervals
and unreliable results for this model. 3.6 x 10 4 ppmwould yield a unit risk estimate of about 28 per ppm, which
provides a lower bound on the unit risk estimate for this model.
fUsing regression coefficient from low-exposure segment of two-piece log-linear spline model with knot at
500 ppm x days; see text and Appendix D. Each of the ECoi values is below the value of 0.0064 ppm roughly
corresponding to the knot of 500 ppm x days [(500 ppm x days) x (10 m3/20 m3) x (240 d/365 d)/(365
d/yr x 70 yr) = 0.0064 ppm] and, thus, appropriately in the range of the low-exposure segment.
sFor unit risk estimates below 1, convert to risk per ppb (e.g., 1.34 per ppm = 1.34 x 10 3 per ppb).
degression coefficient derived from linear regression of categorical Cox regression results from Table 4-2. Each
of the ECoi values is appropriately below the value of 0.064 ppm roughly corresponding to the value of about
5,000 ppm x days (see footnote d for calculation) above which the linear regression model of the categorical results
does not apply (see Figure 4-4).
yr = year.
4.1.1.3. Prediction of Lifetime Extra Risk of Lymphohematopoietic Cancer Incidence
EPA cancer risk estimates are typically derived to represent an upper bound on increased risk of
cancer incidence, as from laboratory animal incidence data. Cancer data from epidemiologic studies are
commonly mortality data, as is the case in the Steenland et al. (2004) study. For tumor sites with low
survival rates, mortality-based estimates are reasonable approximations of cancer incidence risk;
4-29
-------
however, for many lymphohematopoietic cancers, the survival rate is substantial, and incidence risk
estimates are preferred by the EPA (U.S. EPA. 2005a).
Therefore, another calculation was done using the same regression coefficients presented above
(see Section 4.1.1.2), but with age-specific lymphoid cancer incidence rates for the relevant
subcategories of lymphohematopoietic cancer (NHL, myeloma, and lymphocytic leukemia) for
2008-2012 from SEER [Howlader et al. (2014); Tables 18.7, 19.7, 13.12: both sexes, all races] in place
of the lymphoid cancer mortality rates in the life-table analysis. SEER collects good-quality cancer
incidence data from a variety of geographical areas in the United States. The incidence data used here
are from "SEER 18," a registry of eighteen states, regions, and cities covering about 28% of the U.S.
population.
The incidence risk calculation assumes that (1) lymphoid cancer incidence and mortality have
the same exposure-response relationship for the relative rate of effect from EtO exposure and (2) the
incidence data are for first occurrences of primary lymphoid cancer or that relapses and secondary
lymphoid cancers provide a negligible contribution. The latter assumption is probably sound; the former
assumption is potentially more problematic. Because various lymphoid cancer subtypes with different
survival rates are included in the categorization of lymphoid cancers, a bias could occur if the
EtO-associated relative rates of the subtypes differ, resulting in either an underestimation or
overestimation of the extra risk for lymphoid cancer incidence.30 Potential concern that the incidence
risk estimates might be overestimated would come primarily from the inclusion of multiple myeloma
because that subtype has the lowest incidence mortality ratios. Therefore, if that subtype were driving
the increased mortality observed for the lymphoid cancer grouping, then including the incidence rates
for the other subtypes, which have higher incidence mortality ratios, might inflate the incidence risk
estimates. Multiple myelomas, however, constitute only 25% of the lymphoid cancer cases in the
cohort, and there is no evidence that multiple myeloma is driving the EtO-induced excess in lymphoid
cancer mortality.31 Thus, using the total lymphoid cancer incidence rates is not expected to result in an
overestimation of the incidence risk estimates; if anything, the incidence risks would likely be diluted
with the inclusion of the multiple myeloma rates. The incidence risk calculation also relies on the fact
30Sielkenand Valdez-Flores (2009) rejected the assumption that lymphohematopoietic cancer incidence and mortality have
the same exposure-response relationship, reporting that, except at high exposure levels, the exposure-response data in the
male workers in the NIOSH cohort are consistent with a decreased survival time and suggesting that this could explain the
observed increases in mortality. However, they do not establish that this is what is occurring, and the mechanistic data
support an exposure-related increase in incident cancers. See Appendix A.2.20 for a more detailed discussion of this issue.
31 According to data from SEER (www.sccr.canccr.aov). 25% is below the proportion of multiple myeloma deaths one would
expect based on age-adjusted U.S. background mortality rates of multiple myeloma, NHL, and chronic lymphocytic
leukemia, and these three subtypes have the same pattern for mortality rates increasing as a function of age mostly above age
50, so the comparison with lifetime background rates is reasonable. In addition, the low proportion of multiple myeloma
deaths in the lymphoid cancer sub grouping cannot be attributed to an underrepresentation of blacks, who have incidence rates
of multiple myeloma more than twice those of whites Chttp: //seer, cancer, gov/statfac ts/html/mul mv. html). in the cohort
because blacks comprise 16% of the cohort versus 12.3% in the U.S. population
4-30
-------
that the lymphoid cancer incidence rates (more specifically, the differential rates obtained by subtracting
the mortality rates from the incidence rates) are small when compared with the all-cause mortality
rates.32
The resulting ECoi and LECoi estimates for lymphoid cancer incidence from the various models
examined are presented in Table 4-7. The unit risk estimate for lymphoid cancer incidence from the
selected two-piece linear spline model with the knot at 1,600 ppm x days is 5.26 per ppm. This unit risk
estimate is about three times the unit risk estimate from the most similar alternative model, the
two-piece log-linear spline model with the knot at 1,600 ppm x days and about 5.4 times the estimate
from the linear regression of categorical results. ECoi, LECoi, and unit risk estimates from the other
models considered are presented for comparison only, to illustrate the differences in model behavior at
the low end of the exposure-response range. As discussed above, these models were deemed unsuitable
for the derivation of risks from (low) environmental exposure levels. The highly supralinear two-piece
linear and log-linear spline models with the knot at 100 ppm x days yield unit risk estimates about 20
and 7 times the preferred estimate, respectively. On the other hand, the linear model and the more
gradually supralinear log-linear model with a square-root transformation, neither of which provide a
statistically significant fit to the overall data, yield unit risk estimates 1.6% and 3.7% of the preferred
estimate, respectively.
Overall, as discussed above, the preferred estimate for the unit risk for lymphoid cancer is the
estimate of 5.26 per ppm (2.87 x 10-3 per (j,g/m3) derived, using incidence rates for the cause-specific
background rates, from the two-piece linear spline model with the knot at 1,600 ppm x days. This
incidence unit risk estimate is about 2.6 times the mortality-based estimate from the same model.
Sensitivity analyses were conducted to investigate the influence of lag period, knot selection, and
upper-bound estimation approach on the unit risk estimates from the selected two-piece linear spline
model. The sensitivity analyses are detailed in Sections D.3.5, D.3.6, and D.3.8 of Appendix D. In
brief, for the two-piece linear spline model with the knot at 1,600 ppm x days, the unit risk estimates for
different lag periods (0, 5, 10, and 20 years) ranged from about 48% less than (10-year lag) to about
190%) greater than (i.e., 2.9 times) (no lag) the estimate for the selected model (15-year lag). Varying
the knot by 1,000 ppm x days, especially in the lower direction, also changes the unit risk estimate a
notable amount—the unit risk estimate with the knot at 600 ppm x days was about 210% greater than
'-Siclkcn and Valdez-Flores (2009) suggested that the methods used by the EPA to calculate incidence risk estimates in the
life-table analysis are inappropriate; however, as explained in more detail in Appendix A.2.20, the EPA disagrees. For the
situation where the cause-specific incidence rates are small compared to the all-cause mortality rates, as with lymphoid
cancer, there is no problem, as Sielken and Valdez-Flores (2009) themselves demonstrated. For the situation where the
cause-specific incidence rates are not negligible compared to the all-cause mortality rates, as with breast cancer, an
adjustment was made in the analysis to remove those with incident cases fiomthe population at risk [i.e., those "surviving"
each interval without a diagnosis of breast cancer (see Section 4.1.2.3)]. See Appendix A.2.20 for a more detailed discussion
of this issue.
4-31
-------
(i.e., 3.1 times), and unit risk estimate with the knot at 2,600 ppm x days was about 46% less than, the
unit risk estimate for the selected model (with the knot at 1,600 ppm x days). The unit risk estimate
calculated using a Wald approach was 40% lower than the preferred estimate, which relied on a profile
likelihood approach to estimate the upper bound on the regression coefficient.
As discussed in Section 4.1.1.2, risk estimates based on the results of Steenland's reanalyses of
the all lymphohematopoietic cancer data (see Appendix D and Table 4-2) are also derived for
comparison. The same methodology presented above for the lymphoid cancer incidence results was
used for the all lymphohematopoietic cancer incidence risk estimates, and the same assumptions apply.
Age-specific SEER incidence rates for all lymphohematopoietic cancer for the years 2000-2004 were
used [Tables XIX, IX, XVIII, and XIII in (Ries et al.. 2007); both sexes, all races]. The ECoi, LECoi,
and unit risk estimates for all lymphohematopoietic cancer incidence from the different all
lymphohematopoietic cancer mortality models examined are presented in Table 4-8. The unit risk
estimate for all lymphohematopoietic cancer incidence based on the linear regression of the categorical
results for both sexes using cumulative exposure with a 15-year lag is 1.34 per ppm and that based on
the log-linear spline model (knot at 500 ppm x days) is 7.58 per ppm. For comparing with the lymphoid
cancer estimates, the most comparable model is the linear regression of categorical results—the unit risk
estimate for all lymphohematopoietic cancer incidence from the linear regression model of the
categorical results is 38% higher than the unit risk estimate for lymphoid cancer incidence from the
linear regression model (see Table 4-7). The unit risk estimate for all lymphohematopoietic cancer
incidence from the log-linear spline model (knot at 500 ppm x days) is 44% higher than the unit risk
estimate for lymphoid cancer incidence from the selected linear spline model (knot at 1,600 ppm x days)
(see Table 4-7).
4.1.2. Risk Estimates for Breast Cancer
4.1.2.1. Breast Cancer Results from the NIOSH Study
The Steenland et al. (2004) study discussed above in Section 4.1.1.1 also presented results from
exposure-response analyses for breast cancer mortality in female workers. In addition, Steenland et al.
(2003) presented results of a breast cancer incidence study of a subcohort of the female workers from
the NIOSH cohort. In addition to the analyses in the Steenland et al. (2003) and Steenland et al. (2004)
papers, Steenland did subsequent analyses of the breast cancer mortality data set for the EPA and
Steenland and Deddens did additional analyses of the breast cancer incidence data set; these are
discussed below and reported in Sections D.l and D.2 of Appendix D, respectively.
Unit risk estimates are developed for both breast cancer mortality (see Section 4.1.2.2) and breast
cancer incidence (see Section 4.1.2.3). The incidence estimates are strongly preferred over the mortality
estimates for a number of reasons. First, unit risk estimates are intended to reflect cancer incidence
rather than mortality; thus, incidence estimates are generally preferred over mortality estimates. Second,
4-32
-------
in the case of these specific estimates, the incidence estimates based on the subcohort of workers with
interviews are preferred because they are based on a larger number of cases than the mortality estimates,
and data on potential breast cancer risk factors are taken into account. Section 4.1.2.2 containing the
derivation of the mortality estimates has been retained in this final assessment for completeness and for
comparison with the incidence estimates. However, because of the preference for the incidence data, no
further modeling of the mortality data was conducted after the 2014 external review draft; thus, no
additional models of the continuous exposure data were explored.
4.1.2.2. Prediction of Lifetime Extra Risk of Breast Cancer Mortality
Results from the Cox regression models presented by Steenland et al. (2004), with some
reanalyses reported by Steenland in Appendix D (see Section D.2), are summarized in Table 4-9. These
models were considered for the derivation of unit risk estimates for breast cancer mortality in females
from continuous environmental exposure to EtO, applying the methodologies described in
Section 4.1.1.2.
Table 4-9. Cox regression results for breast cancer mortality in females in
the National Institute for Occupational Safety and Health cohort,3 for models
presented in Steenland et al. (2004)
Exposure variableb
/j-value'
Coefficient (SE)
(per ppm x day)
ORs by category11 (95% CI)
Cumulative exposure, 20-yr
lag?
0.10
0.00001.22 x 10-5
(6.41 x 10"6)
Log cumulative exposure,
20-yr lagf
0.02
0.084 (0.035)
Categorical cumulative
exposure, 20-yr lagf
0.09
1.00
1.76 (0.91-3.43)
1.77 (0.88-3.56)
1.97 (0.94-4.06)
3.13 (1.42-6.92)
aBased on 103 breast cancer (ICD-9 174,175) deaths.
bCumulative exposure is inppm x days.
7?-valucs based on likelihood ratio test.
dExposure categories are 0; >0-646; 647-2,779; 2,780-12,321; >12,322 ppm x days.
eFrom reanalyses in SectionD.2 of Appendix D; Steenland et al. (20041 reported the Cox regression results for
cumulative exposure with no lag.
'From Table 8 of Steenland etal. (20041.
yr = year.
U.S. age-specific all-cause mortality rates for 2000 for females of all race groups combined
(Minino et al., 2002) were used to specify the all-cause background mortality rates in the actuarial
program (life-table analysis). The National Center for Health Statistics 1997-2001 cause-specific
4-33
-------
background mortality rates for invasive breast cancers in females were obtained from a SEER report
(Ries et al.. 2004). The risks were computed up to age 85 for continuous exposures to EtO, conversions
were made between occupational EtO exposures and continuous environmental exposures, and 95%
UCLs were calculated for the relative rates, as described above.
The only statistically significant Cox regression model presented by Steenland et al. (2004) for
breast cancer mortality in females was for log cumulative exposure with a 20-year lag (p = 0.01).
However, as for the lymphohematopoietic cancers in Section 4.1.1, using the log cumulative exposure
model to estimate the risks from low environmental exposures is problematic because this model is
highly supralinear and results are unstable for low exposures (i.e., small changes in exposure correspond
to large changes in risk; see Figure 4-5). The cumulative exposure model, which is typically used and
which is stable at low exposures, was nearly statistically significant (p = 0.06 with a 20-year lag; see
Section D.2 of Appendix D) in terms of the global fit to the data; however, the Cox regression model
with cumulative exposure is sublinear, does not reflect the apparent supralinearity of the breast cancer
mortality data, and provides a poor local fit to the lower-exposure region (see Figure 4-5).
In a 2006 external review draft of this assessment (U.S. EPA. 2006a). which relied on the
original published results of Steenland et al. (2004). the EPA proposed that the best way to reflect the
exposure-response relationship in the lower exposure region, which is the region of interest for
low-exposure extrapolation, was to do a weighted linear regression of the results from the Cox
regression model with categorical cumulative exposure and a 20-year lag. In addition, the highest
exposure group was not included in the regression to alleviate some of the "plateauing" in the
exposure-response relationship at higher exposure levels and to provide a better fit to the lower exposure
data. Linear modeling of categorical epidemiologic data and elimination of the highest exposure
group(s) in certain circumstances to obtain a better fit of low-exposure data are both standard techniques
used in EPA dose-response assessments (U.S. EPA. 2005a).
For the weighted linear regression, the results from the Cox regression model with categorical
cumulative exposure (and a 20-year lag) presented in Table 4-9 were used, excluding the highest
exposure group, and the approach discussed above for the lymphoid cancers (see Section 4.1.1.2).33
Mean and median exposures for the cumulative exposure groups were provided by Steenland (see
Appendix D).34 See Table 4-10 for the results obtained from the weighted linear regression of the
categorical results and mean exposures and Figure 4-5 for a depiction of the resulting linear regression
model.
33Equations for this weighted linear regression approach are presented in Rothnian (1986) and summarized in Appendix F.
34Mean exposures for females with a 20-year lag for the categorical exposure quartiles in Table 8 of Steenland etal. (20041
were 276, 1,453, 5,869, and 26,391 ppm x days. Median values were 250, 1,340, 5,300, and 26,676 ppm x days. These
values are for the risk sets but should provide a good approximation to the full cohort values.
4-34
-------
3.5
3
2.5
2
1.5
1
0
5000
10000
15000
20000
25000
30000
35000
cumulative exposure (ppm*days)
eA(B*logexp)
linear
• categorical
eA(B*exp)
spline700
eA(B*exp): Cox regression results for RR = e(P exP°sure); eA(B*logexp): Cox regression results for RR = e(P ["(exposure));
categorical: Cox regression results for RR = e(P exP°sure) with categorical exposures, plotted at the mean cumulative
exposure; linear: weighted linear regression of categorical results, excluding highest exposure group (see text);
spline700: two-piece log-linear spline model with knot at 700 ppm x days (see text). [Note that, with the exception of
the categorical results and the linear regression of the categorical results, the different models have different implicitly
estimated baseline risks; thus, they are not strictly comparable to each other in terms of RR values (i.e., along the
j'-axis). They are, however, comparable in terms of general shape.]
Source: Steenland reanalyses with 20-year lag; see Section D.2 of Appendix D (except for linear regression of the
categorical results, which was done by the EPA).
Figure 4-5. RR estimate for breast cancer mortality vs. occupational cumulative exposure (with 20-year lag).
-------
However, as in Section 4.1.1.2 for the similarly supralinear lymphohematopoietic cancer data,
for the subsequent draft assessment, the EPA pursued modeling the individual exposure data as an
alternative to modeling the published grouped data (U.S. EPA. 2014a. b). Consequently, a two-piece
spline model was considered as an alternative way to address the supralinearity of the full data set (while
avoiding the extreme low-exposure curvature obtained with the log cumulative exposure model).
Steenland was commissioned to do the spline analyses using the full data set with cumulative exposure
as a continuous variable. His findings are reported in Section D.2 of Appendix D, and the results for the
breast cancer mortality analyses are summarized below. (For this final assessment, the model selection
for breast cancer mortality was reconsidered in light of the objectives used for model selection for
lymphoid cancer [see Section 4.1.1.2] and breast cancer incidence [see Section 4.1.2.3]; however,
because the breast cancer mortality results are presented solely for comparison with the preferred
incidence results, no additional modeling of the breast cancer mortality data was pursued and the unit
risk estimates presented below were not updated to reflect more recent background mortality rates.)
For the two-piece log-linear spline modeling approach, as described in Section 4.1.1.2 and
discussed more fully in Appendix D, the breast cancer mortality exposure-response data (cumulative
exposure, with a 20-year lag) were fitted such that log RR is a function of two lines joined at a knot.
The shape of the two-piece log-linear spline model, in particular the slope in the low-exposure region,
depends on the location of the knot. Knot selection was made by trying different knots over a
reasonable range and choosing the one that resulted in the largest model likelihood. For the breast
cancer mortality data, the range examined for knot selection was from 0 to 25,000 ppm x days, using
increments of 100 ppm x days to 7,000 ppm x days and increments of 1,000 ppm x days above
7,000 ppm x days. The largest model likelihood was observed with the knot at 700 ppm x days,
although, as noted above, the model likelihood did not change much across the various trial knots (see
Figure D-9 of Appendix D). Parameter estimates for this model are presented in Table 4-10. The
p-value of the two-piece spline model exceeded 0.05, although minimally (p = 0.067). This two-piece
spline model was selected as the preferred model for breast cancer mortality primarily because it uses
the individual-level exposure data and this model form is more tuned to local behavior than the other
model forms considered.
The two-piece spline model with the knot at 700 ppm x days and the actuarial program (life-table
analysis) were used to estimate the exposure level (ECX) and the associated 95% lower confidence limit
(LECX) corresponding to an extra risk of 1% (x = 0.01). As discussed in Section 4.1.1.2, a 1% extra risk
level is a more reasonable response level for defining the POD for these epidemiologic data than 10%.
4-36
-------
Table 4-10. Exposure-response modeling results for breast cancer mortality
in females in the National Institute for Occupational Safety and Health cohort
for models not presented by Steenland et al. (2004)
Model"
/>-valueb
Coefficient (SE)
(per ppm x day)
Two-piece log-linear spline with maximum
likelihood (knot at 700 ppm x days)
0.067
low-exposure spline segment:
Pi =6.88 x lO-4 (4.17 x 10-4)
Linear regression of categorical results, excluding
the highest exposure quartile
0.09
2.01 x 10"4 (1.20 x 10-4)
"All with cumulative exposure in ppm x days as the exposure variable and with a 20-yr lag; based on 103 breast
cancer deaths.
'/^-values from likelihood ratio test, except for linear regression of categorical results, where Wald p-valuc is
reported.
Source: Additional analyses performed by Steenland (see SectionD.2 of Appendix D), except for the linear
regression of the categorical results, which was performed by the EPA.
Because EtO is DNA-reactive and has direct mutagenic activity (see Section 3.3.3), which is one
of the cases cited by the EPA's Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a) for the
use of linear low-dose extrapolation, a linear low-exposure extrapolation was performed. (Linear
low-exposure extrapolation is also the default approach used in the absence of sufficient evidence for a
nonlinear mode of action, which is also the case for EtO [see Section 3.4].) The ECoi, LECoi, and
inhalation unit risk estimates calculated for breast cancer mortality from the two-piece spline model with
the knot at 700 ppm x days are presented in Table 4-11, along with estimates from some of the other
models for comparison. The resulting unit risk estimate for breast cancer mortality based on the
two-piece spline model with the knot at 700 ppm x days using cumulative exposure with a 20-year lag is
2.12 per ppm. This unit risk estimate is about four times the unit risk estimate from the linear regression
of the categorical results. The standard Cox regression cumulative exposure model, with its extreme
sublinearity in the lower exposure region, yields a substantially lower unit risk estimate (<2% of that
from the selected model), while the log cumulative exposure Cox regression model, with its extreme
supralinearity and steep slope in the lower exposure region, would have resulted in a much higher unit
risk estimate, had it been possible to reliably calculate one. Converting the units, the unit risk estimate
of 2.12 per ppm for breast cancer mortality from the two-piece spline model with the knot at
700 ppm x days corresponds to a unit risk estimate of 1.16 x 10"3 per [j,g/m3.
This unit risk estimate for breast cancer mortality is slightly higher than the unit risk estimate
derived for breast cancer incidence below (see Section 4.1.2.3), which is contrary to expectations.
Confidence is higher in the breast cancer incidence estimate, suggesting that this mortality estimate may
be an overestimate of the unit risk for breast cancer mortality.
4-37
-------
Table 4-11. ECoi, LECoi, and unit risk estimates for breast cancer mortality
in females3
Model
ECoi (ppm)
LECoi (ppm)
Unit risk (per ppm)
Log cumulative exposure,
20-yr lag15
1.12 x 10-3
__C
c
Cumulative exposure, 20-yr
lag1
0.530
0.285
0.035P
Low-exposure log-linear
spline, cumulative exposure
with knot at 700 ppm x days,
20-yr lagf
9.41 x 10-3
4.71 x 10-3
2.12
Categorical cumulative
exposure, 20-yr lag®
0.0387
0.0195
0.513
aFrom lifetime continuous exposure. Unit risk = 0.01/LECoi.
bFrom Table 8 of Steenland et al. (2004). Cox regression model.
Estimated exposure levels were so low (<3. 6 x 10-4 ppm) that the cumulative exposures in some of the age
intervals in the life-table analysis were <1 ppm x day, resulting in In cumulative exposures of <0 for those intervals
and unreliable results for this model. 3.6 x 104 ppm would yield a unit risk estimate of about 28 per ppm, which
provides a lower bound on the unit risk estimate for this model.
dFrom Steenland's reanalyses (see Table D-22 of Appendix D), Cox regression model.
eThis unit risk estimate is not considered a good estimate of risks from (low) environmental exposure levels (see
text).
Trom low-exposure segment of two-piece log-linear spline model with largest model likelihood and a knot at
700 ppm x days; see text and Table D-21 of Appendix D. The ECoi value is below the value of 0.009 ppm roughly
corresponding to the knot of 700 ppm x days [(700 ppm x days) x (10 m3/20 in') x (240 d/365 d)/(365
d/yr x 70 yr) = 0.0013 ppm] and, thus, appropriately in the range of the low-exposure segment.
"Regression coefficient derived from linear regression of categorical Cox regression results from Table 8 of
Steenland et al. (2004). as described in this section The ECoi value is appropriately below the value of 0.064 ppm
roughly corresponding to the value of about 5,000 ppm x days (see footnote f for calculation) above which the
linear regression model of the categorical results does not apply (see Figure 4-5).
yr = year.
4.1.2.3. Prediction of Lifetime Extra Risk of Breast Cancer Incidence
As discussed in Section 4.1.1.3, risk estimates for cancer incidence are preferred to estimates for
cancer mortality, especially for cancer types with relatively high survival rates, such as breast cancer. In
the case of female breast cancer in the NIOSH cohort, there is a corresponding incidence study
(Steenland et al., 2003) with exposure-response results for breast cancer incidence, so one can estimate
cancer incidence risks directly rather than estimate them from mortality data. The incidence study used
a (sub)cohort of 7,576 (76%) of the female workers from the original cohort. Cohort eligibility for the
incidence study was restricted to the female workers who had been employed at 1 of the 13 plants with
exposure estimates for at least 1 year, owing to cost considerations and the greater difficulties in locating
workers with short-term employment. Interviews were sought from all the women in the incidence
study cohort or their next-of-kin (18% of the cohort had died). Completed interviews were obtained for
4-38
-------
5,139 (68%) of the 7,576 women in the cohort. The investigators also attempted to acquire breast cancer
incidence data for the cohort from cancer registries (available for 9 of the 11 states in which the plants
were located) and death certificates; thus, results are presented for both the full cohort (n = 7,576) and
the subcohort of women with interviews (// = 5,139). For additional details and discussion of the
Steenland et al. (2003) study, see Section A.2.16 of Appendix A.
Steenland et al. (2003) identified 319 incident cases of breast cancer in the cohort through 1998.
Interview (questionnaire) data were available for 73% (233 cases). Six percent of the breast cancers
were carcinoma in situ (20 cases). Steenland et al. (2003) performed internal exposure-response
analyses similar to those described in their 2004 paper and in Section 4.1.1.1 above. Controls for each
case were selected from the cohort members without breast cancer at the age of diagnosis of the case.
Cases and controls were matched on race. Of the potential confounders evaluated for those with
interviews, only parity and breast cancer in a first-degree relative were important predictors of breast
cancer, and only these variables were included in the final models for the subcohort analyses. In situ
cases were included with invasive breast cancer cases in the analyses; however, the in situ cases
represent just 6% of the total, and excluding them reportedly did not greatly affect the results.
From the Steenland et al. (2003) internal analyses (Cox regression) using the full cohort, the
best-fitting model with exposure as a continuous variable was for (natural) log cumulative exposure,
lagged 15 years (p = 0.05). Duration of exposure, lagged 15 years, provided a slightly better fitting
model (Steenland et al.. 2003). Models using maximum or average exposure did not fit as well. In
addition, use of a threshold model did not provide a statistically significant improvement in fit. For
internal analyses using the subcohort with interviews, the cumulative exposure and log cumulative
exposure models, both lagged 15 years, and the log cumulative exposure model with no lag all fit almost
equally well, and the duration of exposure (also lagged 15 years) model fit slightly better (Steenland et
al.. 2003). Results of the Cox regression analyses for the cumulative and log cumulative exposure
models, with 15-year lags, are shown in Table 4-12. Cumulative exposure is the preferred basis for
cancer unit risk estimates. The models using duration of exposure are less useful for estimating
exposure-related risks, duration of exposure and cumulative exposure are correlated, and the fits for the
duration models are only marginally better than those with cumulative exposure. In addition,
cumulative exposure with no lag was considered less biologically realistic than cumulative exposure
with the 15-year lag because some lag period would be expected for the development of breast cancer.
For this final assessment, the EPA revisited the issue of lag selection for the breast cancer incidence
data. After considering model fit for cumulative exposure with different lag periods across a larger
number of models than was previously evaluated with different lags, the EPA again selected 15 years as
the lag period to use for the exposure-response analyses (see Section D.1.2 of Appendix D).
4-39
-------
Table 4-12. Cox regression results for breast cancer incidence in females
from the National Institute for Occupational Safety and Health cohort, for the
models presented by Steenland et al. (2003)a'b
Cohort
Exposure variable0
Coefficient (SE)
(per ppm x day),
/j-valucd
ORs by category6 (95% CI)
Full incidence study
cohort
n = 7,576
319 cases
Cumulative exposure, 15-yr
lag
5.4 x 10-6
(3.5 x 10-6),
p = Q.U
Log cumulative exposure,
15-yrlag
0.037 (0.019),
p = 0.05
Categorical cumulative
exposure, 15-yr lag
1.00, 1.07 (0.72-1.59), 1.00
(0.67-1.50), 1.24 (0.85-1.90), 1.17
(0.78-1.78), 1.74(1.16-2.65)
Subcohort with
interviews
n = 5,139
233 cases
Cumulative exposure, 15-yr
lag
9.5 x 10"6
(4.1 x 10"6),
p = 0.02
Log cumulative exposure,
15-yrlag
0.050 (0.023),
p = 0.03
Categorical cumulative
exposure, 15-yr lag
1.00, 1.06 (0.66-1.71), 0.99
(0.61-1.60), 1.24 (0.76-2.00), 1.42
(0.88-2.29), 1.87 (1.12-3.10)
aInvasive breast cancer (ICD-9 174) and carcinoma in situ (ICD-9 233.0).
bCases and controls matched on age and race (white/nonwhite). Full cohort models include cumulative exposure
and categorical variable for yr of birth (quartiles). Subcohort models include cumulative exposure, categorical
variables for yr of birth (quartiles), breast cancer in first-degree relative, and parity.
Cumulative exposure is in ppm x days.
'/^-values for exposure variable from Wald test, as reported by Steenland et al. C2003).
eExposure categories are 0, >0-647, 647-2,026, 2,026-4,919, 4,919-14,620, >14,620 ppm x days.
'/?-valuc for the addition of the categorical exposure variables = 0.11 (email dated 5 March 2010 from Kyle
Steenland, Emory University, to Jennifer Jinot, EPA).
Source: Tables 4 and 5 of Steenland et al. (2003).
yr = year.
Sensitivity of the results to choice of lag period is examined in Sections D.1.5 and D.1.6 of Appendix D
and summarized at the end of this section (see Section 4.1.2.3).
Although risk estimates based on the full cohort results are calculated for comparison, the
preferred estimates are those based on the subcohort with interviews because the subcohort should have
more complete case ascertainment and has additional information available on potential breast cancer
confounders.
For the actuarial program (life-table analysis), U.S. age-specific all-cause mortality rates for
2008-2012 for females of all race groups combined (CDC. 2015) were used to specify the all-cause
4-40
-------
background mortality rates. Because breast cancer incidence rates (more specifically, the differential
rates obtained by subtracting the mortality rates from the incidence rates) are not negligible compared to
all-cause mortality rates, the all-cause mortality rates in the life-table analysis (used to calculate the
population at risk) were adjusted to reflect women dying or being diagnosed with breast cancer in a
given age interval. All-cause mortality rates and breast cancer incidence rates were summed, and breast
cancer mortality rates were subtracted so that those dying of breast cancer were not counted twice (i.e.,
as deaths and as incident cases of breast cancer). The National Center for Health Statistics 2008-2012
mortality rates for invasive breast cancer (ICD-10 50) in females were obtained from a SEER report
(Howlader et al.. 2014). The SEER report also provided SEER-18 incidence rates for invasive and in
situ breast cancer. The Cox regression results reported by Steenland et al. (2003) are for invasive and in
situ breast cancers combined. It is consistent with the EPA's Guidelines for Carcinogen Risk
Assessment (U.S. EPA. 2005a) to combine these two tumor types because the in situ tumors can
progress to invasive tumors. Thus, the primary risk calculations in this assessment use the sum of
invasive and in situ breast cancer incidence rates for the cause-specific background rates. Comparison
calculations were performed using just the invasive breast cancer incidence rates for the cause-specific
rates; this issue is further discussed in Section 4.1.3 on sources of uncertainty. The risks were computed
up to age 85 for continuous exposures to EtO, conversions were made between occupational EtO
exposures and continuous environmental exposures, and 95% UCLs were calculated for the relative
rates, as described in Section 4.1.1.2 above.
For breast cancer incidence in both the full cohort (see Figure 4-6) and the subcohort with
interviews (see Figure 4-7), the low-exposure categorical results suggest a more linear low-exposure
exposure-response relationship than that obtained with either the continuous variable log cumulative
exposure (supralinear) or cumulative exposure (sublinear) Cox regression models. Thus, as with the
lymphohematopoietic cancer and the breast cancer mortality results above, the EPA proposed in the
2006 draft assessment (U.S. EPA. 2006a), which relied on the original published results of Steenland et
al. (2003). that the best way to reflect the data in the lower exposure region, which is the region of
interest for low-exposure extrapolation, was to do a weighted linear regression of the results from the
model with categorical cumulative exposure (with a 15-year lag). In addition, the highest exposure
group was not included in the regression to provide a better fit to the lower-exposure data.35 Linear
modeling of categorical (i.e., grouped) epidemiologic data and elimination of the highest exposure
35The RR estimates for the highest exposure quintiles suggest somewhat supralinear exposure-response relationships for both
the full cohort and the subcohort with interviews, and supralinearity is evidenced in the subcohort with interviews by the
strong influence of the top 5% of cumulative exposures on dampening the slope of the (cumulative exposure) Cox regression
model (see Section D.l and Figure D-4 of Appendix D). Moreover, there is more uncertainty in using the mean cumulative
exposure to represent the range of exposures in a highest exposure categorical group because such groups contain a wider
range of exposures; for example, for the subcohort with interviews, the highest exposure quintile contains exposures ranging
from about 14,500 ppm x days to over 250,000 ppm x days.
4-41
-------
group(s) under certain circumstances to obtain a better fit of low-exposure data are both standard
techniques used in EPA dose-response assessments (U.S. EPA. 2012. 2005a).
However, as in Section 4.1.1.2 for the lymphohematopoietic cancer data, for the subsequent draft
assessment (U.S. EPA. 2014a. b), the EPA explored additional analyses using the individual data rather
than relying on the published grouped data. Consequently, a two-piece spline model was considered as
an alternative way to address the apparent supralinearity of the full data set (while avoiding the extreme
low-exposure curvature imposed by the log cumulative exposure Cox regression model), and Steenland
was commissioned to do the spline analyses. His findings are reported in Appendix D (see Section D. 1),
and the results for the breast cancer incidence analyses are summarized below. Note that, for the
two-piece spline analyses, only the data from the subcohort with interviews and for the invasive and in
situ breast cancers combined were analyzed because this was the preferred data set, as discussed above.
Steenland also employed a cubic spline model as a semiparametric approach to visualize the
underlying exposure-response relationship; however, this approach produces an overly complicated
function for an empirical model, as opposed to a biologically based model, and was not used for risk
assessment purposes. In addition, Steenland investigated the use of a Cox regression model with a
square-root transformation of cumulative exposure; however, this approach, although less extreme than
using the log transformation of cumulative exposure, also yields a notably supralinear model, which can
result in unstable low-exposure risk estimates (i.e., small changes in exposure correspond to large
changes in risk; see Figure 4-7). The model results for both the cubic spline and square-root
transformation models are included in Appendix D, Section D.l. The cubic spline is not considered
further here, and the square-root transformation model was not considered further in the 2014 draft
assessment but was reexamined for the current assessment. The EPA chose to pursue the development
of two-piece spline models to attempt to avoid the problem of unstable risk estimates from supralinear
curvature in the low-exposure region because these models provide a more general and systematic
approach to modeling supralinear exposure-response data, as opposed to using random, arbitrary
power-transformations of the exposure variable. The SAB panel that reviewed the 2014 draft
assessment (U.S. EPA. 2014a, b) supported the EPA's use of two-piece spline models, recommending
prioritizing models that allow "more local fits in the low exposure range," such as spline models (SAB,
2015).
4-42
-------
1.
1.70
1.60
1.50
1.40
1.30
1.20
1.10
1.00
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
cumulative exposure (ppm* days)
eA((3*exp)
— — eA((3*logexp)
linear
• categorical
eA(P*exp): RR = e(P exposure); eA(P*logexp): RR=e(P hl(exP°sure)); categorical: RR=e(P exposure) with categorical
exposures, plotted at the mean cumulative exposure; linear: weighted linear regression of categorical results, excluding
highest exposure group (see text). [Note that, with the exception of the categorical results and the linear regression of
the categorical results, the various models have different implicitly estimated baseline risks; thus, they are not strictly
comparable to each other in terms of RR values (i.e., along the j'-axis). They are, however, comparable in terms of
general shape.]
Source: Steenland et al. (2003) (except for linear regression of the categorical results, which was done by the EPA).
Figure 4-6. RR estimate for breast cancer incidence in full cohort vs. occupational cumulative exposure (with
15-year lag).
-------
2.20
2.00
a>
71 1.80
-------
For the two-piece log-linear spline modeling approach, as described in Section 4.1.1.2 and
discussed more fully in Appendix D, the breast cancer incidence exposure-response data (cumulative
exposure, with a 15-year lag) were modeled such that log RR is a function of two lines joined at a knot.
The shape of the two-piece spline model, in particular the slope in the low-exposure region, depends on
the location of the knot. The knot was selected by evaluating different knots from 100 to
15,000 ppm x days in increments of 100 ppm x days and then choosing the one that resulted in the best
(largest) model likelihood. The model likelihood did not actually change much across the different trial
knots (see Figure D-l of Appendix D), but it did change slightly, and a knot of 5,800 ppm x days was
chosen for the breast cancer incidence data based on the largest likelihood. The two-piece log-linear
spline model with this knot provided a statistically significant fit to the data (p = 0.01 for the addition of
the exposure terms; see Table D-8 in Appendix D), as well as a good visual fit (see Figure 4-7).
A two-piece linear spline model was also fitted, using the approach of Langholz and Richardson
(2010). who describe methods for fitting nonlog-linear relative hazard models, or "excess relative risk
(ERR)" models. This two-piece linear spline model is similar to the log-linear spline model discussed
above; however, for the linear spline model, the underlying exposure-response function for the splines is
a linear model [i.e., RR = (1 + P x exposure) x exp(E(P x covariates)), where P are the parameters being
estimated, exposure is modeled linearly, and the nonexposure covariates are modeled multiplicatively].36
In the case of the two-piece linear spline model, the knot was obtained by considering possible knots up
to 10,000 ppm x days in increments of 500 ppm x days and then interpolating where appropriate.
A knot of 5,750 ppm x days yielded the largest likelihood (see Figure D-l of Appendix D) for
the two-piece linear spline model. The two-piece linear spline model with this knot provided a
statistically significant fit to the data (p = 0.003; p = 0.014 for the addition of the exposure terms), as
well as a good visual fit (see Figure 4-7). This model had essentially the same AIC as the log-linear
spline model (1,954.4 vs. 1,954.5).37 See Table 4-13 and Section D.l of Appendix D for parameter
estimates and fit statistics for the two spline models. Of the two spline models, the two-piece linear
spline model was selected as the preferred model for the unit risk estimates for breast cancer incidence
primarily because linearity is a desirable property to have in risk assessment models. For example,
linear low-dose extrapolation can occur without a discontinuity between the model in the observable
range and low-dose extrapolation from the point of departure, and the unit risk estimate is not dependent
36For this final assessment, the EPA conducted further analyses. The breast cancer incidence data, which contain protected
personal information on the study participants and are not available to the public, were no longer available to Steenland, who
is no longer at NIOSH. Thus, the EPA arranged withNIOSH to undertake the new breast cancer incidence analyses, and
revised and extended linear exposure-response analyses were conducted by James Deddens of NIOSH, who was also one of
the coauthors of the Steenland et al. (2003, 2004) studies of the NIOSH cohort of EtO sterilizer workers. The details and
comprehensive results of Deddens' analyses are summarized in SectionD.l of AppendixD.
37For the breast cancer incidence data, S AS proc NLMIXED was used for the linear models and proc PHREG was used for
the log-linear models, and the discrepancies that were observed in AIC values between the linear and log-linear models for
the lymphoid cancer data (see footnote 25) were not apparent; thus, these AICs are directly comparable.
4-45
-------
on the risk level chosen for determination of the point of departure, at least within the exposure range of
the first spline segment for a spline model. In addition, with an overall exposure-response relationship
that is supralinear, it seems contradictory to use sublinear model forms for the increments represented by
the spline pieces.
For comparison, linear RR (ERR) models with cumulative exposure, log cumulative exposure,
and a square root transformation of cumulative exposure as continuous variables were also investigated
using the approach of Langholz and Richardson (2010). and these models all fit better than the
corresponding log-linear models based on AIC (see Table 4-14 and Section D. 1 of Appendix D). The
linear and log-linear square root exposure models had marginally lower AICs than the two-piece linear
spline model; however, the two-piece linear spline model is preferred for the reasons discussed above
and in Table 4-14, including the greater flexibility of spline models, which allows more local fit in the
low exposure range. Risk estimates based on the linear models with cumulative exposure and with the
square-root transformation of cumulative exposure are developed for comparison, but the linear model
with the log transformation of cumulative exposure had an inferior fit to that of the linear model with the
square-root transformation (AIC of 1,956.8 vs. 1,952.5; see Table D-2 in Appendix D) and was not
considered further. For more details of the breast cancer incidence exposure-response modeling, see
Section D.l of Appendix D.
4-46
-------
Table 4-13. Exposure-response modeling results for breast cancer incidence
in females from the National Institute for Occupational Safety and Health cohort
for models not presented by Steenland et al. (2003)
Model"
/>-valueb
Coefficient0 (SEd)
(per ppm x day)
Full incidence study cohort1
Linear regression of categorical results, excluding
the highest exposure quintile
0.33
2.64 x 10-5 (2.69 x 10-5)
Subcohort with interviews'
Two-piece log-linear spline (knot at
5,800 ppm x days)
0.01
B1 =7.70 x 10-5 (3.17 x 10-5)
B2 = -7.24 x 10-5
Two-piece linear spline (knot at 5,750 ppm x days)
0.01
B1 = 8.98 x 10-5 (UB1 = L84 x 10-4)d
B2 = -7.79 x 10"5
Cox regression with square root cumulative
exposure
0.006
3.49 x 10-3(1.18 x lO-3)
Linear
0.01
2.30 x 10"5 (UB = 4.67 x 10"5)d
Linear with square root cumulative exposure
0.004
5.53 x lO-3 (ub = i 07 x io-2)d
Linear regression of categorical results, excluding
the highest exposure quintile
0.16
5.17 x 10-5(3.69 x lO"5)
"All with cumulative exposure in ppm x days as the exposure variable and with a 15-yr lag.
' /rvalue for addition of exposure variables from likelihood ratio test, except for the linear regressions of categorical
results, where Wald /^-values are reported.
Tor the two-piece spline models, for exposures below the knot, RR = 1 + (B1 x exp); for exposures above the knot,
RR = 1 + (B1 x exp + B2 x (exp - knot)).
dFor linear models of continuous exposure, the profile likelihood 95% (one-sided) upper bound (UB).
e319 breast cancer cases.
f233 breast cancer cases.
Source: Additional analyses performed by Steenland and Deddens (see SectionD.l of Appendix D), except for the
linear regressions of categorical results, which were performed by the EPA using the equations of Rothman (1986)
presented in Appendix F.
Risk estimates based on the original linear regression analyses of the categorical results are also
presented for comparison. For the approach of using a weighted linear regression of the results from the
Cox regression model with categorical cumulative exposure (and a 15-year lag), excluding the highest
exposure group, the weights used for the ORs were the inverses of the variances, which were calculated
from the confidence intervals.38 Mean and median exposures for the cumulative exposure groups for the
full cohort were provided by Steenland (email dated April 21, 2004, from Kyle Steenland, Emory
38Equations for this weighted linear regression approach are presented in Rothman (1986) and summarized in Appendix F.
4-47
-------
University, to Jennifer Jinot, EPA).39 The mean values were used for the weighted regression analysis
because the (arithmetic) mean exposures best represent the model's linear relationship between exposure
and cancer response. Differences between means and medians were not large for the females, especially
for the lower four quintiles. If the median values had been used, a slightly larger regression coefficient
would have been obtained40, resulting in slightly larger risk estimates. Although the exposure values are
for risk sets from the full cohort, they should be reasonably close to the values for the subcohort with
interviews. See Table 4-13 for the results from the weighted linear regressions of the categorical results
and Figures 4-6 and 4-7 for a depiction of the resulting linear regression models.
As the subcohort with interviews from the NIOSH incidence study cohort provides the preferred
data set for the derivation of unit risk estimates for breast cancer, a summary of all the models
considered for modeling the breast cancer exposure-response data from the subcohort and the judgments
made about model selection is provided in Table 4-14. See Figure 4-7 for visual representations of the
models. See Tables 4-12 and 4-13 and Section D.l of Appendix D for parameter estimates,/(-values,
and other fit statistics.
To facilitate a visual comparison of the models, select models are replotted against the
categorical data in deciles in Figure 4-8. The linear and log-linear two-piece spline models are included,
as spline models were the preferred model form due to their ability to allow more local fits in the low
exposure range. Also included are the linear model with the square-root transformation of cumulative
exposure, as this model had the lowest AIC of the models considered, and the linear model with
untransformed cumulative exposure because this model, being a linear model of the full continuous
exposure data, is expected to provide a lower bound on the likely low-exposure slope, as the overall
exposure-response relationship is supralinear. As can be seen in Figure 4-8, the spline models appear to
have the best fit to the lower-exposure data, which are of the greatest interest in deriving a unit risk for
estimating risk from environmental exposures. The linear square-root model imposes a supralinear
curvature at low exposures (i.e., it has a low-exposure slope that becomes increasingly steep as
exposures decrease); thus, a unit risk estimate derived from this model is highly dependent on the extra
risk level chosen for the point of departure, and very steep low-exposure slopes and large unit risk
estimates can result. It also appears from Figure 4-8 that the linear model has a poorer fit (too shallow)
to the lower-exposure data than either of the two-piece spline models. This is consistent with the
analysis presented in Section D. 1 of Appendix D showing the strong influence of the upper tail of
cumulative exposures on the results of the cumulative exposure Cox regression model. The responses in
the upper tail of exposures are relatively dampened, such that when the highest 5% of exposures are
39Mean exposures for females with a 15-year lag for the exposure categories in Table 3 of Steenland et al. (2003) were 280,
1,241, 3,304, 8,423, and 36,022 ppm x days. Median values were 253, 1,193, 3,241, 7,741, and 26,597 ppm x days. These
values are for the risk sets but should provide a good approximation to the full cohort values.
"'8% greater regression coefficient for the subcohort, and 9% greater regression coefficient for the full cohort.
4-48
-------
excluded, the slope of the Cox regression model is substantially increased (e.g., at 10,000 ppm x days,
the RR estimate increases from about 1.1 to almost 1.5; see Figure D-4 in Appendix D). This strong
influence of the upper tail of exposures would similarly attenuate the slope of the linear model. The
two-piece spline models, on the other hand, are more flexible, and the influence of the upper tail of
exposures would be primarily on the upper spline segment; thus, the two-piece models are able to
provide a better fit to the lower-exposure data.
4-49
-------
Table 4-14. Models considered for modeling the exposure-response data for
breast cancer incidence in females in the subcohort with interviews from the
National Institute for Occupational Safety and Health incidence study cohort for
the derivation of unit risk estimates
Model3
AICb
Comments
Two-piece spline models
Two-piece linear spline model
(knot at 5,750 ppm x days)
1,954.360
SELECTED. Good overall statistical fit and good visual fit,
including local fit to low-exposure range; linear model; AIC
within two units of lowest AIC of models considered.
Two-piece log-linear spline
model (knot at
5,800 ppm x days)
1,954.485
Good overall statistical fit and good visual fit, including local fit
to low-exposure range; preference given to the two-piece linear
spline model primarily because it has the advantageous property
of linearity, but it also has a marginally better statistical fit (lower
AIC).
Lincar(ERR) models (RR = 1 + |i x exposure)
Linear model with square-root
transformation of cumulative
exposure
1,952.501
Good overall statistical fit and lowest AIC; low-exposure slope
becomes increasingly steep as exposures decrease, and large unit
risk estimates can result; preference given to the two-piece spline
models because they have a better ability to provide a good local
fit to the low-exposure range.
Linear model with
untransformed cumulative
exposure
1,954.526
Good overall statistical fit but poorer local fit to low-exposure
range than the two-piece spline models; higher AIC than selected
model.
Log-linear (Cox regression) models (RR = c11 x <,xP(lslmi)
Log-linear model with
square-root transformation of
exposure
1,953.028
Good overall statistical fit; low-exposure slope becomes
increasingly steep as exposures decrease, and large unit risk
estimates can result; preference given to the two-piece spline
models because they have a better ability to provide a good local
fit to the low-exposure range.
Log-linear model with (natural)
log cumulative exposure
1,956.176
Good overall statistical fit but poor local fit to low-exposure
range; low-exposure slope becomes increasingly steep as
exposures decrease, and large unit risk estimates can result; higher
AIC than selected model.
Log-linear model (standard Cox
regression)
1,956.675
Good overall statistical fit but poor local fit to low-exposure range
(too shallow); AIC exceeds that of selected model by >2.
Linear regression of categorical results
Linear regression of categorical
results, excluding the highest
exposure quintile
C
Not statistically significant, as one might expect because the
approach, which is based on categorical data, has low statistical
power; preference given to models that treated exposure as a
continuous variable and that also provided reasonable
representations of the low-exposure region
"All with cumulative exposure as the exposure variable, except where noted, and with a 15-yr lag, and all with
exposure as a continuous variable except for the linear regression of categorical results.
bAIC = 2p-2LL, where p = number of parameters and LL = ln(likelihood), assuming two exposure parameters for
the two-piece spline models.
°Not calculated.
4-50
-------
2.4
2.2
2.0
1.8
(0
-------
In conclusion, the two-piece linear spline model with knot at 5,750 ppm x days was selected for
the derivation of the unit risk estimates for breast cancer incidence. Selection of this model is consistent
with the model selction objectives for this assessment (see Section 4.1.1.2). First, the model uses the
individual-level exposure data. Second, the spline model is more tuned to local behavior than the other
model forms considered. Third, the principle of parsimony was applied, both in the use of a preselected
knot (with sensitivity analyses conducted to examine the sensitivity of the unit risk estimates to this
preselection; see Section D.1.7 of Appendix D) and in the use of AICs to compare models, but without
over-reliance on this statistic. Fourth, the general model shape is biologically plausible and the selected
model is consistent with the data. In addition, the same model can be used to derive the estimates of
extra risk for occupational exposure scenarios.
The proportional hazards assumption for the selected model (two-piece linear spline model with
knot at 5,750 ppm x days) was tested by evaluating the significance of an age-interaction term for each
spline regression coefficient, and neither interaction term was statistically significant (see Section D.1.9
of Appendix D).
The exposure level (ECX) and the associated 95% lower confidence limit (LECX) corresponding
to an extra risk of 1% (x = 0.01) for breast cancer incidence in females (based on invasive + in situ
tumors in the subcohort with interviews) for the models discussed above were estimated using the
actuarial program (life-table analysis). As noted in Section 4.1.1.2, a 1% extra risk level is a more
reasonable response level for defining the POD for these epidemiologic data than a 10% level. The
results are presented in Table 4-15. For comparison, the results for some models using the data from the
full cohort of the breast cancer incidence study are also presented, but the subcohort with interviews is
the preferred data set.
4-52
-------
Table 4-15. ECoi, LECoi, and unit risk estimates for breast cancer incidence
(invasive and in situ) in females from various models3
Modelb
With interviews
Full cohort
ECoi (ppm)
LECoic (ppm)
Unit risk
(per ppm)
ECoi (ppm)
LECoic
(ppm)
Unit risk
(per ppm)
Two-piccc spline models
Low-exposure linear
splined
0.0138
6.75 x 10-3
1.48e
Low-exposure
log-linear splined
0.0155
9.27 x 10-3
1.08e
Lincar(ERR) models (RR = 1 + |i x exposure)
Linear model with
square root of
cumulative exposures
2.76 x 10-3
7.42 x 10-4
13.5e-h
Linear model with
untransformed
cumulative exposures
0.0540
0.0266
0.376
Log-linear (Cox regression) models (RR = c'1 x (S|"MI'1)
Log-linear model with
square root of
cumulative exposure1
6.49 x 10-3
2.68 x 10-3
3.73e
Log-linear model with
(natural) log
cumulative exposure!
k
k
k
k
k
k
Log-linear model
(standard Cox
regression^
0.126
0.0737
0.136h
0.222
0.107
0.0935h
Linear regression of categorical results
Linear regression of
categorical results,
excluding highest
exposure quintilei1
0.0240
0.0110
0.909
0.0469
0.0176
0.568
aAll-cause mortality adjusted (to dying of something other than breast cancer or developing breast cancer). Unit
risk= 0.01/LECoi.
bAll with cumulative exposure as the exposure variable, except where noted, and with a 15-yr lag, and all with
exposure as a continuous variable except for the linear regression of categorical results.
Confidence intervals used in deriving the LEC01S were estimated employing the Wald approachfor the log-linear
RR models and a profile likelihood approach, which allows for asymmetric CIs, for the linear RR models
CLangholz and Richardson. 20101.
dFrom low-exposure segment of two-piece spline analysis; see text and Table D-4 of Appendix D for log-linear
model or Table D-10 for linear model. The EC0i value is below the value of 0.074 ppm roughly corresponding to a
knot of 5,750 ppm x days [(5,750 ppm x days) x (10 rrf^O m3) x (240 d/365 d)/(365 d/yr x 70 yr) = 0.074 ppm],
and thus, appropriately in the range of the low-exposure segment.
Tor unit risk estimates above 1, one can convert to risk per ppb (e.g., 1.75 perppm= 1.75 x 10"3perppb).
''Not estimated; two-piece spline models, linear RR models, and log-linear RR model with square-root
transformation of exposure not developed for the full cohort.
4-53
-------
Table 4-15. EC01, LECOl, and unit risk estimates for breast cancer
incidence (invasive and in situ) in females from various models3 (continued)
gFrom linear analyses in Section D. 1.3.2 and Table D-10 of AppendixD.
This unit risk estimate is not considered a good estimate of risks from (low) environmental exposure levels (see
text).
'From Table D-6 of Appenidx D.
JFrom Tables 4 and 5 of Steenland et al. (2003). Cox regression models.
kEstimated exposure levels were so low (<3. 6 x 104 ppm) that the cumulative exposures in some of the age
intervals in the life-table analysis were <1 ppm x day, resulting in In cumulative exposures of <0 for those intervals
and unreliable results for this model. 3.6 x 10"4 ppm would yield a unit risk estimate of about 28 per ppm, which
provides a lower bound on the unit risk estimate for this model.
'Regression coefficient derived fromlinear regression of categorical results, as described in Section4.1.2.3.
Because EtO is DNA-reactive and has direct mutagenic activity (see Section 3.3.3), which is one
of the cases cited by the EPA's Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a) for the
use of linear low-dose extrapolation, a linear low-exposure extrapolation was performed. (Linear
low-exposure extrapolation is also the default approach used in the absence of sufficient evidence for a
nonlinear mode of action, which is also the case for EtO [see Section 3.4].) The inhalation unit risk
estimates for the different breast cancer incidence models considered are presented in Table 4-15.
As discussed above, the unit risk estimate based on the two-piece linear spline model using
cumulative exposure with a 15-year lag (i.e., 1.48 per ppm, or 1.48 x 10"3 per ppb) is the preferred
estimate for breast cancer incidence. The two-piece log-linear spline model resulted in a unit risk
estimate of 1.08 per ppm, while the linear regression of categorical results yielded a unit risk estimate of
0.909 per ppm and the continuous exposure linear model produced a unit risk estimate of 0.376 per ppm;
these alternate estimates are about 73%, 61%, and 25%, respectively, of the estimate based on the
preferred two-piece linear spline model.
ECoi, LECoi, and unit risk estimates from the other models examined are presented for
comparison only, to illustrate the differences in model behavior at the low end of the exposure-response
range. The unit risk estimates from these other models are not considered reliable estimates of risk from
(low) environmental exposures because, although these models provide an adequate global fit to the
overall data, they do not allow a good local fit to the low-exposure data (see Figure 4-7). As discussed
above, the log-linear log cumulative exposure model and both the linear and log-linear square-root
transformation models impose a low-exposure supralinear curvature that can result in inappropriately
large unit risk estimates. For example, for the log-linear log cumulative exposure model, a reliable unit
risk estimate could not be calculated from the 1% extra risk level, but the substantially lower ECoi
estimate (<3. 6 x 10~4 ppm) from that model compared to that from the two-piece linear spline model
(0.0138 ppm) indicates a much higher unit risk. The standard Cox regression cumulative exposure
model, on the other hand, with its global sub linearity, has an overly shallow slope in the low-exposure
range, yielding a notably higher ECoi estimate (0.126 ppm) than that from the two-piece linear spline
4-54
-------
model (0.0138 ppm) and a correspondingly lower unit risk estimate (0.136 per ppm vs 1.48 per ppm).
The linear untransformed exposure model might be considered a lower bound to a range of credible unit
risk estimates—unit risk estimates below that value (0.376 per ppm) are considered unlikely from the
available data, given that for a global linear model, the high-exposure results would dampen the
low-exposure slope (as discussed above with respect to the log-linear [standard Cox regression] model
[and see Section D. 1 and Figure D-4 of Appendix D]). There is no clear upper bound to a range of
credible unit risk estimates due to the supralinear curvature of the alternative models. The next highest
unit risk estimates are 3.73 per ppm and 13.5 per ppm, from the log-linear and linear square-root
transformation models, respectively, but there is no basis for considering either of those estimates an
upper bound to a range of credible estimates.
As discussed above, the primary risk calculations for breast cancer incidence were based on
invasive and in situ tumors in the subcohort of women with interviews, and the primary model was the
two-piece linear spline model. For this assessment, the two-piece spline analyses were not performed
with the full cohort. However, in the 2006 draft assessment (U. S. EPA. 2006a). comparison analyses
were done. Using the linear regression of the categorical results, the comparable unit risk estimate for
the full cohort was about 40% lower than the estimate based on the subcohort with interviews. A lower
estimate from the full cohort is consistent with the expectation that there was an under-ascertainment of
cases in the full cohort, as discussed above. Using the two-piece linear spline model, the corresponding
unit risk estimate derived based on the subcohort results but using invasive breast cancer only for the
background incidence rates was about 20% lower than the estimate based on invasive and in situ tumors,
reflecting the difference between incidence rates for invasive breast cancer only and for combined in situ
and invasive breast cancer.
Sensitivity analyses were also conducted to investigate the influence of lag period, inclusion of
covariates, knot selection, and upper-bound estimation approach on the unit risk estimates from the
selected two-piece linear spline model. The sensitivity analyses are detailed in Sections D.1.6, D.1.7,
D. 1.8, and D. 1.10 of Appendix D. In brief, for the two-piece linear spline model with the knot at
5,750 ppm x days, the unit risk estimates for different lag periods (0, 5, 10, 15, and 20 years) ranged
from about 35% less than (10-year lag) to about 21% greater than (20-year lag) the estimate for the
selected model (15-year lag). (Note that the two-piece linear spline model with the knot at
5,750 ppm x days and a 20-year lag had a slightly better fit, based on log likelihood and AIC, than the
model with a 15-year lag [see Table D-12 of Appendix D].) Exclusion of covariates produced very little
difference in the unit risk estimates—excluding parity and both parity and breast cancer in a first-degree
relative changed the unit risk estimate by only about 1%. Similarly, varying the knot by
1,000 ppm x days resulted in little difference in the unit risk estimates—the unit risk estimate with the
knot at 4,750 ppm x days was about 14% greater than, and unit risk estimate with the knot at
6,750 ppm x days was about 11% less than, the unit risk estimate for the selected model (with the knot
4-55
-------
at 5,750 ppm x days). The unit risk estimate calculated using a Wald approach was about 3% lower
than the preferred estimate, which relied on a profile likelihood approach to estimate the upper bound on
the regression coefficient.
This unit risk estimate for breast cancer incidence is slightly lower than the unit risk estimate
derived for breast cancer mortality above (see Section 4.1.2.2), which is contrary to expectations.
Confidence is higher in the breast cancer incidence estimate because it is based on more cases,
especially in the lower-exposure region, and a better-fitting model.
The life-table analysis takes into account competing risks and the occurrence of different
cumulative exposures and different cause-specific background risks at different ages. A crude
approximation of the general approach for obtaining ECoi and LECoi estimates without the use of the
life-table component of the analysis is presented here for illustration. In this crude approach, an estimate
of the lifetime background risk of developing breast cancer is used rather than age-specific rates.
According to SEER data, this lifetime background incidence risk estimate (Ro) is 12.3%.41 From this
and eq 4-2, an estimate of the RR associated with a 1% extra risk can be calculated as RR = (0.99 x Ro
+ 0.01)/Ro = 1.07. Then, a maximum likelihood estimate of the cumulative exposure associated with
this RR can be calculated from the low-exposure spline segment from the selected two-piece linear
spline model as exposure = (RR - 1)/Pi. This quantity is an occupational cumulative exposure in ppm x
days. To convert to environmental ppm x years, multiply by (10 m3 breathed at work/day)/(20 m3/day)
and (240 days worked/year)/(365 days/year), as discussed in Section 4.1.1.2, and then divide by
365 days/year. Because the life-table analysis is based on actual demographic rates, this crude
approximation uses an average U. S. life expectancy of 80 years42 rather than the EPA default average
lifespan of 70 years, for a more appropriate comparison. With a 15-year lag, this means dividing the
cumulative exposure by 65 years to get the continuous lifetime exposure level associated with a 1%
extra risk (ECoi). The LECoi is obtained using the same calculations but with the profile likelihood
upper bound on betal in place of betal. With betal of 8.98 x 10"5 per ppm x day and a profile
likelihood upper bound on betal of 1.84 x 10"4 per ppm x day from the selected two-piece linear spline
model, these calculations yield an ECoi of 0.0110 ppm and an LECoi of 0.00538 ppm. In comparison to
the estimates presented for the selected two-piece linear spline model in Table 4-15, this crude approach
yields ECoi and LECoi estimates that are both about 20% lower, which would correspond to a unit risk
estimate about 26% higher.
In summary, the unit risk estimate of 1.48 per ppm (1.48 x 10"3 per ppb) is the preferred
estimate for female breast cancer risk because it is based on incidence data versus mortality data, it is
based on more cases (n = 233) than the mortality estimate (n = 103), and information on personal breast
"Based on2010-2012 data; http://seer.cancer.gov/statfacts/html/breast.html.
42The overall U.S. life expectancy in womenin2014 was 81.2 years (Murphy et al.. 20151.
4-56
-------
cancer risk factors obtained from the interviews is taken into account. Furthermore, the two-piece linear
spline model, which uses the complete data set with exposure as a continuous variable, was statistically
significant, had an AIC within two units of the model with the lowest AIC, and provided a good visual
fit to the lower-exposure data. Converting the units, 1.48 per ppm corresponds to a unit risk estimate of
8.09 x 10"4 per [j,g/m3.
4.1.3. Total Cancer Risk Estimates
According to the EPA's Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a), cancer
risk estimates are intended to reflect total cancer risk, not site-specific cancer risk; therefore, an
additional calculation was made to estimate the combined risk for (incident) lymphoid and breast
cancers because females would be at risk for both cancer types. The unit risk estimates for both of the
individual models for these cancers were derived from linear RR models and are based on profile
likelihood upper-bound estimates of the regression coefficient (Lanuholz and Richardson. 2010). It was
not possible to derive the total cancer unit risk estimate using a profile likelihood approach; thus, a Wald
approach was employed to estimate the combined risk.
To derive a total cancer unit risk estimate, it is assumed that the cancer types are independent,
which is a reasonable assumption for the cancer types involved. In addition, to employ the Wald
approach, it is assumed that the risk estimates are approximately normally distributed. For breast
cancer, this is reasonable, as the Wald-based estimate was only 3% less than the profile-likelihood-based
unit risk estimate. For lymphoid cancer, the Wald-based estimate was 40% less than the
profile-likelihood-based unit risk estimate by 40%, indicating that the total cancer unit risk estimate will
also be underestimated by using the Wald approach. In fact, the total cancer unit risk estimate
calculated using the Wald-based unit risk estimates and Wald standard errors (SEs) (see Sections D. 1.10
and D.3.8 of Appendix D) was less than the profile-likelihood-based unit risk estimate for lymphoid
cancer alone (4.03 vs. 5.26 per ppm); thus, Wald-type SEs were approximated from the profile
likelihood upper bounds, and these approximated SEs were used with the profile-likelihood-based unit
risk estimates in the Wald approach to derive the total cancer unit risk estimate.
Under the Wald approach, one can estimate the 95% UCL (one-sided) on the total risk as the
95%) UCL on the sum of the maximum likelihood estimates of the risk estimates according to the
formula
95% UCL = MLE + 1,645(SE), (4-3)
where MLE is the MLE of total cancer risk (i.e., the sum of the individual MLEs) and the SE of the sum
of the MLEs is the square root of the sum of the individual variances (i.e., the variance of the sum is the
4-57
-------
sum of the variances, and the SE is the square root of the variance). Because both models are linear in
the range around the PODs, the combining-risk calculations can be done directly from the unit risk and
0.01/ECoi estimates rather than having to do the calculations at a common exposure level near where the
ECoi and LECoi for the combined risk would be.
First, the 0.01/ECoi estimates were calculated and an ECoi of 4.85 x 103 ppm for the total cancer
risk (i.e., lymphoid cancer incidence + breast cancer incidence) was estimated, as summarized in
Table 4-16.
Table 4-16. Calculation of ECoi for total cancer risk
Cancer type
ECoi (ppm)
0.01/ECoi (per ppm)
ECoi for total cancer risk
(ppm)
Lymphoid
0.00748
1.34
--
Breast
0.0138
0.725
--
Total3
--
2.06
0.00485
"The total 0.01/ECoi value equals the sum of the individual 0.01/ECoi values; the ECoi for the total cancer risk then
equals 0.01/(0.01/ECoi).
Then, a unit risk estimate of 6.06 per ppm for the total cancer risk (i.e., lymphoid cancer
incidence + breast cancer incidence) was derived, as shown in Table 4-17. An LECoi estimate of
1.65 x 10"3 ppm for the total cancer risk can be calculated as 0.01/(6.06 per ppm).
Table 4-17. Calculation of total cancer unit risk estimate
Cancer type
Unit risk
estimate3
(per ppm)
0.01/ECoi
(per ppm)
SEb
(per ppm)
Variance
Total cancer
unit risk
estimate
(per ppm)
LECoi for total
cancer risk0
(ppm)
Lymphoid
5.26
1.34
2.38
5.69
--
--
Breast
1.48
0.725
0.459
0.211
--
--
Total
--
2.06
[2.43]d
5.90
6.06e
0.00165
aProfile-likelihood-based unit risk estimates from the selected linear two-piece spline models.
bSE ~ (unit risk—0.01/ECoi)/1.645; Wald-type SEs were approximated from the profile-likelihood-based unit risk
estimates (i.e., profile likelihood upper bounds).
°The LECoi for the total cancer risk equals 0.01/(total cancer unit risk estimate).
dThe SE of the total cancer risk is calculated as the square root of the sum of the variances (next column), not as the
sum of the SEs.
eTotal cancer unit risk = 2.06 + 1.645 x 2.43.
4-58
-------
Thus, the total cancer unit risk estimate is 6.1 per ppm (or 6.1 x 10"3 per ppb; 3.3 x 10"3 per
(j,g/m3). As can be seen in Table 4-17, lymphoid cancer contributes about 2/3 of the risk to the sum of
the MLEs and somewhat more (between about 75 and 85%) to the total cancer unit risk estimate (i.e., to
the UCL on the sum).
Recall that this is the unit risk estimate derived under the assumption that RR is independent of
age (see Section 4.1.1.2). The preferred assumption of increased early-life susceptibility, in accordance
with the EPA's Supplemental Guidance (U. S. EPA. 2005b). is considered in Section 4.4.
While there are uncertainties regarding the assumption of a normal distribution of risk estimates,
the resulting unit risk estimate is appropriately bounded in the roughly 1.1-fold range between the sum
of the unit risk estimate for lymphoid cancer plus the 0.01/ECoi estimate for breast cancer (5.98 per
ppm), which provides a lower bound for the upper bound on the sum of the individual MLEs (i.e.,
0.01/ECoi estimates), and the sum of the individual 95% UCLs (i.e., unit risk estimates) (6.74 per ppm).
Thus, any inaccuracy in the total cancer risk estimate resulting from the approach used to combine risk
estimates across cancer types is relatively minor.
4.1.4. Sources of Uncertainty in the Human-Data-Based Cancer Risk Estimates
Discussion of the sources of uncertainty in the human-data-based unit risk estimates derived in
Sections 4.1.1.2, 4.1.2.2, and 4.1.2.3 above is organized into three subsections: (1) limitations in the
human database, (2) uncertainties that stem from the epidemiological study used as the basis for the unit
risk estimates and its analysis, and (3) uncertainties associated specifically with the total cancer
(incidence) unit risk estimate.
4.1.4.1. Limitations in the Human Database
The availability of suitable human data from which to derive unit risk estimates for EtO
eliminates one of the primary sources of uncertainty inherent in most unit risk estimates—the
uncertainty associated with interspecies extrapolation. Nonetheless, there are limitations in the human
database on cancer riskfrom EtO exposure that contribute uncertainty to the unit risk estimates
(estimates based on rodent cancer data and the sources of uncertainty pertaining to those estimates are
discussed in Section 4.2 below). The sources of uncertainty related to these limitations include
availability of only a single study with superior characteristics, the derivation of unit risk estimates for
the general population from an occupational study, and in the case of the lymphoid cancer (incidence)
unit risk estimate, the derivation of the incidence estimate from mortality data.
4.1.4.1.1. Single study with superior characteristics
Three independent epidemiology studies conducted exposure-response analyses based on
exposure estimates for the individual workers; however, the NIOSH study (Steenland et al., 2004;
4-59
-------
Steenland et al.. 2003) was judged to be substantially superior to the other two studies (Mikoczv et al..
2011; Swaen et al.. 2009) for the purposes of deriving a unit risk estimate, as discussed at the beginning
of Section 4.1. Although only one study was used for the unit risk estimation from human data, it is a
large study that included workers from 13 different sterilization facilities in different states, decreasing
the likelihood that the results are overly influenced by uncontrolled confounding related to either
location or a specific facility.
The limitations and sources of uncertainty discussed below notwithstanding, the NIOSH study is
considered a high-quality study for the purposes of deriving a unit risk estimate. The NIOSH study is a
large longitudinal cohort study that developed individual-worker exposure estimates using detailed
employment histories and a state-of-the-art regression model for retrospectively estimating exposures
based on time-period-specific plant and job variables. In addition, internal comparisons were used for
estimating risk from EtO exposure, coexposures to other chemicals in the facilities were reportedly
minimal, and in the analysis of the breast cancer incidence subcohort with interviews, various
nonexposure-related breast cancer risk factors were taken into account. The NIOSH study comprises a
large cohort that has been followed for a sufficient length of time for cancer detection. The cohort
includes 17,530 workers with exposure estimates, about 55% of whom are women, contributing a total
of over 400,000 exposed person-years to the most recent follow-up (Steenland et al.. 2004). The mean
follow-up time was about 27 years. Thus, although the unit risk estimates are based on a single study,
there is relatively high confidence in that study.
The UCC study (Swaen et al.. 2009) is considered largely uninformative in terms of assessing
the unit risk estimates derived from the NIOSH study because of the crude exposure assessment used in
the UCC study and because of differences in the exposure-response analyses conducted [e.g., in their
internal analyses, Swaen et al. (2009) only used Cox regression models in cumulative exposure (i.e.,
sublinear models), and may have over-adjusted by including age at hire in their models] (see
Section A.2.20 of Appendix A). In addition, the Swaen et al. (2009) study was restricted to males, so
there are no female breast cancer data available, and there were only 17 lymphoid cancer deaths
compared to 53 in the NIOSH study.
The Mikoczv et al. (2011) study of sterilizer workers, although small, appeared to have a
well-done exposure assessment; however, the data reported by Mikoczv et al. (2011) were not well
suited for the derivation of a unit risk estimate. Thus, crude comparison analyses were done to evaluate
whether or not the exposure-response models of the NIOSH study that were used to derive unit risk
estimates in this assessment gave predictions consistent with the Mikoczv et al. (201 1) internal
incidence ratios (IIRs) for the two highest exposure quartiles (see Section J.2.2 of Appendix J). The
predicted values for lymphoid cancer were within the 95% CIs for the IIRs for lymphohematopoietic
cancer reported by Mikoczv et al. (2011). The predicted values for breast cancer incidence, however,
were below the lower limit of the 95% CIs for the IIRs for breast cancer, suggesting that the Mikoczv et
4-60
-------
al. (201 1) results are consistent with a higher unit risk estimate for breast cancer incidence than the one
derived in this assessment. The reasons for the discrepancies are unknown, although it might be noted
that a less rigorous approach was used to estimate historical exposure levels for the plants in the
Mikoczy et al. (2011) study than the regression model that was developed for the NIOSH study and that
there were many fewer breast cancer cases in the Mikoczy et al. (201 1) study (41 incident cases [33 with
>15 years since time of first exposure] vs. 233 cases [at least 170 with >15 years since time of first
exposure] in NIOSH's subcohort with interviews).
4.1.4.1.2. High-to-low dose extrapolation
The human-based estimates are also less affected by high-dose to low-dose extrapolation than are
rodent-based estimates, and thus, uncertainty from that source is less than for estimates from rodent
studies. For example, the average exposure in the NIOSH cohort was more than 10 times lower than the
lowest exposure level in a rodent bioassay after adjustment to continuous lifetime exposure.
Nonetheless, uncertainty remains in the extrapolation from occupational exposures to lower
environmental exposures.43 Although the actual exposure-response relationship at low exposure levels
is unknown, the clear evidence of EtO mutagenicity supports the linear low-exposure extrapolation that
was used (U.S. EPA. 2005a). The linear low-exposure extrapolation from the 95% lower bound on the
exposure level associated with the 1% extra risk level is considered to be a plausible upper bound on the
risk at lower exposure levels. Actual low-exposure risks are expected to be lower, to an unknown
extent.
Because of the existence of endogenous EtO (see Section 3.3.3.1), several members of the SAB
panel that reviewed the EPA's 2006 external review draft assessment (U. S. EPA. 2006a) felt that the
exposure-response relationship for cancer at low exposures would be nonlinear and suggested that it
would be consistent with the EPA's 2005 Guidelines for Carcinogen Risk Assessment (U. S. EPA.
43Even though lifetime cumulative exposures from environmental sources of EtO may overlap the low end of the range of
lagged cumulative exposures from occupational sources, the exposure-response model is based on the full range of the
occupational exposure data and not just the data in the range of environmental exposures, which are too sparse to model on
their own Even with a two-piece spline model which gives a more local fit to the low-exposure data, there is uncertainty
about the exposure-response relationship specifically in the range of environmental exposures. The point of departure
(LECoi) is intended to be at the low end of the "observable" range (i.e., the range of exposures for which the study might be
able to detect a significant increase in risk), but it is still substantially above typical environmental exposure levels (according
to the EPA's 2005 National Air Toxics Assessment data, the average exposure concentration of EtO fromall sources
[including background] in the United States is 0.0062 (ig/m3 [3.4 x 106 ppm]; the average background concentration is
0.0044 (ig/m3 [2.4 x 10 6 ppm]), and thus, there is uncertainty about the low-exposure extrapolation from the points of
departure (6.75 x 10 3 ppm, or 12 (ig'm3, for breast cancer incidence and 1.9 x 10 3 ppm, or 3.5 (ig'm3, for lymphoid cancer
incidence). For lymphoid cancer, only 2 of the 13 cases below the knot (-15%) are also below the point of departure. For
breast cancer, about 25% of the cases below the knot are also below the point of departure, roughly corresponding to the
lowest 1.5 deciles (see Table D-l of Appendix D), but as one can see from Table D-3 of Appendix D, the variability of the
low-exposure data is such that the two lowest deciles both have RR estimates <1, and thus, are not by themselves consistent
with the unit risk estimate, illustrating the uncertainty that still exists in the low-exposure extrapolation.
4-61
-------
2005a) to present a nonlinear approach for "extrapolation" to lower exposures (SAB. 2007). The EPA
considered this suggestion but judged that the support for a nonlinear approach was inadequate. In brief,
as discussed in Sections 3.1 through 3.3.3, EtO is a DNA-reactive, mutagenic, multisite carcinogen in
humans and experimental species; as such, it has the hallmarks of a compound for which low-dose linear
extrapolation is strongly supported under the EPA's Guidelines for Carcinogen Risk Assessment (U.S.
EPA. 2005a). The EPA's Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a) provide for
presenting alternate approaches when those alternatives have significant biological support; however,
the EPA's analysis of the arguments for using a nonlinear approach presented on page 23 and in
Appendix C of the SAB report (SAB. 2007) did not find these arguments to be persuasive. The
arguments posited by the SAB panel members who supported using a nonlinear approach were largely
that (1) DNA adducts may show a nonlinear response when identical adducts are formed endogenously
and (2) mutations do not have linear relationships with exposure but exhibit an "inflection point."
However, as discussed in Section 3.3.3.1, recent data from Marsden et al. (2009) support a linear
exposure-response relationship for EtO exposure and DNA adducts (p < 0.05) and demonstrate increases
of DNA adducts from exogenous EtO exposure above those from endogenous EtO for very low
exposures to exogenous EtO, providing evidence against the first argument. Moreover, Appendix C of
the SAB report (SAB. 2007) presents two EtO-specific mutation data sets in support of the second
argument; however, the EPA's analysis of these data sets found that they are in fact consistent with
low-dose linearity. See the response to this comment under Charge Question 2.b in Appendix H for a
more comprehensive discussion of the EPA's consideration and rejection of a nonlinear approach and
for the details of the EPA's analysis of the two EtO mutation data sets.
The EPA also considered more recent (2013) public comments proposing nonlinear modes of
action for EtO carcinogenicity; however, the EPA found these hypothetical modes of action to be
speculative at this time (see Appendix K and Section J.3.2 of Appendix J).
4.1.4.1.3. Generalizability of estimates from a worker population
The NIOSH data represent an industrial worker cohort that is generally healthier than the U. S.
population at large. The healthy-worker effect is often an issue in occupational epidemiology studies,
but the use of internal exposure-response analyses help address this concern, at least partially. The unit
risk estimates derived from the NIOSH worker cohort data could underestimate the cancer risk for the
general population to an unknown extent, although the impact is expected to be relatively low for the
majority of the population.
Industrial workers can also differ from the general population in factors other than health status.
In terms of representing the general population, the NIOSH study cohort was relatively diverse. It
contained both female (55%) and male (45%) workers, and the workers were 79% white, 16% black,
and 5% "other."
4-62
-------
The workers were, however, all adults, and a related area of uncertainty pertains to the
assumption that RR is independent of age, which is a common assumption in the dose-response
modeling of epidemiological data and is an underlying assumption in the Cox regression model. For the
NIOSH worker cohort, the proportional hazards model assumption of RR being independent of age was
tested by checking the significance of an interaction between age and cumulative exposure, and none of
the models had a significant interaction term (see Sections D. 1.9 and D.3.7 of Appendix D). This
suggests that, for adults at least, the assumption that RR is independent of age is valid. However, the
worker cohort contains no children and is uninformative on the issue of early-life susceptibility. In the
absence of data on early-life susceptibility, the EPA's Supplemental Guidance (U. S. EPA. 2005b)
recommends that increased early-life susceptibility be assumed for carcinogens with a mutagenic mode
of action, and the conclusion was made in Section 3.4 that the weight of evidence supports a mutagenic
mode of action for EtO. Thus, in accordance with the Supplemental Guidance, the alternate assumption
of increased early-life susceptibility is preferred as the basis for risk estimates in this assessment, and
risk estimates derived under this preferred assumption, and intended for the application of age-dependent
adjustment factors (ADAFs), are presented in Section 4.4.
4.1.4.1.4. Derivation of lymphoid cancer incidence estimates from mortality data
The study reported by Steenland et al. (2004) is a retrospective mortality study, and cancer
incidence data are not available for lymphohematopoietic cancer [for breast cancer, a separate incidence
study (Steenland et al., 2003) was available]. This limitation was addressed quantitatively in the
life-table analysis by using incidence rates instead of mortality rates for the cause-specific background
rates in order to derive unit risk estimates for lymphoid cancer incidence from the exposure-response
modeling results from the mortality data. Although assumptions are made in this approach, as discussed
in Section 4.1.1.3, the resulting incidence-based estimates are considered to be better estimates of cancer
incidence risk than are the mortality-based estimates. The incidence unit risk estimate for lymphoid
cancers is about 160% higher than (i.e., 2.6 times) the mortality-based estimate, which seems reasonable
given the relatively high survival rates for lymphoid cancers (according to SEER data from 2006-2012
rwww.seer.cancer.gov!. 5-year survival rates are about 71% for NHL; 83% for chronic lymphocytic
leukemia, which is the vast majority of the lymphocytic leukemias in adults; and 49% for multiple
myeloma).
4.1.4.1.5. Possibility that not all potential cancer sites are reflected in the unit risk estimate
The two types of cancer identified in epidemiology studies of EtO exposure as being of concern
to humans were also associated with EtO exposure in rodent studies (lymphohematopoietic cancers
female mice and in male and female rats and mammary gland tumors in female mice). However, the
rodent data suggest associations between EtO exposure and other tumor types as well, and although site
4-63
-------
concordance across species is not generally assumed, it is possible that the NIOSH study, despite its
relatively large size and long follow-up (mean length of follow-up was 26.8 years), had insufficient
power to observe small increases in risk in certain other sites (not limited to those in which tumors were
observed in rodents). For example, the tumor site with the highest potency estimate in both male and
female mice was the lung. In the NIOSH study, one cannot rule out a small increase in the risk of lung
cancer, which has a high background rate, thus making small increases difficult to detect.
4.1.4.2. Sources of Uncertainty Stemming from the NIOSH Studies and Their Analyses
Other sources of uncertainty arise from the epidemiologic studies and their analyses [Steenland
et al. (2004);Steenland et al. (2003); and Steenland and Deddens analyses in Appendix D], including the
retrospective estimation of EtO exposures in the cohort and the exposure-response modeling of the
epidemiologic data. Sources of uncertainty pertaining to the exposure-response modeling are discussed
first for lymphoid cancer and then for breast cancer and include endpoint refinement, model selection,
lag time, exposure metric, and potential confounding or modifying factors. Although these are common
areas of uncertainty in epidemiologic studies, they were generally well addressed in the NIOSH studies.
4.1.4.2.1. Exposure estimation
Regarding exposure estimation, the NIOSH investigators conducted a detailed retrospective
exposure assessment to estimate the individual worker exposures. They used extensive data from
18 facilities, including charcoal tube measurement data from 1976 to 1985, to develop and test a
regression model for estimating EtO exposure levels associated with different jobs (exposure
categories), facilities, and time periods (Hornung et al., 1994; Greife et al., 1988) (see also Section A.2.8
for more details about the development and evaluation of the regression model). The model accounted
for 85% of the variation in average EtO exposure levels in an independent set of test data (from the
18 facilities, 6, with measurement data from 1979-1985, were randomly selected for model evaluation,
and the other 12 were used for model development). In addition, the modeled estimates were not highly
biased nor biased in one direction when compared to the predictions of a panel of 11 industrial hygiene
experts familiar with EtO levels in the sterilization industry.
The regression model was used to develop an exposure matrix stratified by time and exposure
category for each facility in the cohort study. Detailed work history data for the individual workers were
collected for the 1987 follow-up and used in conjunction with the facility-specific exposure matrices to
derive cumulative exposure estimates for the individual workers (Steenland et al., 1991). Thus, although
measurement data were not available for most of the time that the cohort was exposed (exposures started
in 1938 for some workers), exposure levels for those early time periods could be estimated from the
regression model based on variables for which historical data were available (e.g., plant- and
year-specific sterilizer volume), which served as a surrogate measure for the amount of EtO used.
4-64
-------
Another variable, calendar year, served as a surrogate for general improvements in work practices after
the human health effects of EtO became a matter of concern in the late 1970s. This variable captured
decreases in exposure after the late 1970s that were unaccounted for by the other variables. For the
years before 1978, when human health effects of EtO were not a large concern, it was assumed that the
other variables more fully accounted for exposure levels and the calendar year variable was fixed at the
1978 level. While this assumption is impossible to corroborate, it is reasonable, and the calendar year
variable provides a means of dealing with general work practice improvements that are otherwise
difficult to quantify.
For the extended follow-up through 1998 (Steenland et al., 2004; Steenland et al., 2003). the
exposure assessment conducted in 1987 was not updated; however, additional information on the date
last employed was obtained for those workers still employed and exposed at the time of the original
work history collection for the plants still using EtO (25% of the cohort). It was then assumed that
exposure for these workers continued until the date of last employment and that their exposure level
stayed the same as that in their last job held at the time of the original data collection. When the
investigators compared cumulative exposures estimated with and without the extended work histories,
they found little difference because exposure levels were very low by the mid-1980s and, therefore, had
little impact on cumulative exposure (Steenland et al., 2004; Steenland et al.. 2003).
Estimated exposure levels in the NIOSH sterilizer plants are in the same rough range as those
reported for the Swedish sterilizer plants (Hagmaret al.. 1991). but it is difficult to compare more
specifically because it is a wide range of exposure levels across different plants, time periods, and jobs.
At the high end, two of the NIOSH plants had jobs with historical exposure levels as high as those
estimated for the Mikoczy et al. (2011) study for the earliest time periods in that study (Hagmar et al.,
1991), but most of the NIOSH plants had lower estimated exposure levels (see Table J-4 of Appendix J).
In summary, the EPA has relatively high confidence in the NIOSH exposure assessment because
of the use of a well-validated regression model for developing exposure estimates for earlier time
periods and jobs for which measurements were not available. Nonetheless, errors in retrospective
exposure assignments are inevitable, and exposure estimation is a primary source of uncertainty in the
unit risk estimates. Thus, the unit risk estimates based on the NIOSH study could over-predict or
under-predict the true risks to an unknown extent.
4.1.4.2.2. Lymphoid cancer mortality analyses
Grouping of lymphohematopoietic cancer subtypes
With respect to the lymphohematopoietic cancer response, it is not clear exactly which
lymphohematopoietic cancer subtypes are related to EtO exposure, so analyses were done for both
lymphoid cancers and all lymphohematopoietic cancers (Steenland et al., 2004). The associations
observed for all lymphohematopoietic cancers was largely driven by the lymphoid cancer responses, and
4-65
-------
biologically, there is stronger support for an etiologic role for EtO in the development of the more
closely related lymphoid cancers than in the development of the more diverse cancers in the aggregate
all lymphohematopoietic cancer grouping; thus, the lymphoid cancer analysis is the preferred analysis
for the lymphohematopoietic cancers. Unit risk estimates for all lymphohematopoietic cancer mortality
and incidence are roughly 100 and 40% greater, respectively, than those for the lymphoid cancer.
Exposure-response modeling
Modeling the exposure-response relationship for lymphoid cancer is limited by the small number
of cases (n = 53) and complicated by the supralinearity of the data (i.e., the response rises relatively
rapidly and then plateaus). As discussed in Section 4.1.1.2, after considering multiple models, the EPA
selected the two-piece linear spline model with the knot at 1,600 ppm x days (cumulative exposure with
a 15-year lag). This model was selected based on multiple objectives, such as less reliance on AIC,
prioritization of models with more local fit in the low-exposure region, and weighing of biological and
statistical considerations (see Section 4.1.1.2). Although the selected two-piece spline model was
considered to be a reasonable approach for reflecting the exposure-response relationship at the lower end
of the exposure range, which is of primary importance for the derivation of unit risk estimates, there is
uncertainty regarding the exposure-response model. However, the model uncertainty is not as great as
might be inferred from the large range of unit risk estimates derived from the various models that were
investigated (see Table 4-7). Models with better global fits than the selected model (based on AIC;
Table 4-6), such as the linear and log-linear log cumulative exposure models and the two-piece spline
models with knots at 100 ppm x days, had much steeper low-exposure slopes (see Figure 4-3) and
correspondingly higher unit risk estimates (at a minimum, seven times that of the selected model) (see
Table 4-7); however, those models do not provide a good reflection of the exposure-response
relationship in the low-exposure range. Conversely, the linear (ERR) model and the (sublinear) Cox
regression model had much shallower low-exposure slopes (see Figure 4-3), providing poor local fits in
that region as well as poor global fits (see Table 4-6).
Of the three models that appeared to have the best potential to provide a good fit to the
low-exposure region (see Figure 4-3), the selected model had the highest unit risk estimate—about 2.3
times the unit risk estimate from the two-piece log-linear spline model with the knot at
1,600 ppm x days and 5.4 times the unit risk estimate from the linear regression of the categorical
results. The ECois are similar between the two two-piece spline models with the knot at
1,600 ppm x days, with the one from the log-linear spline model just 8% greater than that from the
selected linear spline model; however, the LECois are more divergent (2.3 times) (see Table 4-7). This
suggests that some of the difference in the unit risk estimates between the two spline models is more
related to parameter estimate uncertainty for the linear spline model than to model uncertainty.
4-66
-------
An inherent uncertainty in the two-piece spline models is in the selection of the knot, and the
location of the knot is critical in defining the low-exposure slope. The model likelihood was used to
provide a statistical basis for knot selection; although, as shown in Figure D-14 of Appendix D, the
likelihood did not generally change appreciably over a range of possible knots. Ultimately, the
two-piece linear spline model with the knot at 1,600 ppm x days was selected, based on multiple
objectives (see Section 4.1.1.2), although this was not the model with the knot at the maximum
likelihood (lowest AIC) but rather one with the knot at a local maximum likelihood. The two-piece
linear spline with the knot at the maximum likelihood (100 ppm x days) yielded a unit risk estimate
about 20 times that of the same model with the selected knot; however, the former model was rejected
for not providing a good fit to the low-exposure region. Moreover, the difference in AIC between the
two models is an insubstantial 0.7 units. Sensitivity analyses revealed that knots of ± 1,000 ppm x days
from the selected knot of 1,600 ppm x days produced unit risk estimates that were about three times
greater than that of the selected model for the knot at 600 ppm x days and about 50% lower than that of
the selected model for the knot at 2,600 ppm x days (see Section D.3.6 of Appendix D).
Estimation of upper bounds
Another area of uncertainty in the unit risk estimate pertaining to the exposure-response
modeling is the estimation of the 95% (one-sided) upper bound for the selected model. According to
Langholz and Richardson (2010). the distribution of estimated parameters in nonlog-linear models
(hazard functions), such as the linear spline model, is often not symmetrical (because beta is constrained
so that the hazard cannot be less than 0) and profile likelihood confidence intervals are recommended as
being more accurate than Wald-type intervals. Using a Wald approach yields a unit risk estimate about
40% lower than the preferred profile likelihood-based estimate (see Section D.3.8 of Appendix D).
Exposure timing
A further area of uncertainty related to the exposure-response modeling is the lag period. The
best-fitting models presented by Steenland et al. (2004) for lymphohematopoietic cancer mortality had a
15-year lag (lag periods of 0, 5, 10, 15, and 20 years were considered). A 15-year lag period means that
exposures in the 15 years prior to death or the end of follow-up are not taken into account. After
revisiting the issue of lag selection for the lymphoid cancer mortality data and considering model fit for
cumulative exposure with different lag periods across a larger number of models than was previously
evaluated with different lags, the EPA again selected 15 years as the lag period to use for the
exposure-response analyses (see Section D.3.2 of Appendix D). Sensitivity of the results to choice of
lag period is examined in Section D.3.5 of Appendix D. In brief, for the two-piece linear spline model
with the knot at 1,600 ppm x days, the unit risk estimates for different lag periods (0, 5, 10, and
20 years) ranged from about 48% less than (10-year lag) to about 190% greater than (i.e., 2.9 times) (no
4-67
-------
lag) the estimate for the selected model (15-year lag). These alternative lags, however, all resulted in
models with poor fits to the exposure-response data (p-values of 0.12, 0.23, 0.21, and 0.35 for inclusion
of the exposure terms for lags of 0, 5, 10, and 20 years, respectively), with the possible exception of the
untagged model, which was considered less biologically realistic than the lagged models.
In addition, the analyses of the NIOSH investigators indicate that the regression coefficient for
cumulative exposure might have decreased with increasing follow-up, suggesting that the higher
exposure levels encountered by the workers in the more distant past might be having less of an impact
on more recent risk. The regression coefficient for lymphoid cancers was 1.2 x 10"5 per ppm x day, for
both sexes with a 10-year lag, in the 1987 follow-up (Stayner et al.. 1993) versus 4.7 x 10"6 per
ppm x day, for both sexes with a 15-year lag, in the 1998 follow-up (see Steenland reanalyses in
Appendix D). A similar decrease was found in the regression coefficient for cumulative exposure for all
lymphohematopoietic cancers. The life-table analysis used in this dose-response assessment assumes
exposure accrues over the full lifetime for the cumulative exposure metric. If, in fact, exposures in the
distant past cease to have a meaningful impact on the risk of lymphohematopoietic cancers, this
approach would tend to overestimate the unit risk. Thus, a comparison analysis was conducted to
evaluate the impact of ignoring exposures over 55 years in the past in the life-table analysis. The actual
value of such a cut point, if warranted, is unknown. A value less than 55 years might not be appropriate
because exposures for some of the workers began in 1943, so any diminution of potency for past
exposures occurring since 1943 is already reflected in the regression coefficient with follow-up through
1998, at least for those workers. The comparison analysis for lymphoid cancer yielded an LECoi of
2.62 x 10"3 ppm and a unit risk estimate of 3.82 per ppm, which is about 27% less than the estimate
obtained from the unrestricted life-table analysis. Because the appropriate cut point for excluding past
exposures, if any, is unknown, the estimate from the full life-table analysis is preferred. In any event,
the preferred estimate is not appreciably different from the estimate from the analysis which considered
only the most recent 55 years of exposure in the life-table analysis.
Exposure metrics
In general, the ideal dose metric reflects the biologically relevant tissue dose of the active
compound over time. For EtO and lymphoid cancer, the ideal dose metric is unknown. Several
surrogate dose metrics (cumulative exposure, duration of exposure, maximum [8-hour TWA] exposure,
and average exposure) were analyzed by Steenland et al. (2004). and cumulative exposure was the best
predictor of mortality from lymphohematopoietic cancers. Cumulative exposure is considered a good
measure of total exposure because it integrates exposure (levels) over time.
4-68
-------
Confounding factors
The important potential modifying/confounding factors of age, sex, race, and calendar time were
taken into account in the analysis, and the plants included in this cohort were specifically selected for the
absence of any known confounding exposures (Stavner et al.. 1993).
4.1.4.2.3. Breast cancer mortality analyses
With respect to the breast cancer mortality response (Steenland et al., 2004), the
exposure-response modeling was based on 103 deaths. As for the lymphohematopoietic cancer
responses, the exposure-response modeling is complicated by the supralinearity of the data. As
discussed in Section 4.1.2.2, after considering multiple models, the EPA selected the two-piece
log-linear spline model with the knot at 700 ppm x days (cumulative exposure with a 20-year lag). This
model was selected based on multiple objectives, such as prioritization of models fit to the
individual-level data and with more local fit in the low-exposure region. The unit risk estimate from this
model is about four times the estimate from the linear regression model of the categorical results. The
Cox regression models with cumulative and log cumulative exposure indicated unit risk estimates
substantially lower and higher, respectively, than the estimate from the selected model, but those models
provided poor fits to the lower-exposure region.
For the lag period, the best-fitting model had a lag of 20 years, which was the longest lag period
investigated. This is a commonly used lag period for solid tumors, which typically have longer latency
periods than lymphohematopoietic cancers. The Steenland et al. (2004) analysis took into account age,
race, and calendar time. Other risk factors for breast cancer could not be included in the mortality
analysis, but many of these factors were considered in the breast cancer incidence study (Steenland et
al., 2003), and the preferred breast cancer risk estimates are based on the breast cancer incidence data.
As discussed below, however, inclusion of these breast cancer risk factors had little impact on the unit
risk estimate for breast cancer incidence, suggesting that these factors are not confounding or modifying
the exposure-response relationship for breast cancer.
The unit risk estimate for breast cancer mortality is slightly higher than the unit risk estimate
derived for breast cancer incidence (see Section 4.1.2.3), which is contrary to expectations. Confidence
is higher in the breast cancer incidence estimate, suggesting that the mortality estimate may be an
overestimate of the unit risk for breast cancer mortality.
4.1.4.2.4. Breast cancer incidence analyses
Cohort selection
Steenland et al. (2003) conducted an incidence study for breast cancer; therefore, it was not
necessary to calculate unit risk estimates for breast cancer incidence indirectly from the mortality data as
was done for the lymphohematopoietic cancers. Further advantages to using the results from the
4-69
-------
incidence study are that more cases were available for the exposure-response modeling (319 cases), and
that the investigators were able to include data on potential confounders in the modeling for the
subcohort with interviews (233 cases). The results for the subcohort with interviews are used for the
primary breast cancer unit risk calculations because, in addition to including the data on potential
confounders, the subcohort is considered to have full ascertainment of the breast cancer cases, whereas
the full cohort for the incidence study has incomplete case ascertainment, as illustrated by the fact that
death certificates were the only source of case ascertainment for 14% of the cases. Complete interviews
were available for only 68% of the 7,576 women in the full incidence cohort, and thus, some potential
exists for participation selection bias in the subcohort. There is, however, no basis for considering
participation to be associated with breast cancer orEtO exposure, and the major reason for
nonparticipation was a failure to locate (22% of full incidence cohort) and not lack of response (3% of
cohort) or refusal to participate (7% of cohort). Unit risk estimates based on the full cohort were
calculated for comparison with the subcohort estimates using the original linear regression analyses of
the categorical results (U.S. EPA. 2006a). [Because of the preference for the subcohort with interviews,
for which there was SAB concurrence (SAB. 2015). no further modeling of the continuous exposure
data for the full cohort was done, as was done for the subcohort.] The unit risk estimate from the linear
regression analysis of the categorical results based on the full cohort was about 40% lower than the
corresponding estimate from the subcohort (U.S. EPA. 2006a).
Exposure-response modeling
As discussed in Section 4.1.2.3, after considering multiple models for the breast cancer incidence
data in the subcohort with interviews, the EPA selected the two-piece linear spline model with the knot
at 5,750 ppm x days (cumulative exposure with a 15-year lag). The use of a two-piece spline model is
not intended to imply that an abrupt change in biological response occurs at the knot but, rather, to allow
description of an exposure-response relationship in which the slope of the relationship differs notably in
the low-exposure versus high-exposure regions. The two-piece model is used here primarily for its
representation of the low-exposure data, which are key for the derivation of unit risk estimates. The
main uncertainty in the two-piece spline models is in the selection of the knot, and the location of the
knot is critical in defining the low-exposure slope. The model likelihood was used to provide a
statistical basis for knot selection; although, as shown in Appendix D (see Figure D-l), the likelihood
did not change appreciably over a range of possible knots. Thus, because of the importance of knot
selection, a sensitivity analysis was done to examine the impacts of selecting different knots (see
Section D.1.7 of Appendix D). For the sensitivity analysis, the two-piece linear spline model was run
with knots 1,000 ppm x days below and above the selected knot. For breast cancer incidence, this
sensitivity analysis revealed little difference in the unit risk estimates, yielding estimates about 14%
4-70
-------
higher than and 11% lower than, respectively, the estimate obtained with the selected knot of
5,750 ppm x days.
As can be seen in Figure 4-7 and Table 4-15, there is substantial variation in the low-exposure
slopes and in the unit risk estimates from the different models considered. With the exception of the
cumulative exposure and log cumulative exposure Cox regression models, which had both the poorest
global fits (in terms of AIC) and local fits to the lower-exposure range of the models considered, the unit
risk estimates have about a 35-fold range. At the lower end of that range, the linear cumulative exposure
model is considered a lower bound on the likely low-exposure slope, given the overall supralinearity of
the exposure-response data (as indicated by the apparent plateauing with the highest exposure group and
evidenced by the strong influence of the top 5% of cumulative exposures on dampening the slope of the
[cumulative exposure] Cox regression model [see Section D. 1.3.1 and Figure D-4 of Appendix D]). At
the upper end of the range, the square-root transformation models have less dramatic supralinearity than
the log cumulative exposure model, and although the low-exposure curvature is largely imposed by the
data at higher exposures, it is difficult to know to what extent, if any, the low-exposure slope is
over-estimated by those models at the ECoi. The remaining models, the two two-piece spline models
and the linear regression of the categorical results, gave similar results, spanning a 1.6-fold range. The
selected model, which has the highest unit risk estimate in this narrow range, is the best-fitting (in terms
of AIC) of the continuous exposure models represented in the range.
Although the reason for the observed supralinear exposure-response relationship is unknown, it
is worth noting that the results of the Swedish sterilizer worker study reported by Mikoczy et al. (201 1).
although limited and consistent with a higher unit risk estimate for breast cancer incidence than that
obtained from the NIOSH study results, support the general supralinear exposure-response relationship
observed in the NIOSH study (see Section J.2.2 of Appendix J).
Estimation of upper bounds
Another area of uncertainty in the unit risk estimate pertaining to the exposure-response
modeling is the estimation of the 95% (one-sided) upper bound for the selected model. According to
Langholz and Richardson (2010). the distribution of estimated parameters in nonlog-linear models
(hazard functions), such as the linear spline model, is often not symmetrical (because beta is constrained
so that the hazard cannot be less than 0) and profile likelihood confidence intervals are recommended as
being more accurate than Wald-type intervals. Using a Wald approach yields a unit risk estimate about
3% lower than the preferred profile-likelihood-based estimate (see Section D. 1.10 of Appendix D).
Lag time
A further area of uncertainty related to the exposure-response modeling is the lag period. The
best-fitting models presented by Steenland et al. (2003) for breast cancer incidence generally had a
4-71
-------
15-year lag (lag periods of 0, 5, 10, 15, and 20 years were considered). After revisiting the issue of lag
selection for the breast cancer incidence data and considering model fit for cumulative exposure with
different lag periods across a larger number of models than was previously evaluated with different lags,
the EPA again selected 15 years as the lag period to use for the exposure-response analyses (see Section
D. 1.2 of Appendix D). Sensitivity of the unit risk estimates to choice of lag period is examined in
Section D. 1.6 of Appendix D. In brief, for the two-piece linear spline model with the knot at
5,750 ppm x days, the unit risk estimates for different lag periods (0, 5, 10, 15, and 20 years) ranged
from about 35% less than (10-year lag) to about 21% greater than (20-year lag) the estimate for the
selected model (15-year lag). Of these specific models, the model with the 20-year lag was a
better-fitting model than the selected model, based on log likelihood. The models for lags of 0, 5, and
10 years had ^-values > 0.05 for inclusion of the exposure terms (0.11, 0.057, and 0.080, respectively).
Exposure metric
With respect to dose metrics for breast cancer incidence, models using duration of exposure
provided better model fits than those using cumulative exposure (Steenland et al.. 2003). However,
duration is less useful for estimating unit risks and the cumulative exposure models also provided
statistically significant fits to the data; thus, the cumulative exposure metric was used for the quantitative
risk estimates. Cumulative exposure is considered a good measure of total exposure because it
integrates exposure (levels) over time. Models using peak (highest one-time exposure) or average
exposure (cumulative exposure divided by duration) did not fit as well.
Confounding factors
Regarding potential confounders/modifying factors, analyses for the full cohort and the
subcohort with interviews were adjusted for age, race, and calendar time. In addition, exposures to other
chemicals in these plants were reportedly minimal. For the subcohort with interviews, a number of
specific breast cancer risk factors were investigated, including body mass index, breast cancer in a
first-degree relative, parity, age at menopause, age at menarche, socioeconomic status, and diet;
however, only parity and breast cancer in a first-degree relative were determined to be important
predictors of breast cancer and were included in the final models. Exclusion of these covariates
produced very little difference in the unit risk estimates—excluding parity and both parity and breast
cancer in a first-degree relative changed the unit risk estimate by only about 1%. Presumably, these
covariates were not associated with exposure, and thus, although they were associated with breast cancer
incidence, they did not confound the exposure-response analyses.
4-72
-------
Endpoint definition
An area of uncertainty in the life-table analysis for breast cancer incidence pertains to the rates
used for the cause-specific background rate. The regression coefficients presented by Steenland et al.
(2003) represent invasive and in situ cases combined, where 6% of the cases are in situ, and the
preferred unit risk estimates in this assessment are calculated similarly using background rates for
invasive and in situ cases combined. The regression coefficients for invasive and in situ cases combined
should be good approximations for regression coefficients for invasive cases alone; however, it is
uncertain how well they reflect the exposure-response relationships for in situ cases alone. Diagnosed
cases of in situ breast cancer would presumably be remedied and not progress to invasive breast cancer,
so double-counting is unlikely to be a significant problem. Carcinoma in situ is a risk factor for invasive
breast cancer; however, this observation is most likely explained by the fact that these two types of
breast cancer have other breast cancer risk factors in common, some of which have been considered in
the subcohort analysis. One might hypothesize that EtO exposure could cause a more rapid progression
to invasive tumors; however, there is no specific evidence that this occurs. On the other hand, there is
some indication that in situ cases in the incidence study might have been diagnosed at relatively low
rates in comparison to the invasive cases. Steenland et al. (2003) reported that 6% of the cases in their
study are in situ; according to the National Cancer Institute, however, ductal carcinoma in situ accounted
for about 18% of newly diagnosed cases of breast cancer in 1998 (NCI. 2004).
There are several possible explanations for this difference. One is that it reflects differences in
diagnosis with calendar time because the rate of diagnosis of carcinoma in situ has increased over time
with increased use of mammography. Another is that the difference is partially a reflection of the age
distribution in the cohort because the proportion of new cases diagnosed as carcinoma in situ varies by
age. A third possible explanation is that the low proportion of in situ cases is at least partially a
consequence of underascertainment of cases because in situ cases will not be reported on death
certificates, although, even if all 20 in situ cases were in the subcohort with interviews, that would still
be only 8.6% of the cases. In any event, this is a relatively minor source of uncertainty, and a
comparison of the unit risk estimates using invasive + in situ breast cancer background rates and
invasive-only background rates found that the estimate using only invasive breast cancer background
rates was about 20% lower than the preferred estimate based on the invasive + in situ background rates.
Comparison with breast cancer mortality estimate
The unit risk estimate for breast cancer incidence is slightly lower than the unit risk estimate
derived for breast cancer mortality, which is contrary to expectations. It is likely that the mortality
estimate is an overestimate of the unit risk for breast cancer mortality. Confidence is higher in the breast
cancer incidence estimate because it is based on more cases (233 cases versus 103 deaths), especially in
4-73
-------
the lower-exposure region (e.g., 68 exposed cases with cumulative exposures < 2,207 ppm x days versus
33 exposed deaths with cumulative exposures < 2,780 ppm x days), and a better-fitting model.
4.1.4.3. Uncertainties Associated with the Total Cancer Unit Risk Estimate
To obtain the risk estimate for total cancer risk (6.1 per ppm, 6.1 x 10"3 per ppb, or 3.3 x 10"3 per
(j,g/m3), the preferred estimates for lymphoid cancer incidence and breast cancer incidence were
combined (see Section 4.1.3). While there are uncertainties in the approach used to combine the
individual estimates, the resulting unit risk estimate is appropriately bounded in the roughly 1.1-fold
range between the sum of the unit risk estimate for lymphoid cancer plus the 0.01/ECoi estimate for
breast cancer (5.98 per ppm), which provides a lower bound for the upper bound on the sum of the
individual MLEs of risk and the sum of the individual 95% UCLs. Thus, any inaccuracy in the total
cancer unit risk estimate resulting from the approach used to combine risk estimates across cancer types
is relatively minor. Because the breast cancer component of the total cancer risk estimate applies only to
females, the total cancer risk estimate is expected to overestimate the cancer risk to males somewhat (the
preferred unit risk estimate for lymphoid cancer alone was 5.26 per ppm [or 2.87 x 10"3 per (J,g/m3],
which is about 87% of the total cancer risk estimate).
4.1.4.4. Conclusion Regarding Uncertainties
As discussed above, most of the sources of uncertainty inherent in the unit risk estimates (e.g.,
uncertainty about the lag times) have little potential quantitative impact on the values of the estimates.
The primary sources of uncertainty—exposure estimation, model uncertainty, and low-dose
extrapolation—have potentially larger impacts. Retrospective exposure estimation is an inevitable
source of uncertainty in this type of epidemiology study; however, the NIOSH investigators put
extensive effort into addressing this issue by developing a state-of-the-art regression model to estimate
unknown historical exposure levels using variables, such as sterilizer size, for which historical data were
available (see Section 4.1 and Section A.2.8 of Appendix A). Uncertainty pertaining to the
exposure-response models is a particular concern with these data sets because the supralinear nature of
the exposure-response relationships makes it more difficult to estimate the low-exposure slopes.
Nonetheless, the two-piece spline models used in this assessment are considered to provide a reasonable
basis for the derivation of unit risk estimates, especially for breast cancer incidence, for which more data
were available in the lower-exposure range. Low-dose extrapolation is another inevitable source of
uncertainty in the derivation of unit risk estimates; however, for EtO, the use of linear low-exposure
extrapolation is strongly supported by the conclusion that EtO carcinogenicity has a mutagenic mode of
action (see Section 3.4.3).
In summary, despite these uncertainties, the inhalation cancer unit risk estimate of 6.1 per ppm
(or 3.3 x 10"3 per (J,g/m3) for the total cancer risk from lymphoid cancer incidence and female breast
4-74
-------
cancer incidence has the advantages of being based on human data from a large, high-quality
epidemiologic study with individual exposure estimates for each worker. Furthermore, the breast cancer
component of the risk estimate, which is almost 25% of the total cancer risk, is based on a substantial
number of incident cases (233 total, the vast majority of which were in the exposure range below the
knot of 5,750 ppm x days [at least 102 exposed and 164 total cases; see Table D-l of Appendix D]).
Thus, there is relatively high confidence in the total cancer unit risk estimate.
4.1.5. Summary of Unit Risk Estimates Derived from Human Data
Under the common assumption that RR is independent of age (which was validated over the age
range of the cohort), an inhalation unit risk estimate for lymphoid cancer incidence of 5.26 per ppm (or
5.26 x 10"3 per ppb; 2.87 x 10-3 per (j,g/m3) was calculated using a life-table analysis and a two-piece
linear spline model for excess lymphoid cancer mortality from a high-quality occupational epidemiology
study. Similarly, an inhalation unit risk estimate for female breast cancer incidence of 1.48 per ppm (or
1.48 x 10"3 per ppb; 8.09 x io_4 per (j,g/m3) was calculated using a life-table analysis and two-piece
linear spline model for excess breast cancer incidence from the same high-quality occupational
epidemiology study. The two-piece linear spline analysis was undertaken to address the supralinearity
of the exposure-response data in the two data sets. Low-dose linear extrapolation was used, as
warranted by the clear mutagenicity of EtO. An ECoi estimate of 0.0049 ppm, a LECoi estimate of
0.0017 ppm, and a unit risk estimate of 6.1 per ppm (or 6.1 x 10"3 per ppb; 3.3 x 10"3 per (J,g/m3) were
obtained for the total cancer risk combined across both cancer types. Despite the uncertainties discussed
above, this inhalation unit risk estimate has the advantages of being based on human data from a
high-quality epidemiologic study with individual exposure estimates for each worker.
In the absence of data on early-life susceptibility, the EPA's Supplemental Guidance (U. S. EPA.
2005b) recommends that increased early-life susceptibility be assumed for carcinogens with a mutagenic
mode of action, and the conclusion was made in Section 3.4 that the weight of evidence supports a
mutagenic mode of action for EtO. Thus, in accordance with the Supplemental Guidance, the alternate
assumption of increased early-life susceptibility is preferred as the basis for risk estimates in this
assessment, and risk estimates derived under this preferred assumption are presented in Section 4.4.
Other than the use of the alternate assumption about early-life susceptibility, the approach used to derive
the estimates presented in Section 4.4 is identical to the approach used for the estimates derived here in
Section 4.1, and the comparisons made between various options and the issues and uncertainties
discussed here in Section 4.1 are applicable to the estimates derived in Section 4.4.
4-75
-------
4.2. INHALATION UNIT RISK DERIVED FROM LABORATORY ANIMAL DATA
4.2.1. Overall Approach
Lifetime animal cancer bioassays of inhaled EtO have been carried out in three laboratories, as
described in Section 3.2. The data from these reports are presented in Tables 3-3 through 3-5. These
studies have also been reviewed by the I ARC (1994b) and Health Canada (2001). Health Canada
calculated the EDos for each data set using the benchmark dose methodology. The Ethylene Oxide
Industry Council report (EPIC. 2001) tabulated only lymphatic tumors because they constituted the
predominant risk.
The overall approach in this derivation is to find a unit risk for each of the bioassays—keeping
data on males and females separate—from data on the incidence of all tumor types and then to use the
maximum of these values as the summary measure of the unit risk from animal studies (i.e., the unit risk
represents the most sensitive species and sex). The unit risk for the animals in these bioassays is
converted to a unit risk in humans by first determining the continuous exposures in humans that are
equivalent to the rodent bioassay exposures and then by assuming that the lifetime incidence in humans
is equivalent to lifetime incidence in rodents, as is commonly accepted in interspecies risk
extrapolations. For cross-species scaling of exposure levels (see Section 4.2.2 below), an assumption of
ppm equivalence is used; thus, no interspecies conversion is needed for the exposure concentrations.
Bioassay exposure levels are adjusted to equivalent continuous exposures by multiplying by (hours of
exposure/24 hours) and by (5/7) for the number of days exposed per week. The unit risk in humans (risk
per unit air concentration) is then assumed to be numerically equal to that in rodents (after adjustment to
continuous exposures); the calculations from the rodent bioassay data are shown in Tables 3-3 through
3-5.
4.2.2. Cross-Species Scaling
In the absence of chemical-specific information, the EPA's 1994 inhalation dosimetry methods
(U.S. EPA. 1994) provide standard methods and default scaling factors for cross-species scaling. Under
the EPA's methodology, EtO would be considered a Category 2 gas because it is reactive and water
soluble and has clear systemic distribution and effects. Dosimetry equations for Category 2 gases are
undergoing EPA reevaluation and are not being used at this time. For cross-species scaling of
extrarespiratory effects, current practice is to treat Category 2 gases as Category 3 gases. For
Category 3 gases, ppm equivalence is assumed (i.e., responses across species are equivalent on a ppm
exposure basis), unless the air:blood partition coefficient for the experimental species is less than the
coefficient for humans [U.S. EPA (1994). p. 4-61], In the case of EtO, measured air:blood partition
coefficients are 78 in mice (Fennel! and Brown, 2001), 64 in rats (Krishnan et al., 1992), and 61 in
humans (Csanadv et al., 2000); thus, ppm equivalence for cross-species scaling to humans can be
assumed for extrarespiratory effects observed in mice and rats. The assumption of ppm equivalence is
4-76
-------
further supported by the PBPK modeling of Fennel! and Brown (2001). who reported that simulated
blood AUCs for EtO after 6 hours of exposure to concentrations between 1 ppm and 100 ppm were
similar for mice, rats, and humans and were linearly related to the exposure concentration (see
Section 3.3.1 and Figure 3-2). This modeling was validated against measured blood EtO concentrations
for rodents and humans. For Category 2 gases with respiratory effects, there is no clear guidance on an
interim approach. One suggested approach is to do cross-species scaling using both Category 1 and
Category 3 gas equations and then decide which is most appropriate. In this document, the preferred
approach was to assume ppm equivalence was also valid for the lung tumors in mice because of the clear
systemic distribution of EtO (e.g., see Section 3.1). Treating EtO as a Category 1 gas for cross-species
scaling of the lung tumors would presume that the lung tumors are arising only from the immediate and
direct action of EtO as it comes into first contact with the lung. In fact, some of the EtO dose
contributing to lung tumors is likely attributable to recirculation of systemic EtO through the lung.
If, as a bounding exercise, EtO was treated as a Category 1 gas for the cross-species scaling of
the lung tumor response, equations for estimating the RGDRpu (i.e., the regional gas dose ratio for the
pulmonary region), which acts as an adjustment factor for estimating human equivalent exposure
concentrations from laboratory animal exposure concentrations (adjusted for continuous exposure), are
presented in the EPA's 1994 inhalation dosimetry methods [U.S. EPA (1994). pp. 4-49 to 4-51], These
equations rely on parameters describing mass transport of the gas (EtO) in the extrathoracic and
tracheobronchial regions for both the laboratory animal species (mouse) and humans. Without
experimental data for these parameters, it seems reasonable to estimate RGDRpu using a simplified
equation and the adjusted alveolar ventilation rates of Fennel! and Brown (2001). Fennel! and Brown
(2001) adjusted the alveolar ventilation rates to reflect limited pulmonary uptake of EtO, a phenomenon
commonly observed for highly water-soluble gases (J oh an son and Filser. 1992). The adjusted
ventilation rates were then used by Fennel! and Brown (2001) in their PBPK modeling simulations, and
good fits to blood concentration data were reported for both the mouse and human models. In this
document, the adjusted alveolar ventilation rates were used to estimate the RGDRpu as follows:
RGDRpu = (RGDpu)m/ (RGDpu)h = (Qaiv/ S Apu)m/(Qaiv/ S Apu)h, (4-4)
where:
RGDpu = regional gas dose to the pulmonary region,
Qaiv = (adjusted) alveolar ventilation rate,
SApu = surface area of the pulmonary region, and
the subscripts "m" and "h" denote mouse and human values.
4-77
-------
Then, using adjusted alveolar ventilation rates from Fennel! and Brown (2001) and surface area values
from the EPA rU.S. EPA (1994). p. 4-26],
RGDRpu = [(0.78 L/h)/(0.05 m2)]/[(255 L/h)/(54.0 m2)] = 3.3.
(4-5)
Using this value for the RGDRpu would increase the human equivalent concentration about threefold,
resulting in a decreased risk for lung tumors of about threefold, as a lower bound. The true value of the
RGDRpu is expected to be between 1 and 3, and any adjustment to the lung tumor risks would still be
expected to result in unit risk estimates roughly within the range of the rodent unit risk estimates derived
later in Section 4.2 under the assumption of ppm equivalence.
4.2.3. Dose-Response Modeling Methods
In this document the following steps were used:
1. Extract the incidence data presented in the original studies. In order to crudely adjust for
early mortality in the analysis of the NTP (1987) data, the incidence data have been corrected
for a specific tumor type by eliminating the animals that died prior to the occurrence of the
first tumor or prior to 52 weeks, whichever was earlier. It was not possible to make this
adjustment with the other studies where data on individual animals were not available. With
these exceptions, the tumor incidence data in Tables 3-3 through 3-5 match the original data.
2. Fit the multistage model to the dose-response data using the Tox Risk program. The
likelihood-ratio test was used to determine the lowest value of the multistage polynomial
degree that provided the best fit to the data while requiring selection of the most
parsimonious model. In this procedure, if a good fit to the data in the neighborhood of the
POD is not obtained with the multistage model because of a nonmonotonic reduction in risk
at the highest dose tested (as sometimes occurs when there is early mortality from other
causes), that data point is eliminated and the model is fit again to the remaining data. Such a
deletion was found necessary in two cases (mammary tumors in the NTP study and
mononuclear cell leukemia in the Lynch study). The goodness-of-fit measures for the
dose-response curves and the parameters derived from them are shown in Appendix G.
In the NTP bioassay, where the individual animal data were available, a time-to-tumor
analysis was undertaken to account for early mortality. The general model used in this
analysis is the multistage Weibull model:
P(d,t) = 1 - exp[-(q0 + qid + q2d2 + ... + qkdk) x (t - to)2],
(4-6)
4-78
-------
where P(d,t) represents the probability of a tumor by age t (in bioassay weeks) for dose d
(i.e., human equivalent exposure), and the parameter ranges are restricted as follows: z > 1,
to > 0, and qi > 0 for i = 0, 1, ..., k. The parameter to represents the time between when a
potentially fatal tumor becomes observable and when it causes death. The analyses were
conducted using the computer software ToxRisk version 3.5, which is based on methods
developed by Krewski et al. (1983). Parameters are estimated in Tox Risk using the method
of maximum likelihood.
Tumor types can be categorized by tumor context as either fatal or incidental. Incidental
tumors are those tumors thought not to have caused the death of an animal, whereas fatal
tumors are thought to have resulted in animal death. Tumors at all sites were treated as
incidental (although it was recognized that this may not have been the case, the experimental
data are not detailed enough to conclude otherwise). The parameter to was set equal to 0
because there were insufficient data to reliably estimate it.
The likelihood-ratio test was used to determine the lowest value of the multistage polynomial
degree k that provided the best fit to the data while requiring selection of the most
parsimonious model. The one-stage Weibull (i.e., k = 1) was determined to be the most
optimal value for all the tumor types analyzed.
3. Select the POD and calculate the unit riskfor each tumor site. The effective concentration
that causes a 10% extra risk for tumor incidence, ECio, and the 95% lower bound of that
concentration, LECio, are derived from the dose-response model. The LECio is then used as
the POD for a linear low-dose extrapolation, and the unit risk is calculated as 0.1/LECio.
This is the procedure specified in the EPA's Guidelines for Carcinogen Risk Assessment
(U.S. EPA. 2005a) for agents such as EtO that have direct mutagenic activity. See
Section 3.4 for a discussion of the mode of action for EtO. Tables 3-3 through 3-5 present
the unit risk estimates for the individual tumor sites in each bioassay.
4. Develop a unit risk estimate based on the incidence of all tumors combined. This method
assumes that occurrences of tumors at multiple sites are independent, and further, that the
risk estimate for each tumor type is normally distributed. Then, at a given exposure level, the
MLEs of extra risk due to each tumor type are added to obtain the MLE of total cancer risk.
The variances corresponding to each tumor type are added to give the variance associated
with the sum of the MLEs. The one-sided 95% UCL of the MLE for the combined risk is
then calculated as:
95% UCL = MLE + 1,645(SE), (4-7)
where SE is the standard error and is the square root of the summed variance. (Note that as a
precursor to this step, when Tox Risk is used to fit the incidence of a single tumor type, it
provides the MLE and 95% UCL of extra risk at a specific dose. The standard error in the
MLE is determined using the above formula.) The calculation is repeated for a few exposure
4-79
-------
levels, and the exposure yielding a value of 0.1 for the upper bound on extra risk is
determined by interpolation. The unit risk is then the slope of the linear extrapolation from
this POD. The results are given in Table 4-18.
Table 4-18. Upper-bound unit risks (per jig/m3) obtained by combining
tumor sites
Combination method3
NTP (1987)
female mouse
Lvnch et al. (1984c).
Lvnch et al. (1984a)
male rat
Snellings et al. (1984)b
Male rat
Female rat
UCL on sum of risks0
2.71 x 10"5
4.17 x 10"5
2.19 x 10"5
3.37 x 10"5
Sum of unit risksd
4.12 x 10-5
3.66 x 10-5
2.88 x 10-5
3.54 x 10-5
Time-to-tumor analysis and
UCL on sum of risks0
4.55 x 10"5
-
-
-
"Unit risk in these methods is the slope of the straight line extrapolation from a point of departure at the dose
corresponding to a value of 0.1 for the 95% upper confidence bound on total extra risk.
bIncludes data onbrain tumors from the analysis by Carman et al. (1985). See Table 3-3.
°UCL = 95% upper confidence bound. At a given dose, the MLE of the combined extra risk was determined by
summing the MLE of risk due to each tumor type. The variance associated with this value was determined by
summing over the variances due to each tumor type.
dSum of values in last column of Tables 3-1 through 3-3.
4.2.4. Description of Laboratory Animal Studies
NTP (1987) exposed male and female B6C3Fi mice to concentrations of 0, 50, or 100 ppm for
6 hours per day, 5 days per week, for 102 weeks. An elevated incidence of lung carcinomas was found
in males, and elevated lung carcinomas, malignant lymphomas, uterine adenocarcinomas, and mammary
carcinomas were found in females. These data are shown in Table 3-3.
Lynch et al. (Lynch et al.. 1984a; Lynch et al.. 1984c) exposed male F344 rats to 0, 50, or
100 ppm for 7 hours per day, 5 days per week, for 2 years. They found excess incidence of tumors at
three sites: mononuclear cell leukemia in the spleen, testicular peritoneal mesothelioma, and brain
glioma. In this study the survival in the high-dose group (19%) was less than that of controls (49%),
which reduced the incidence of leukemias. In the animals in the high-dose group that survived to the
termination of the experiment, the incidence of leukemias was statistically significantly higher than for
controls (p < 0.01). The incidence data are shown in Table 3-4, uncorrected for the high-dose-group
mortality. If the individual animal data were available to perform the correction, the incidence would be
higher. Therefore, using these data results in an underestimate of risk.
Snellings et al. (1984) exposed male and female F344 rats to 0, 10, 33, or 100 ppm for 6 hours
per day, 5 days per week, for 2 years and described their results for all sites except the brain. In two
subsequent publications for the same study, (Gartnan et al.. 1986. 1985) described the development of
4-80
-------
brain tumors in a different set of F344 rats. The Snellings et al. (1984) publication reported an elevated
incidence of splenic mononuclear cell leukemia and peritoneal mesothelioma in males and an elevated
incidence of splenic mononuclear cell leukemia in females. The mortality was higher in the 100-ppm
groups than the other three groups for both males and females. The incidences in the animals killed
after 24 months in Snellings et al. (1984) are shown in Table 3-5. Table 3-5 also presents the brain
tumor incidence data for male and female rats from the (Garman et al.. 1986. 1985) publications. The
brain tumor incidence was lower than that of the other tumors, particularly the splenic mononuclear cell
leukemias.
4.2.5. Results of Data Analysis of Laboratory Animal Studies
The unit risks calculated from the individual site-sex-bioassay data sets are presented in
Tables 3-3 through 3-5. The highest unit risk of any individual site is 3.23 x 10"5 per (J,g/m3, which is for
mononuclear cell leukemia in the female rats of the Snellings et al. (1984) study.
Table 4-19 presents the results of the time-to-tumor method applied to the individual animals in
the NTP bioassay, compared with the results from the dose group incidence data in Table 3-3. This
comparison was done for each tumor type separately. The time-to-tumor method of analyzing the
individual animals results in generally higher unit risk estimates than does the analysis of dose group
data, as shown in Table 4-19. The ratio is not large (less than 2.2) across the tumor types.44 The results
also show the extent to which a time-to-tumor analysis of individual animal data increases the risk
estimated from data on dose groups. It is expected that if individual animal data were available for the
Lynch et al. (Lynch et al„ 1984a; Lynch et al., 1984c) and Snellings et al. (1984) bioassays, then the
time-to-tumor analysis would also result in higher estimates because both those studies also showed
early mortality in the highest dose group.
The results of combining tumor types are summarized in Table 4-18. The sums of the individual
unit risks tabulated in Tables 3-3 to 3-5 are given in the second row of Table 4-18. Note that, as
expected, they are greater than the unit risks computed from the upper bound on the sum of risks for all
data sets except for the Lynch et al. (Lynch et al., 1984a; Lynch et al„ 1984c) data. The reason for this
exception is not known, but the differences are small. It is likely that the problem arises from the
methodology used to combine the risks across tumor sites.
"In the case of mammary tumors this ratio is actually less than 1. It must be noted that the incidence at the highest dose
(where the incidence was substantially less than at the intermediate dose) was deleted from the analysis of grouped data,
whereas it was retained in the time-to-tumor analysis. Therefore, the comparison for the mammary tumors is not a strictly
valid comparison of methods.
4-81
-------
Table 4-19. Unit risk values from multistage Weibull" time-to-tumor
modeling of mouse tumor incidence in the NTT (1987) study
Tumor type
Unit risk, 0.1/LECio
(per jig/m3)
from time-to-tumor
analysis
Unit risk,
0.1/LECio
(per jig/m3)
(Table 3-3)b
Ratio of unit risks
time-to-tumo r/ groupe d
data
Males
Lung: alveolar/bronchiolar adenoma
and carcinoma
3.01 x 10"5
2.22 x 10"5
1.4
Females
Lung: alveolar/bronchiolar adenoma
and carcinoma
2.40 x 10"5
1.10 x 10"5
2.2
Malignant lymphoma
1.43 x 10-5
7.18 x 10-6
2.0
Uterine carcinoma
6.69 x 10"6
4.33 x 10"6
1.5
Mammary carcinoma
8.69 x 10-6
1.87 x 10-5
0.5
aP(d,t) = 1 - exp[-(q0 + qid + q2d2 + ... + qkdk) x (t - to)z], where d is inhaled ethylene oxide concentration inppm, t
is weeks until death with tumor. In all cases, k= 1 provided the optimal model.
bIncidence data modeled using multistage model without taking time to tumor into account.
In an attempt to be consistent with the new two-step methodology (i.e., modeling in the
observable range to a POD and then doing a linear extrapolation to zero extra risk at zero exposure), the
exposure concentration at which the sum of the independent tumor site risks yielded a 95% upper bound
on 10% extra risk was estimated and used as the POD. Summing risks in this way results in a POD for
the combined tumor risk that is different (lower) than the points of departure for each individual tumor
site risk. Thus, the risk estimate for the sum is not strictly comparable to the individual risks that
constitute it. These tumor-site-specific risks were based on points of departure individually calculated to
correspond with a 10% extra risk. In any event, adding the upper bound risks of individual tumor sites
should overestimate the upper bound of the sum, and the latter is the preferred measure of the total
cancer risk because it avoids the overestimation. However, for the exceptional Lynch et al. (Lynch et
al.. 1984a; Lynch et al.. 1984c) data, the sum of upper bounds, 3.66 x 10"5 per (.tg/m3, is already an
overestimate of the total risk, and this value is preferred over the anomalously high value of 4.17 x 10~5
per [j,g/m3 corresponding to the upper bound on the sum of risks. The latter value is considered to be an
excessive overestimate and is, therefore, not carried over into the summary Table 4-20. For the
Snellings et al. (1984) data sets, the upper confidence bound on the sum of risks is used in the summary
Table 4-20. The results of the sum-of-risks calculations on the NTP bioassay time-to-tumor data are
included in the third row of Table 4-18. The estimate for the NTP female mice is 4.55 x 10"5 per (J,g/m3,
which is higher than the other two measures of total tumor risk in that bioassay. This value is preferable
to the other measures because it uses the individual animal data available for that bioassay.
4-82
-------
Summary of results. The summary of unit risks from the five data sets is shown in Table 4-20.
The data set giving the highest risk (4.55 x 10~5 per (J,g/m3) is the NTP (1987) data on combined tumors
in female mice. The other values are within about a factor of 2 of the highest value.
Table 4-20. Summary of unit risk estimates (per jig/m3) in animal bioassays
Assay
Males
Females
NTP rigs?). B6C3F, mice
3.01 x 10"5a
4.55 x 10"5b
Lvnchetal. (1984c). Lvnchetal. (1984a).
3.66 x 10-5'°
-
F344 rats
Snellines etal. (1984). F344 rats
2.19 x 10"5d
3.37 x 10"5d
"From time-to-tumor analysis of lung adenomas and carcinomas, Table 4-19.
bUpper bound on sum of risks from the time-to-tumor analysis of the NTP data, Table 4-18.
°Sumof (upper bound) unit risks (see text for explanation), Table 4-18.
dUpper bound on sum of risks, Table 4-18.
The largest sources of uncertainty in the rodent-based unit risk estimates for EtO are interspecies
extrapolation and low-dose extrapolation. Although the unit risk estimates from mouse and rat data are
similar (see Table 4-20), the different EtO-associated cancer sites across rodent species demonstrate
species differences that are not understood and illustrate the existence of interspecies uncertainty that
extends to the extrapolation of rodent-based cancer risk estimates to humans. Regarding low-dose
extrapolation, the clear evidence of EtO mutagenicity supports the linear low-exposure extrapolation
that was used (U.S. EPA. 2005a) (see also Section 4.1.4.1). Additional uncertainties arise from the
dose-response modeling of the data in the observable range and the application of the results to
potentially sensitive human populations.
4.3. SUMMARY OF INHALATION UNIT RISK ESTIMATES—NOT ACCOUNTING FOR
ASSUMED INCREASED EARLY-LIFE SUSCEPTIBILITY
For both humans and laboratory animals, tumors occur at multiple sites. In humans, there was a
combination of tumors having lymphohematopoietic, in particular lymphoid, origins in both sexes and
breast cancer in females, and in rodents, lymphohematopoietic tumors, mammary carcinomas, and
tumors of other sites were observed. From human data, an extra cancer unit risk estimate of
2.87 x 10"3 per [j,g/m3 (5.26 per ppm) was calculated for lymphoid cancer incidence, and a unit risk
estimate of 8.09 x 10-4 per [j,g/m3 (1.48 per ppm) was calculated for breast cancer incidence in females.
The total extra cancer unit risk estimate was 3.3 x 10"3 per [j,g/m3 (6.1 per ppm) for both cancer types
combined (ECoi = 0.0049 ppm; LECoi = 0.0017 ppm). Unit risk estimates derived from the three
4-83
-------
chronic rodent bioassays for EtO ranged from 2.2 x 10"5 per [j,g/m3 to 4.6 x 10~5 per (J,g/m3, roughly two
orders of magnitude lower than the estimates based on human data.
The reasons for the species differences in the carcinogenic potency of EtO are unknown. EtO is
efficiently absorbed into the blood and rapidly distributed to all organs and tissues, and tissue
concentrations in mice, rats, and humans exposed to a particular air concentration (below 100 ppm) of
EtO are expected to be approximately equal (see Section 3.3.1). Thus, the species differences in
carcinogenic potency are likely to be the result of toxicodynamic rather than toxicokinetic differences.
EPA notes that there are differences in tissue sensitivity within species, as well, that cannot be attributed
to toxicokinetic differences. For example, EtO forms protein and DNA adducts in tissues throughout the
body (see Sections 3.3.2 and 3.3.3.1), yet only certain tissues are apparent targets for EtO-induced
carcinogenicity. Factors that may account for some of the differences in tissue susceptibility include
differences in DNA configuration (e.g., access to critical genes) and other epigenetic modifications,
DNA repair capacity, signaling pathways, and rates of cellular turnover. Some of these same factors
could be contributing to interspecies differences. In addition, hormonal differences could account for
some interspecies variability, and humans may encounter more co-exposures that could be contributing
to increased risks as co-carcinogens supporting or promoting processes complementary to EtO-induced
carcinogenicity. For example, alcohol consumption, which is widespread in the U.S., could help
promote EtO-initiated breast cancers. Furthermore, although the reasons for the species differences in
potency are unknown, it is not uncommon for differences in potency to be observed across species.
Adequate human data, if available, are considered to provide a more appropriate basis than
rodent data for estimating human risks (U.S. EPA. 2005a). primarily because uncertainties in
extrapolating quantitative risks from rodents to humans are avoided. Although there is a sizeable
difference between the rodent-based and the human-based estimates, the human data are from a large,
high-quality study, with EtO exposure estimates for the individual workers and little reported exposure
to chemicals other than EtO. Therefore, the total extra cancer unit risk estimate of 3.3 x 10"3 per [j,g/m3
(6.1 per ppm) calculated for lymphoid cancers and breast cancer combined based on the human data is
the preferred estimate, not taking assumed increased early-life susceptibility into account (estimates
accounting for assumed increased early-life susceptibility are presented in Section 4.4).
The unit risk estimate is intended to be an upper bound on cancer risk for use with exposures
below the POD (i.e., the LECoi). The unit risk estimate should not generally be used above the POD;
however, in the case of this total extra cancer unit risk, which is based on cancer type-specific unit risk
estimates from two linear models, the estimate should be valid for exposures up to about 0.021 ppm
(38 (J,g/m3), which is the minimum of the limits for the lymphoid cancer unit risk estimate (0.021 ppm;
see Section 4.1.1.2) and the breast cancer unit risk estimate (0.075 ppm; see Section 4.1.2.3) dictated by
the knot locations.
4-84
-------
Because a mutagenic mode of action for EtO carcinogenicity (see Section 3.3.2) is "sufficiently
supported in (laboratory) animals" and "relevant to humans," and as there are no chemical-specific data
to evaluate the differences between adults and children, increased early-life susceptibility should be
assumed, and if there is early-life exposure, age-dependent adjustment factors (ADAFs) should be
applied, as appropriate, in accordance with the EPA's Supplemental Guidance [(U.S. EPA. 2005b); see
Section 4.4 below for more details on the application of ADAFs],
4.4. ADJUSTMENTS FOR POTENTIAL INCREASED EARLY-LIFE SUSCEPTIBILITY
There are no chemical-specific data on age-specific susceptibility to EtO-induced carcinogenesis.
However, there is sufficient weight of evidence to conclude that EtO operates through a mutagenic mode
of action (see Section 3.4.1). In such circumstances (i.e., the absence of chemical-specific data on
age-specific susceptibility but sufficient evidence of a mutagenic mode of action), the EPA's
Supplemental Guidance for Assessing Susceptibility from Early-Life Exposure to Carcinogens (U.S.
EPA. 2005b) recommends the assumption of increased early-life susceptibility and the application of
ADAFs to adjust for this potential increased susceptibility from early-life exposure. See the
Supplemental Guidance for detailed information on the general application of these adjustment factors.
In brief, the Supplemental Guidance establishes default ADAFs for three specific age groups. The
current ADAFs and their age groupings are 10 for <2 years, 3 for 2 to <16 years, and 1 for 16 years and
above (U. S. EPA. 2005b). For risk assessments based on specific exposure assessments, the 10-fold and
3-fold adjustments to the unit risk estimates are to be combined with age-specific exposure estimates
when estimating cancer risks from early-life (<16 years of age) exposure.
These ADAFs, however, were formulated based on comparisons of the ratios of cancer potency
estimates from juvenile-only exposures to cancer potency estimates from adult-only exposures, from
rodent bioassay data sets with appropriate exposure scenarios, and they are designed to be applied to
cancer potency estimates derived from adult-only exposures. Thus, alternate life-table analyses were
conducted to derive comparable adult-exposure-only unit risk estimates to which ADAFs would be
applied to account for early-life exposure.45 For these alternate life-table analyses, it was assumed that
RR is independent of age for adults, which represent the life stage for which the exposure-response data
and the Cox regression modeling results from the NIOSH cohort study specifically pertain, but that there
45Inthis assessment, adult-exposure-only unit risk estimates refer to estimates derived fiomthe life-table analysis assuming
exposure only for ages >16 years. The adult-exposure-only unit risk estimates are merely intermediate values in the
calculation of adult-based unit risk estimates and should not be used in any risk calculations. Adult-based unit risk estimates
refer to estimates derived after rescaling the adult-exposure-only unit risk estimates to a (70-year) lifetime, as described later
in Section 4.4. The adult-based unit risk estimates are intended to be used in ADAF calculations (U.S. EPA. 2005b') for the
computation of extra risk estimates for specific exposure scenarios. Note that the unit risk estimates in this section, which are
derived under an assumption of increased early-life susceptibility, supersede those that were derived in Section 4.1 under the
assumption that RR is independent of age.
4-85
-------
is increased early-life susceptibility, based on the weight-of-evidence-based conclusion that EtO
carcinogenicity has a mutagenic mode of action (see Section 3.4), which supersedes the assumption that
RR is independent of age for all ages including children.
In the alternate analyses, exposure in the life table was taken to start at age 16 years, the age cut
point that was established in the EPA's Supplemental Guidance (U.S. EPA. 2005b). to derive an
adult-exposure-only unit risk estimate to which ADAFs would be applied to account for early-life
exposure. Other than the age at which exposure was initiated, the life-table analyses are identical to
those conducted for the results presented in Section 4.1. Incidence estimates are preferred over
mortality estimates; thus, adult-exposure-only unit risk estimates were derived for cancer incidence for
both lymphoid and breast cancers. Alternate estimates were not derived for all lymphohematopoietic
cancers because lymphoid cancer was the preferred lymphohematopoietic endpoint (see Section 4.1.1.2).
For each cancer endpoint, the same exposure-response model was used as that which was selected for
the unit risk estimates in Section 4.1 (i.e., two-piece linear spline model with the knot at
1,600 ppm x days for lymphoid cancer and two-piece linear spline model with the knot at
5,750 ppm x days for breast cancer). The results are presented in Table 4-21 along with the unit risk
estimates derived assuming that RR was independent of age for all ages (see Section 4.1) for
comparison. As can be seen in Table 4-21, the unit risk estimates for adult-only exposures range from
about 67 to about 70% of the unit risk estimates derived under the assumption of age independence
across all ages.
Table 4-21. ECoi, LECoi, and unit risk estimates for adult-only exposures3
Cancer response
ECoi (ppm)
LECoi
(ppm)
Adult-exposure-
only unit risk
estimateb
(per ppm)
Lifetime-exposure unit risk estimate
under assumption of age
independence0 (per ppm)
Lymphoid cancer
incidence (both sexes)
0.0107
0.00271
3.69°
5.26d
Breast cancer incidence
(females)
0.0206
0.0101
0.99
1.48d
aThese are intermediate values. See Table 4-24 below for the final adult-based cancer-type-specific unit risk
estimates.
bUnit risk estimate = 0.01/LECoi.
cFromTables 4-7 and 4-15 of Section4.1.
dFor unit risk estimates above 1, one can convert to riskperppb (e.g., 5.26 per ppm= 5.26 x 10"3 per ppb) to obtain
risk estimates below 1.
4-86
-------
According to the EPA's Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a) cancer
risk estimates are intended to reflect total cancer risk, not site-specific cancer risk; therefore, an
additional calculation was made to estimate the combined risk for (incident) lymphoid and breast
cancers from adult-only exposures because females would be at risk for both cancer types. Assuming
that the tumor types are independent and that the risk estimates are approximately normally distributed,
this calculation can be made as described in Section 4.1.3. First, the 0.01/ECoi for the total cancer risk
(i.e., lymphoid cancer incidence + breast cancer incidence) from adult-only exposure was estimated, as
summarized in Table 4-22.
Then, a unit risk estimate for the total cancer risk from adult-only exposure was derived, as
shown in Table 4-23.
Table 4-22. Calculation of ECoi for total cancer (incidence) risk from
adult-only exposure3
Cancer type
ECoi (ppm)
0.01/ECoi (per ppm)
Lymphoid
0.0107
0.935
Breast
0.0206
0.485
Totalb
--
1.42
aThese are intermediate values. See Table 4-25 below for the final adult-based cancer-type-specific 0.01/ECoi
estimates.
bThe total 0.01/ECoi value equals the sum of the individual 0.01/ECoi values.
Table 4-23. Calculation of total cancer unit risk estimate from adult-only
exposure3
Cancer type
Adult-exposure-
only unit risk
estimate
(per ppm)
0.01/ECoi
(per ppm)
SEb(per ppm)
Variance
Adult-exposure-only total
cancer unit risk estimate
(per ppm)
Lymphoid
3.69
0.935
1.67
2.805
--
Breast
0.990
0.485
0.307
0.094
--
Total
--
1.42
[1.70]°
2.90
4.22d
aThese are intermediate values. See Table 4-24 below for the final adult-based cancer type-specific unit risk
estimates.
bSE = (unit risk-0.01/ECoi)/1.645.
The SE of the total cancer risk is calculated as the square root of the sum of the variances (next column), not as the
sum of the SEs.
dTotal cancer unit risk = 1.42 + 1.645 x 1.70.
4-87
-------
Thus, the total cancer unit risk estimate from adult-only exposure is 4.22 per ppm
(2.31 x 10 3 per (J,g/m3). While there are uncertainties regarding the assumption of a normal distribution
of risk estimates, the resulting unit risk estimate is appropriately bounded in the roughly threefold range
between estimates based on the sum of the individual MLEs (i.e., 1.42) and the sum of the individual
95% UCLs (i.e., unit risk estimates, 4.68), or more precisely in this case, between the largest individual
unit risk estimate (i.e., that for lymphoid cancer), which has most of the variance, plus the 0.01/ECoi for
breast cancer (4.18) and the sum of the unit risk estimates (4.68), and so, any inaccuracy in the total
cancer risk estimate resulting from the approach used to combine risk estimates across cancer types is
relatively minor.
When the EPA derives unit risk estimates from rodent bioassay data, there is a blurring of the
distinction between lifetime and adult-only exposures because the relative amount of time that a rodent
spends as a juvenile is negligible (<8%) compared to its lifespan. [According to the EPA's
Supplemental Guidance, puberty begins around 5-7 weeks of age in rats and around 4-6 weeks in mice
(U.S. EPA. 2005b)1. Thus, when exposure in a rodent is initiated at 5-8 weeks, as in the typical rodent
bioassay, and the bioassay is terminated after 104 weeks of exposure, the unit risk estimate derived from
the resulting cancer incidence data is considered a unit risk estimate from lifetime exposure, except
when the ADAFs were formulated and are applied, in which case the same estimate is considered to
apply to adult-only exposure. Yet, when adult exposures are considered in the application of ADAFs,
the adult-exposure-only unit risk estimate is prorated over the full default (average) human lifespan of
70 years, presumably because that is how adult exposures are treated when a unit risk estimate
calculated in the same manner from the same bioassay exposure paradigm is taken as a lifetime unit risk
estimate.
However, in humans, a greater proportion of time is spent in childhood (e.g., 16 of
70 years = 3%), and the distinction between lifetime exposure and adult-only exposure cannot be
ignored when human data are used as the basis for the unit risk estimates. Thus, as described above,
adult-exposure-only unit risk estimates were calculated distinct from the lifetime estimates that were
derived in Section 4.1 under the assumption of age independence for all ages. In addition, the
adult-exposure-only unit risk estimates need to be rescaled to a 70-year lifespan in order to be used in
the ADAF calculations and risk estimate calculations involving less-than-lifetime exposure scenarios in
the standard manner, which includes prorating even adult-based unit risk estimates over 70 years. Thus,
the adult-exposure-only unit risk estimates are multiplied by 70/54 to rescale the 54-year adult period of
the 70-year default lifespan to 70 years. Then, for example, if a risk estimate were calculated for a
less-than-lifetime exposure scenario involving exposure only for the full adult period of 54 years, the
rescaled unit risk estimate would be multiplied by 54/70 in the standard calculation and the
adult-exposure-only unit risk estimate would be appropriately reproduced. Without rescaling the
adult-exposure-only unit risk estimates, the example calculation just described for exposure only for the
4-88
-------
full adult period of 54 years would result in a risk estimate 77% (i.e., 54/70) of that obtained directly
from the adult-exposure-only unit risk estimates, which would be illogical. The rescaled adult-based
unit risk estimates for use in ADAF calculations and risk estimate calculations involving
less-than-lifetime exposure scenarios are presented in Table 4-24. An LECoi estimate for adult-based
total cancer risk can be calculated as 0.01/(adult-based unit risk estimate) = 1.8 x 10-3 ppm (3.3 (J,g/m3),
and an ECoi estimate for adult-based total cancer risk can be calculated as 0.01/(the 0.01/ECoi for total
cancer) x 54/70 = 5.4 x 10"3 ppm (9.9 (j,g/m3).
Table 4-24. Adult-based unit risk estimates for use in calculations involving
age-dependent adjustment factors and less-than-lifetime exposure scenarios
Cancer response3
Adult-based unit risk estimate
(per ppm)b
Adult-based unit risk estimate
(per jig/m3)
Lymphoid cancer incidence
4.78
2.61 x lO"3
Breast cancer incidence
1.28
7.01 x 10-4
Total cancer incidence
5.47
2.99 x 10 3
"Two-piece linear spline model with knot at 1,600 ppm x days for lymphoid cancer incidence; two-piece linear
spline model with knot at 5,750 ppm x days for breast cancer incidence.
bFor unit risk estimates above 1, one can convert to risk per ppb (e.g., 5.47 perppm= 5.47 x 10 3perppb).
An example calculation illustrating the application of the ADAFs to the human-data-derived
adult-based (rescaled as discussed above) unit risk estimate for EtO for a lifetime exposure scenario is
presented below. For inhalation exposures, assuming ppm equivalence across age groups (i.e.,
equivalent risk from equivalent exposure levels, independent of body size), the ADAF calculation is
fairly straightforward.
Total cancer risk from exposure to constant EtO exposure level of 1 [j,g/m3 from ages 0-70 years:
Unit risk Exposure Duration Partial
Age group ADAF (per ug/m3) cone (ug/m3) adjustment risk
0 to <2 years 10 2.99 x 10"3 1 2 years/70 years 8.54 x 10-4
2 to <16 years 3 2.99 x 10-3 1 14 years/70 years 1.79 x 10-3
>16 years 1 2.99 x 10-3 1 54 years/70 years 2.31 x 1Q-3
total lifetime risk = 4.96 x 10-3
The partial risk for each age group is the product of the values in columns 2-5 [e.g.,
10 x (2.99x 10"3) x l x 2/70 = 8.54 x 10"4], and the total risk is the sum of the partial risks.
4-89
-------
This 70-year risk estimate for a constant exposure of 1 [j,g/m3 is equivalent to a lifetime unit risk
estimate of 5.0 x 10"3 per jig/m3 (9.1 per ppm, or 9.1 x 10"3 per ppb), adjusted for potential increased
early-life susceptibility, assuming a 70-year lifetime and constant exposure across age groups. Note that
because of the use of the rescaled adult-based unit risk estimate, the partial risk for the > 16-year age
group is the same as would be obtained for a 1 [j,g/m3 constant exposure directly from the total cancer
adult-exposure-only unit risk estimate of 2.31 x 10"3 per [j,g/m3 that was presented above, as it should be.
In addition to the uncertainties discussed above for the inhalation unit risk estimate, there are
uncertainties in the application of ADAFs to adjust for potential increased early-life susceptibility. The
ADAFs reflect an expectation of increased risk from early-life exposure to carcinogens with a mutagenic
mode of action (U.S. EPA. 2005b), but they are general adjustment factors and are not specific to EtO.
With respect to the breast cancer estimates, for example, evidence suggests that puberty/early adulthood
is a particularly susceptible life stage for breast cancer induction (U.S. EPA. 2005b; Russo and Russo.
1999); however, the EPA has not developed alternate ADAFs to reflect such a pattern of increased
early-life susceptibility, and there is currently no EPA guidance on an alternate approach for adjusting
for early-life susceptibility to potential breast carcinogens.
4.5. INHALATION UNIT RISK ESTIMATES—CONCLUSIONS
For both humans and laboratory animals, tumors occur at multiple sites. In humans, there was a
combination of tumors having lymphohematopoietic, in particular lymphoid, origins in both sexes and
breast cancer in females, and in rodents, lymphohematopoietic tumors, mammary carcinomas, and
tumors of other sites were observed. From human data, an extra cancer unit risk estimate of 2.88 x 10"3
per [j,g/m3 (5.26 x 10"3 per ppb) was calculated for lymphoid cancer incidence, and a unit risk estimate of
8.10 x 10"4 per [j,g/m3 (1.48 x 10"3 per ppb) was calculated for breast cancer incidence in females, under
the assumption that RR is independent of age for all ages (see Section 4.1). The total extra cancer unit
risk estimate was 3.31 x 10-3 per [j.g/m3 (6.06 x 10"3 per ppb) for both cancer types combined). Unitrisk
estimates derived from the three chronic rodent bioassays for EtO ranged from 2.2 x 10"5 per [j,g/m3 to
4.6 x 10"5 per (J,g/m3, roughly two orders of magnitude lower than the estimates based on human data.
Because a mutagenic mode of action for EtO carcinogenicity (see Section 3.4.1) is "sufficiently
supported in (laboratory) animals" and "relevant to humans," and as there are no chemical-specific data
to evaluate the differences between adults and children, increased early-life susceptibility should be
assumed, in accordance with the EPA's Supplemental Guidance (U.S. EPA. 2005b). This assumption of
increased early-life susceptibility supersedes the assumption of age independence under which the
human-data-based estimates presented in the previous paragraph were derived. Thus, as described in
Section 4.4, adult-exposure-only (i.e., ages >16 years) unit risk estimates were calculated from the
4-90
-------
human data under an alternate assumption that RR is independent of age for adults, which represent the
life stage for which the exposure-response data that were modeled pertain. These adult-exposure-only
unit risk estimates were then rescaled to a 70-year basis to derive adult-based unit risk estimates for use
in the standard ADAF calculations and risk estimate calculations involving less-than-lifetime exposure
scenarios. The resulting adult-based unit risk estimates were 2.61 x 10-3 per [j,g/m3 (4.78 x 10~3 per ppb)
for lymphoid cancer incidence and 7.01 x 10"4 per [j,g/m3 (1.28 x 10"3 per ppb) for breast cancer
incidence in females. The adult-based total extra cancer unit risk estimate for use in ADAF calculations
and risk estimate calculations involving less-than-lifetime exposure scenarios was 2.99 x 10~3 per [j,g/m3
(5.47 x 10"3 per ppb) for both cancer types combined.
When using the adult-based unit risk estimates to estimate extra cancer risks for a given exposure
scenario, the ADAFs should be applied, in accordance with the EPA's Supplemental Guidance (U.S.
EPA. 2005b). Applying the ADAFs to obtain a full lifetime unit risk estimate yields
5.47/ppm x [(10 x 2 years/70 years) + (3 x 14/70) + (1 x 54/70)] (4-8)
= 9.08/ppm = 4.96 x 10"3/([j,g/m3).
Applying the ADAFs to the unit risk estimates derived from the three chronic rodent bioassays for EtO
yields estimates ranging from 3.7 x 10~5 per [j,g/m3 to 7.6 x 10~5 per (J,g/m3, again roughly two orders of
magnitude lower than the estimate based on human data.
Adequate human data, if available, are considered to provide a more appropriate basis than
rodent data for estimating human risks (U.S. EPA. 2005a). primarily because uncertainties in
extrapolating quantitative risks from rodents to humans are avoided. Although there is a sizeable
difference between the rodent-based and the human-based estimates, the human data are from a large,
high-quality study, with EtO exposure estimates for the individual workers and little reported exposure
to chemicals other than EtO. Therefore, the human-based full lifetime total extra cancer unit risk
estimate of 5.0 x 103 per jig/m3 (9.1 x 103 per ppb) calculated for lymphoid cancers and breast cancer
combined and applying the ADAFs is the preferred lifetime unit risk estimate.46 For less-than-lifetime
exposure scenarios, the human-data-derived (rescaled) adult-based unit risk estimate of 3.0 x 10"3 per
[j,g/m3 (5.5 x 10"3 per ppb) should be used, in conjunction with the ADAFs if early-life exposures occur.
'"'More precisely, this unit risk estimate reflects the total (upper bound) cancer risk to females and not to the general
population because the breast cancer risk applies only to females. As a practical matter for regulatory purposes, however,
females comprise roughly half the general population, and this unit risk estimate enables risk managers to evaluate the
individual risk for this substantial population group. For the purposes of estimating numbers of cancer cases attributable to
specific exposure levels (e.g., for benefits analyses), it would be more appropriate to use the cancer-specific unit risk
estimates (or central tendency estimates), taking sex into account.
4-91
-------
Although there are uncertainties in this unit risk estimate—primarily related to exposure
misclassification, model uncertainty, and low-dose extrapolation, as discussed in
Section 4.1.4—confidence in the unit risk estimate is relatively high. First, there is high confidence in
the hazard characterization of EtO as "carcinogenic to humans," which is based on strong
epidemiological evidence supplemented by other lines of evidence, such as genotoxicity in both rodents
and humans (see Section 3.5.1). Second, the unit risk estimate is based on human data from a large,
high-quality epidemiology study with individual worker exposures estimated using a high-quality
regression model (see Section 4.1 and Section A.2.8 of Appendix A). Finally, the use of linear
low-exposure extrapolation is strongly supported by the conclusion that EtO carcinogenicity has a
mutagenic mode of action (see Section 3.4.1).
Confidence in the unit risk estimate is particularly high for the breast cancer component, which is
based on over 200 incident cases for which the investigators had information on other potential breast
cancer risk factors (see Section 4.1.2.3). The selected model for the breast cancer incidence data
provided a good global fit as well as a good local fit in the lower exposure range of greatest relevance
for the derivation of a unit risk estimate. The actual unit risk might be higher or lower; however,
considering the continuous-exposure linear model as a lower bound for the supralinear
exposure-response relationship suggests that while a unit risk estimate for breast cancer incidence that is
up to fourfold lower is plausible, unit risk estimates lower than that are considered unlikely from the
available data. Sensitivity analyses for lag time, inclusion of covariates, knot, upper-bound estimation
approach, use of the full incidence cohort, and inclusion of only invasive cancers for the breast cancer
background rates in the life-table indicate that the unit risk estimate is not highly influenced by these
factors, with comparison unit risk estimates differing by at most 40% (see Section 4.1.2.3).
There is lower confidence in the lymphoid cancer component of the unit risk estimate because it
is based on fewer events (53 lymphoid cancer deaths); incidence risk was estimated from mortality data;
and the exposure-response relationship is exceedingly supralinear (see Figure 4-2), complicating the
exposure-response modeling and model selection to a greater extent than for breast cancer incidence.
The selected model had ap-value that minimally exceeded 0.05 (p = 0.07) and it was not the best-fitting
spline model in terms of AIC for knot selection; however, its AIC was within two units of the lowest
AIC of all the models considered, and the selection of this model was consistent with the model
selection objectives for this assessment (see Section 4.1.1.2), including prioritizing models providing
good local fit in the lower-exposure region. The actual unit risk might be higher or lower than that from
the selected model, and there were no clear upper or lower bounds for the apparent exposure-response
relationship provided by other models. Sensitivity analyses for lag time, knot, and upper-bound
estimation approach, indicate that the unit risk estimate for lymphoid cancer is more influenced by these
factors than was the estimate for breast cancer incidence. Comparison unit risk estimates from the
sensitivity analyses ranged from about 50% of the preferred unit risk estimate to about three times that
4-92
-------
estimate (see Section 4.1.1.3). While there is lower confidence in the lymphoid cancer unit risk estimate
than in the breast cancer unit risk estimate, the lymphoid cancer estimate is considered a reasonable
estimate from the available data, and overall, there is relatively high confidence in the total cancer unit
risk estimate.
The unit risk estimate is intended to be an upper bound on cancer risk for use with exposures
below the POD (i.e., the LECoi). The unit risk estimate should not generally be used above the POD;
however, in the case of this total extra cancer unit risk, which is based on cancer type-specific unit risk
estimates from two linear models, the estimate should be valid for exposures up to about 0.021 ppm
(38 (J,g/m3), which is the minimum of the limits for the lymphoid cancer unit risk estimate (0.021 ppm;
see Section 4.1.1.2) and the breast cancer unit risk estimate (0.075 ppm; see Section 4.1.2.3) dictated by
the knot locations (see Section 4.7 for risk estimates based on occupational exposure scenarios).
Using the above full lifetime unit risk estimate of 9.1 x 10"3 per ppb (5.0 x 10"3 per (J,g/m3), the
(lower bound) lifetime chronic exposure level of EtO corresponding to an increased cancer risk of 10"6
can be estimated as follows:
(10"6)/(9.1/ppm) = 1.1 x 10"7 ppm = 1.1 x 10"4 ppb = 2 x 10"4 ^g/m3. (4-9)
The inhalation unit risk estimate presented above, which is calculated based on a linear
extrapolation from the POD (LECoi), is expected to provide an upper bound on the risk of cancer
incidence. For some applications, however, estimates of "central tendency" for the risk below the POD
are desired. Thus, adult-based extra risk estimates per ppm for the cancer incidence responses based on
linear extrapolation from the adult-exposure-only ECoi (i.e., 0.01/ECoi), and rescaled to a 70-year basis
for use in ADAF calculations and risk estimate calculations involving less-than-lifetime exposure
scenarios (see Section 4.4), are reported in Table 4-25. The adult-exposure-only ECois were derived
from the low-dose segments of the two-piece linear spline models for lymphoid cancer and breast cancer
incidence. (Note that, for each of these models, the low-exposure extrapolated estimates are a straight
linear continuation of the linear models used above the PODs, and thus, the statistical properties of the
models are preserved.) These 0.01/ECoi estimates are dependent on the suitability of the models used
for deriving the ECoi estimates as well as on the applicability of the linear low-dose extrapolation. The
assumption of low-dose linearity is supported by the mutagenicity of EtO (see Section 3.4). If these
adult-based 0.01/ECoi estimates are to be used, ADAFs should be applied if early-life exposure occurs,
in accordance with the EPA's Supplemental Guidance (U.S. EPA. 2005b).
As can be seen by comparing the adult-based 0.01/ECoi estimates in Table 4-25 with the
adult-based unit risk estimates (i.e., 0.01/LECoi estimates) in Table 4-24, the 0.01/ECoi estimate is about
25% of the unit risk estimate for lymphoid cancer, about 50% of the unit risk estimate for breast cancer
incidence, and about 33% of the unit risk estimate for total cancer incidence.
4-93
-------
Table 4-25. Adult-based extra risk estimates per ppm based on
adult-exposure-only ECois (0.01/ECoi estimates)3
Cancer response
Adult-exposure-only
ECoi (ppm)
Adult-based
0.01/ECoi (per ppm)bc
Lymphoid cancer incidence (both sexes)
0.0107
1.21d
Breast cancer incidence (females)
0.0206
0.629
Total cancer incidence
1.84d>e
aADAFs should be applied to the adult-based 0.01/ECoi estimates if early-life exposure occurs, in
accordance with the EPA's Supplemental Guidance.
bThese estimates are calculated as 0.01/ECoi for the adult-exposure-only extra risk estimate per ppm
rescaledto a 70-yr basis by multiplying by 70/54 (see Section4.4).
cFor conversion to per |ig/ ml divide by 1,830.
Calculated as the sum of the individual adult-based 0.01/ECois.
Tor unit risk estimates above 1, convert to risk per ppb (e.g., 1.84 perppm= 1.84 x 10 3perppb).
Finally, it should be noted that some investigators have posited that the high and variable
background levels of endogenous EtO-induced DNA damage in the body (see Section 3.3.3.1) may
overwhelm any contribution from low levels of exogenous EtO exposure (Marsden et al., 2009; SAB,
2007). It is true that the existence of these high and variable background levels may make it hard to
observe statistically significant increases in risk from low levels of exogenous exposure. However, there
is clear evidence of carcinogenic hazard from the rodent bioassays and strong evidence from human
studies (see Section 3.5), and the genotoxicity/mutagenicity of EtO (see Section 3.4) supports low-dose
linear extrapolation of risk estimates from those studies (U.S. EPA. 2005a). In fact, as discussed in
Section 3.3.3.1, Marsden et al. (2009), using sensitive detection techniques and an approach designed to
separately quantify endogenous N7-HEG adducts and "exogenous" N7-HEG adducts induced by EtO
treatment in rats, reported (non-significant) increases in exogenous adducts in DNA of spleen and liver
at the lowest dose administered (0.0001 mg/kg injected i.p. daily for 3 days, which is a very low dose
compared to the LOAELs in the carcinogenicity bioassays; see Section C.7 of Appendix C). Marsden et
al. (2009) also reported statistically significant linear dose-response relationships (p < 0.05) for
exogenous adducts in all three tissues examined (spleen, liver, and stomach), although they caution that
their study was not designed to test for linearity and that some of the adduct levels induced at low EtO
concentrations are below the limit of accurate quantitation. Furthermore, while the contributions to
DNA damage from low exogenous EtO exposures may appear "negligible" (Marsden et al., 2009)
compared to those from endogenous EtO exposure, low levels of exogenous EtO may nonetheless be
responsible for additional risk (above background risk) above de minimis risk levels, which are generally
10"6 to 10"4 for cancer. This is not inconsistent with the much higher levels of background cancer risk, to
4-94
-------
which endogenous EtO may contribute, for the two cancer types observed in the human
studies—lymphoid cancers have a background lifetime incidence risk on the order of 3%, while the
background lifetime incidence risk for breast cancer is on the order of 15% 47
Also related to the issue of endogenous EtO, Starr and Swenberg (2013) have proposed an
approach for bounding the cancer risk from low levels of exogenous exposure to chemicals that also
exist endogenously. In brief, Starr and Swenberg (2013) assume that all background cancer risk (Po) for
a specific cancer type is attributable to background levels (Co) of some endogenous adduct (as a marker
of exposure) of the chemical of interest \ Starr and Swenberg (2013) use formaldehyde as an example] in
that tissue; they then use the ratio Po:Co (actually the lower bound on Co) to estimate a linear slope for
risk as a function of endogenous adduct level down to zero adducts, which they claim is a conservative
upper bound on cancer risk from low levels of exogenous exposure (similarly expressed in terms of
adduct levels). However, the EPA disagrees that this approach necessarily yields a conservative bound
on risk because even if the adduct level is an appropriate dose metric for comparing cancer risks, the
approach relies on the assumption that the dose-response relationship over the dose range of endogenous
adducts is linear down to zero adducts. In contrast to this assumption, the EPA considers it highly
plausible that the dose-response relationship over the endogenous range is sublinear (e.g., that the
baseline levels of DNA repair enzymes and other protective systems evolved to deal with endogenous
DNA damage would work more effectively for lower levels of endogenous adducts), that is, that the
slope of the dose-response relationship for risk per adduct would increase as the level of endogenous
adducts increases. If the dose-response relationship over the endogenous range is sublinear, rather than
linear as assumed by Starr and Swenberg (2013). then the approach proposed by Starr and Swenberg
(2013) does not necessarily produce a conservative bound on risk (Crump et al.. 2014).
See Table 4-26 for a summary of key unit risk estimates derived in this assessment. See
Section 4.7 for risk estimates based on occupational exposure scenarios.
47These background lifetime incidence values were obtained from the life-table analysis, based on SEER rates, as discussed
in Sections 4.1.1.3 and 4.1.2.3. For lymphoid cancer, for example, see the value of Ro at the bottom of the life-table analysis
in Appendix E.
4-95
-------
Table 4-26. Summary of key unit risk estimates from this assessment (see
Section 4.7 for risk estimates based on occupational exposure scenarios)
Basis
Inhalation unit risk estimate" (per n«j/m3)b
Full lifetime unit risk estimate'
Total cancer risk based on human data (NIOSH cohort of
sterilizer workers)—lymphoid cancer incidence and breast
cancer incidence in females (two-piece linear spline models)
5.0 x 10J
Adult-based unit risk estimates'1
Total cancer risk based on human data (NIOSH
cohort)—lymphoid cancer incidence and breast cancer
incidence in females (two-piece linear spline models)
3.0 x 10J
Lymphoid cancer incidence based on human data (NIOSH
cohort)—preferred model: two-piece linear spline model
2.6 x 10-3
Breast cancer incidence in females based on human data
(NIOSH cohort)—preferred model: two-piece linear spline
model
7.0 x 10"4
Preferred total cancer incidence risk estimate from rodent data
(female mouse)
4.6 x 10-5
Range of total cancer incidence risk estimates from rodent data
(mouse and rat)
2.2 x 10"5 to 4.6 x 10"5
Adult-based 0.01/ECoiestimates'
Lymphoid cancer incidence based on human data (NIOSH
cohort)—two-piece linear spline model
6.6 x 10"4
Breast cancer incidence in females based on human data
(NIOSH cohort)—two-piece linear spline model
3.4 x 10-4
Total cancer incidence based on human data (NIOSH cohort)
1.0 x 10"3
aStrictly speaking, the values listed in this table are not all unit risk estimates as defined by the EPA, but they are all
potency estimates that, when multiplied by an exposure value, give an estimate of extra cancer risk. These potency
estimates are not intended for use with continuous lifetime exposure levels above 38 (ig/m3. See Section 4.7 for
risk estimates based on occupational exposure scenarios. Preferred estimates are in bold.
bTo convert unit risk estimates to (ppm)1, multiply the ((ig^m3)1 estimates by 1,830 (|ig/m3)/ppm.
°Because the weight of evidence supports a mutagenic mode of action for EtO carcinogenicity and because of the
lack of chemical-specific data, the EPA assumes increased early-life susceptibility and recommends the application
of ADAFs, in accordance with the EPA's Supplemental Guidance (U.S. EPA. 2005b'). for exposure scenarios that
include early-life exposures. For the full lifetime (upper-bound) unit risk estimate presented here, ADAFs have
been applied, as described in Section 4.4.
dThese (upper-bound) unit risk estimates are intended for use in ADAF calculations and less-than-lifetime adult
exposure scenarios (U.S. EPA. 2005^). Note that these are not the same as the unit risk estimates derived directly
from the human data in Section 4.1 under the assumption that RRs are independent of age. Under that assumption,
the key unit risk estimates were 2.9 x 10 3 per (ig/m3 for lymphoid cancer incidence, 8.1 x 10"4 per (ig/m3 for breast
cancer incidence, and 3.3 x 10 3 per (ig/m3 for the combined cancer incidence risk from those two cancers. See
Section 4.4 for the derivation of the adult-based unit risk estimates.
eThese are not upper-bound risk estimates but, rather, estimates based on linear extrapolation from the ECoi.
ADAFs should be applied if early-life exposure occurs, in accordance with the EPA's Supplemental Guidance (U.S.
EPA. 2005b).
4-96
-------
4.6. COMPARISON WITH OTHER PUBLISHED RISK ESTIMATES
The unit risk values derived in this document are compared with other recent risk estimates
presented in the published literature (see Table 4-27).
4.6.1. Unit Risk Estimates Based on Human Studies
Kirman et al. (2004) used leukemia data only and pooled data from both the Stayner et al. (1993)
and the UCC studies (Teta et al., 1999; Teta et al., 1993). Based on the assumption that leukemias are
due to chromosome translocations, requiring two independent events (chromosome breaks), Kirman et
al. (2004) proposed that two independent EtO-induced events are required for EtO-induced leukemias
and used a dose-squared model, yielding a unit risk value of 4.5 x 10"8 ((j,g/m3)_1 as their preferred
estimate.
The Kirman et al. (2004) values are different from those in the current document because of the
different assumptions inherent in the Kirman et al. (2004) approach and because the study used
unpublished data from earlier follow-ups of the two cohorts. A key difference is that the EPA uses a
linear model or a two-piece linear spline model rather than a quadratic (dose-squared) model in the
range of observation. Then, the EPA uses a higher extra risk level (1%) for establishing the POD,
whereas Kirman et al. (2004) used a risk level of 10"5 for their best estimate and a risk range of 10"4 to
10"6 for their range of values. The extra risk level and the corresponding POD are not critical with the
linear model; however, with the quadratic model used by Kirman et al. (2004), the lower the risk level
(and hence the POD), the greater the impact of the quadratic model and the lower the resulting unit risk
estimates.
In addition, the EPA (1) uses data for lymphoid cancers (and female breast cancers) rather than
leukemias, (2) includes ages up to 85 years in the life-table analysis rather than stopping at 70 years,
(3) calculates unit risk estimates for cancer incidence as well as mortality, (4) uses a lower bound as the
POD rather than the maximum likelihood estimate, (5) uses the results of lagged analyses rather than
untagged analyses, and (6) uses adult-based unit risk estimates in conjunction with ADAFs (see
Section 4.4) to derive the lifetime unit risk estimates.
Another key difference is that Kirman et al. (2004) relied on earlier NIOSH results (Stayner et
al., 1993), whereas the EPA uses the results of NIOSH's more recent follow-up of the cohort (Steenland
et al., 2004). Kirman et al. (2004) asserted that a quadratic dose-response model provided the best fit to
the data in the observable range and that this provides support for their assumed mode of action.
However, the 2004 NIOSH data for lymphohematopoietic cancers suggest a supralinear
exposure-response relationship (see Section 4.1.1.2 and Figures 4-2 and 4-4), which is inconsistent with
a dose-squared model. Furthermore, the EPA's review of the mode of action evidence does not support
the mode of action assumed by Kirman et al. (2004) (see Section 3.4).
4-97
-------
Table 4-27. Comparison of unit risk estimates3
Assessments
Data source
Inhalation unit risk estimateb
(per jig/m3)
Based on human data
EPA (this document)0
Lymphoid cancer incidence in sterilizer
workers (NIOSH cohort)d
7.2 x 10"4
Breast cancer incidence in female sterilizer
workers (NIOSH cohort)0
1.4 x 10-3
Total cancer risk based on the NIOSH data
1.8 x 10-3
Ki rutin etal. (2004)
Leukemia mortality in combined NIOSH
and UCC cohorts (earlier follow-ups)
4.5 x 10-8
Range of 1.4 x 10 8to 1.4 x l0 7f
Valdez-Flores etal. (2010)
Multiple individual cancer endpoints,
including all lymp ho hematopoietic,
lymphoid, and breast cancers, in combined
updated NIOSH and updated UCC cohorts
5.5 x 10-7 to 1.6 x 10_6s
Based on rodent data
EPA (this document)0
Female mouse tumors
7.6 x 10-5
Ki rutin etal. (2004)
Mononuclear cell leukemia in
rats and lymphomas in mice
2.6 x lO 8 to 1.5 x 10"5h
aUpper-bound estimates except where indicated that estimates are based onEC values (i.e., estimates with footnotes
f and g).
bBecause the weight of evidence supports a mutagenic mode of action for EtO carcinogenicity, and in the absence
of chemical-specific data, the EPA assumes increased early-life susceptibility, in accordance with the EPA's
Supplemental Guidance (U.S. EPA. 2005b'). and for the EPA lifetime unit risk estimates presented in this table,
ADAFs have been applied, as described in Section 4.4. The corresponding adult-based unit risk estimates are
4.4 x 10"4 (|ig/m3)"1 for human-based lymphoid cancer incidence, 8.2 x 10 4 ((ig/m3)1 for human-based breast
cancer incidence, 1.1 x 10 3 ((ig/m3)1 for human-based total cancer incidence, and 4.6 x 10s (^g/m3)-1 for
rodent-based total cancer incidence. The non-EPA estimates in the table are shown as reported and do not account
for potential increased early-life susceptibility for lifetime exposures that include childhood, with the exception of
the Valdez-Flores etal. (2010) estimates, which are purported to include the ADAFs, but the ADAFs were in fact
misapplied and have essentially no impact (see Appendix A.2.20).
°See Table 4-26 in Section 4.5 for a more complete summary of estimates from this assessment. See Section 4.7 for
risk estimates for occupational exposure scenarios.
dFor lymphoid cancer mortality, the ADAF-adjusted lifetime unit risk estimate is 3.3 x 10 4 (jxg/m?)1 and the
adult-based unit risk estimate is 2.0 x 10 4(^g/m3)"1.
Tor breast cancer mortality, the ADAF-adjusted lifetime unit risk estimate is 4.0 x 10 4 ((ig/m3)1 and the
adult-based unit risk estimate is 2.4 x 10 4 ((ig/m3)1.
'Estimates based on linear extrapolation fro mECOOOl-ECOOOOOl obtained from the quadratic model.
gEstimates based on range of EC(l/million)s of 0.001-0.003 ppm obtained from the model RR = e(|3 xe>Posure) for
relevant cancer endpoints.
hEstimates based on quadratic extrapolation model below the observable range of the data (i.e., below the LECio or
LECoi obtained using multistage model) with various points of departure (LECoi-LECoooooi) for final linear
extrapolation (see Section 4.4.2).
4-98
-------
The Valdez-Flores et al. (2010) unit risk estimates (see Table 4-27) are similarly much lower
than those in the current document because of the different assumptions used. A key difference is that
the EPA uses a linear model or a two-piece linear spline model in the range of observation rather than an
exponential model (RR = ep x exposure), which was used by Valdez-Flores et al. (2010) despite its lack of
fit. Then, the EPA uses a 1% extra risk level for establishing the POD for linear extrapolation, whereas
Valdez-Flores et al. (2010) used a risk level of 10"6. In addition, the EPA (1) includes ages up to
85 years in the life-table analysis rather than stopping at 70 years, (2) calculates unit risk estimates for
cancer incidence as well as mortality, (3) uses a lower bound as the POD rather than the maximum
likelihood estimate, and (4) uses the results of lagged analyses rather than untagged analyses. See
Appendix A.2.20 for a more detailed discussion of the differences between the EPA and Valdez-Flores
et al. (2010) analyses.
4.6.2. Unit Risk Estimates Based on Laboratory Animal Studies
Kirman et al. (2004) also used linear and dose-squared extrapolation models to derive unit risk
estimates based on the rat mononuclear cell leukemia data and the mouse lymphoma data. First, they
used the multistage model to calculate the LECio (LECoi for the male mouse lymphoma data) for the
POD from the observable range. Then, using these PODs for linear extrapolation, Kirman et al. (2004)
obtained a unit risk range of 3.9 x 10~6 (|ig/m3)"' to 1.5 x 10"5 (jig/in3)"1. Alternatively, Kirman et al.
(2004) used a quadratic extrapolation model below the observable range to estimate secondary points of
departure (LECoi~LECoooooi; LECooi~LECoooooi for the male mouse) for final linear low-dose
extrapolation, yielding unit risks ranging from 2.6 x 10"8 (jj.g/m3)-1 to 4.9 x 10~6 ((j.g/m3)"1. These values
are all smaller than the unit risks derived from the rodent data in this document.
4.7. RISK ESTIMATES FOR SOME OCCUPATIONAL EXPOSURE SCENARIOS
The unit risk estimates derived in the preceding sections were developed for environmental
exposure levels, where maximum modeled levels are on the order of 1-2 [j,g/m3 (email dated October 3,
2005, from Mark Morris, EPA, to Jennifer Jinot, EPA), i.e., roughly 0.5-1 ppb, and are not applicable to
higher exposures, including some occupational exposure levels. However, occupational exposure levels
of EtO are of concern to the EPA when EtO is used as a pesticide (e.g., sterilizing agent or fumigant).
The occupational exposure scenarios of interest to the EPA include some cumulative exposures
corresponding to exposure levels in the nonlinear range of some of the models (i.e., above the maximum
exposure level at which the low-dose-linear unit risk estimates apply). Therefore, extra risk estimates
were calculated for a number of occupational exposure scenarios of possible concern. Extra risk
estimates are estimates of the extra cancer risk above background and are the same type of estimate that
one gets from multiplying a unit risk estimate by an exposure level. In this case, the exposure level is
used directly in the exposure-response model, thus accounting for any nonlinearities in the model above
4-99
-------
the range of exposure levels for which the linear unit risk estimate is applicable. For these occupational
exposure scenarios, exposure-response models based on data from the NIOSH cohort were used in
conjunction with the life-table program, as previously discussed in Section 4.1. The NIOSH cohort is
especially relevant to the EPA's ethylene oxide concerns because it is a cohort of sterilizer workers. A
35-year exposure occurring between ages 20 and 55 years was assumed, and exposure levels ranging
from 0.1 to 1 ppm 8-hour TWA were examined (i.e., ranging from about 1,300 to 13,000 ppm x days).
(Note that the current Occupational Safety and Health Administration Permissible Exposure Limit is
1 ppm [8-hour TWA].)
4.7.1. Extra Risk Estimates for Lymphoid Cancer
For lymphoid cancer mortality in both sexes, the same model of the Steenland et al. (2004) data
that was selected for the derivation of the unit risk estimate was used for the estimation of the extra risks
associated with the occupational exposure scenarios, consistent with the model selection objectives for
this assessment (see Section 4.1.1.2). The selected model is the two-piece linear spline model with the
knot at 1,600 ppm x days (cumulative exposure, with a 15-year lag) (see Section 4.1.1.2 and Section D.3
of Appendix D). While this model was considered to provide the best local fit to the data in the
low-exposure region for the purposes of deriving a unit risk estimate for low-exposure extrapolation, the
model appears to underestimate the risk at higher exposures of concern for the occupational exposure
scenarios, as compared to models with better global fits to the full range of the data. Figure 4-3, for
example, shows that the four models with the best global fits in terms of AIC (the linear and log-linear
two-piece spline models with knots at 100 ppm x days and the linear and log-linear log cumulative
exposure models; see Table 4-6) all indicate higher RR estimates than the selected model for the
occupational exposure range of interest (1,277.5 to 12,775 ppm x days). Thus, estimates from the
best-fitting48 log cumulative exposure Cox regression model (with 15-year lag; see Section 4.1.1.2 and
Section D.3 in Appendix D) are also presented for comparison. The next "best-fitting" (in terms of
AIC) model across the full range of exposures is the log cumulative exposure linear model (see
Section 4.1.1.2 and Section D.3.3 of Appendix D), which would yield even higher extra risk estimates
than the log cumulative exposure Cox regression model across the range of occupational exposure
scenarios of interest (see Figure 4-3),49 but only the latter model is considered for comparisons here.
The extra risk results for lymphoid cancer mortality and incidence in both sexes for the selected
two-piece linear spline model with the knot at 1,600 ppm x days and the log cumulative exposure Cox
1xIn terms of AIC, after accounting for the 0.4 unit discrepancy between linear and log-linear models; see footnote 25 in
Section4.1.1.2.
49For example, the MLEs of extra risk from the log cumulative exposure linear model would range from about 22% higher
than that from the log cumulative exposure Cox regression model for the 0.1 ppm 8-hour TWA to about 5% higher for the
1 ppm 8-hour TWA.
4-100
-------
regression model, for comparison, are presented in Table 4-28. For the lymphoid cancer incidence
estimates, the exposure-response relationship was assumed to be the same as for mortality (see
Section 4.1.1.3). The models used to derive the extra risk estimates presented in Table 4-28 for
lymphoid cancer for the occupational exposure scenarios are displayed in Figure 4-9 over the range of
occupational cumulative exposures of interest; the categorical results are included for comparison.
The 95% (one-sided) upper bounds for the two continuous exposure models were estimated
using a Wald approach. While this approach is appropriate for the log cumulative exposure Cox
regression model, a profile likelihood approach, which allows for asymmetric CIs, would have been
preferred for the linear spline model (Langholz and Richardson. 2010). However, a formula for
applying the profile likelihood approach in the range of the second spline segment was not available for
the life-table analysis; thus, the upper bounds for the two-piece linear spline model were approximated
using a Wald approach. A comparison of the 95% upper bounds on RR derived for the linear spline
model using the two different approaches shows that the Wald upper-bound estimates are about halfway
between the MLE RR estimates and the profile likelihood upper-bound estimates (see Figure D-17 in
Appendix D). In the range of cumulative exposures of interest for the occupational scenarios considered
in this assessment (i.e., up to 12,775 ppm x days, with a 15-year lag), the Wald-based upper-bound
estimates are about 67% of the profile-likelihood-based upper-bound estimates.
As can be seen in Table 4-28, the extra risk estimates from the selected two-piece spline model
are consistently below those of the log cumulative exposure Cox regression model. The MLEs for
lymphoid cancer mortality and incidence from the log cumulative exposure Cox regression model range
from about 15% higher than those from the two-piece spline model at 0.2 ppm 8-hour TWA to about
40% higher at 1 ppm 8-hour TWA. The 95% (one-sided) UCLs on the extra risks range from about 20%
higher for mortality and incidence from the log cumulative exposure Cox regression model at 0.2 ppm
8-hour TWA to about 70% higher for mortality and 60% higher for incidence at 1 ppm 8-hour TWA.
As noted above, the log cumulative exposure linear model would yield even higher extra risk estimates
than the log cumulative exposure Cox regression model across the range of occupational exposure
scenarios of interest (see Figure 4-3). So, while the two-piece spline model was considered to provide
the best local fit to the data in the low-exposure region for the purposes of deriving a unit risk estimate
for low-exposure extrapolation, and is used here for the occupational extra risk estimates for the sake of
consistency, following the model selection objectives for this assessment (see Section 4.1.1.2), these
results confirm that the model underestimates the risks at higher exposures of concern for the
occupational exposure scenarios, compared to models with better global fits to the data. This
underestimation is compounded for the upper-bound estimates by the use of the Wald approach rather
than the more appropriate profile likelihood approach.
4-101
-------
Table 4-28. Extra risk estimates for lymphoid cancer in both sexes for various occupational exposure levels3
8-hr TWA
(ppm)
Lymphoid cancer mortality
Lymphoid cancer incidenceb
Cox regression model of log
cumulative exposure0
Linear spline model with knot at
1,600 ppm x daysd
Cox regression model of log
cumulative exposure0
Linear spline model with knot at
1,600 ppm x daysd
MLE
95% UCL
MLE
95% UCL
MLE
95% UCL
MLE
95% UCL
0.1
0.012
0.029
0.0092
0.022
0.034
0.078
0.024
0.056
0.2
0.014
0.035
0.012
0.029
0.039
0.092
0.033
0.076
0.3
0.015
0.038
0.013
0.029
0.042
0.10
0.034
0.078
0.4
0.016
0.041
0.013
0.030
0.044
0.11
0.035
0.079
0.5
0.017
0.043
0.013
0.030
0.046
0.11
0.035
0.080
0.6
0.017
0.045
0.013
0.030
0.047
0.12
0.036
0.081
0.7
0.018
0.047
0.013
0.030
0.049
0.12
0.036
0.081
0.8
0.018
0.048
0.013
0.030
0.050
0.13
0.037
0.081
0.9
0.019
0.049
0.013
0.030
0.051
0.13
0.037
0.081
1.0
0.019
0.051
0.014
0.030
0.052
0.13
0.037
0.082
g "Assuming a 35-yr exposure between ages 20 and 55 years (see Section 4.7).
bAssumes same exposure-response relationship as for lymphoid cancer mortality.
°Fromthe best-fitting50 log cumulative exposure Cox regression model for lymphoid cancer mortality in both sexes; 15-yr lag (see Appendix D; see also Section 4.1.1.2).
dTwo-piece linear spline model withknotat 1,600 ppm x days (cumulative exposure, with 15-yr lag) (see Section 4.1.1.2).
hr = hour; MLE = maximum likelihood estimate; UCL = (one-sided) upper confidence limit.
50Best-fitting nonspline model based on AIC, after accounting for the 0.4 unit difference in AICs between linear and log-linear models (see Table 4-6 and
footnote 25).
-------
3.5
¦--- log-linear log exposure
• categorical
— linear spline 1600
1
0.5
0 T 2000 4000 6000 8000 10000 12000 | 14000
0.1 ppm x 35 years Cumulative exposure (ppmx days) 1 ppm x 35 years
t
Lymphoid cancer models (see Section 4.1.1.2): log cumulative exposure Cox regression model; categorical results; two-piece
linear spline model with knot at 1,600 ppm x days. [Note that, with the exception of the categorical results, the various models
have different implicitly estimated baseline risks; thus, they are not strictly comparable to each other in terms of RR values
(i.e., along the _y-axis). They are, however, comparable in terms of general shape.]
Figure 4-9. RR estimates for lymphoid cancer from occupational EtO exposures (with 15-year lag).
-------
Finally, MLEs and Wald UCLs of extra risk for the two-piece linear spline model with the knot
at 1,600 ppm x days and different lag periods (0, 5, 10, and 20 years) were calculated for the
occupational exposure scenarios, for comparison with the results with the 15-year lag (see Section D.3.9
and Table D-43 of Appendix D). The MLEs ranged from about 25% of (5-year lag) to just over 80% of
(no lag) the estimates for the selected model (15-year lag). The 95% (one-sided) upper bounds of extra
risk ranged from about 45% of (5-year lag) to just over 5% greater than (no lag) the estimates for the
selected model. Of these models, the model with no lag was the best-fitting model after the selected
model (15-year lag), based on log likelihood (and AIC) (see Table D-38), and that is the model that had
the most similar MLEs and UCLs to the selected model. The models for lags of 5, 10, and 20 years each
had /^-values > 0.20 for inclusion of the exposure terms, indicating an inadequate fit to the data.
4.7.2. Extra Risk Estimates for Breast Cancer
For breast cancer, incidence data were available from the NIOSH incidence study; thus, only
incidence estimates were calculated. In addition to being the preferred type of cancer risk estimate, the
breast cancer incidence risk estimates are based on more cases than were available in the mortality study
and the incidence data (for the subcohort with interviews) are adjusted for a number of breast cancer risk
factors (see Section 4.1.2.3). In terms of the incidence data, the subcohort data are preferred to the full
cohort data because the subcohort data are adjusted for these potential confounders and because the full
cohort data have incomplete ascertainment of breast cancer cases.
For breast cancer incidence in the subcohort with interviews, a number of Cox regression
exposure-response models from the Steenland et al. (2003) breast cancer incidence study fit almost
equally well (see Section 4.1.2.3). These include a log cumulative exposure model and a cumulative
exposure model, both with a 15-year lag, and a log cumulative exposure model with no lag. The latter
model was not considered further because the inclusion of a 15-year lag for the development of breast
cancer was considered more biologically realistic than not including a lag. Steenland et al. (2003) also
provided a duration-of-exposure Cox regression model with a marginally better fit; however, models
using duration of exposure are less useful for estimating exposure-related risks, and duration of exposure
and cumulative exposure are correlated. Thus, only the (lagged) cumulative exposure models are
considered here. The 2014 draft assessment (U.S. EPA. 2014a. b) provides extra risk estimates from the
cumulative exposure and log cumulative exposure Cox regression models.
In the current assessment, estimates from those two models are omitted in favor of estimates
from the linear cumulative exposure and square root of cumulative exposure models (with a 15-yer lag;
see Section 4.1.2.3), which provide better overall fits to the data (based on lower AIC values) and better
local fits to the data in the range of the exposure scenarios of interest than do those log-linear models.
The two-piece linear spline model (with a 15-year lag; see Section 4.1.2.3) was also used to calculate
extra risk estimates. This was the preferred model for the derivation of unit risk estimates in
4-104
-------
Section 4.1.2.3, and it is the preferred model for the derivation of the extra risk estimates for the
occupational exposure scenarios presented here. Selection of the two-piece linear spline model as the
preferred model is consistent with the model selection objectives for this assessment, which included
prioritizing models which allow for better local fits to the low-exposure range and using the same model
for deriving both the unit risk estimate and the extra risk estimates for the occupational exposure
scenarios (see Section 4.1.1.2). Estimates from the other two models are provided for comparison. The
linear square-root model provided the best global fit (lowest AIC), and the linear model might be
considered a lower bound on a range of credible estimates for the occupational exposure scenarios of
interest (see Section 4.1.2.3).
The 95% (one-sided) upper bounds for the linear and linear square-root models were estimated
using a profile likelihood approach (Langholz and Richardson. 2010). which allows for asymmetric CIs.
The 95% (one-sided) upper bounds for the two-piece linear spline model were approximated using a
Wald approach because a formula for applying the profile likelihood approach in the range of the second
spline segment was not available for the life-table analysis. A comparison of the 95% (one-sided) upper
bounds on RR derived using the two different approaches shows that the results differ little in the range
of cumulative exposures of interest for the occupational scenarios considered in this assessment (i.e., up
to 12,775 ppm x days, with a 15-year lag; see Figure D-8 in Appendix D). The Wald approach
underestimates the upper bounds compared to the profile-likelihood approach for all exposures, but
never by more than about 4%.
The extra risk estimates for breast cancer incidence in females from the linear cumulative
exposure and square root of cumulative exposure models and the two-piece linear spline model for the
occupational exposure scenarios of interest are presented in Table 4-29, and these models are displayed
in Figure 4-10 over the corresponding range of occupational cumulative exposures.
As can be seen in Table 4-29, the extra risk estimates from the linear model are well below those
of the two-piece linear spline and linear square-root models. The MLEs and 95% (one-sided) UCLs on
the extra risks from the two-piece linear spline and linear square-root models are fairly similar over
much of the exposure range of interest. For the lowest exposure levels (0.1 to 0.3 ppm 8-hour TWAs),
both the MLEs and UCLs are higher for the linear square-root model, but by less than twofold. For the
higher exposure levels, both the MLEs and UCLs are virtually indistinguishable (differ by <10 %)
between the linear square-root model and the two-piece linear spline model.
4-105
-------
Table 4-29. Extra risk estimates for breast cancer incidence in females for various occupational exposure levelsa'b
8-hr TWA (ppm)
Linear sqrt cumulative exposure model0
Linear modeld
Two-piece linear spline model®
MLE
95%UCLf
MLE
95%UCLf
MLE
95 % U CLg
0.1
0.025
0.047
0.0033
0.0067
0.013
0.025
0.2
0.035
0.066
0.0066
0.013
0.025
0.050
0.3
0.042
0.080
0.099
0.020
0.038
0.074
0.4
0.049
0.091
0.013
0.026
0.050
0.097
0.5
0.054
0.10
0.016
0.033
0.059
0.11
0.6
0.059
0.11
0.020
0.039
0.064
0.12
0.7
0.064
0.12
0.023
0.046
0.068
0.13
0.8
0.068
0.13
0.026
0.052
0.071
0.13
0.9
0.072
0.13
0.029
0.058
0.074
0.13
1.0
0.075
0.14
0.032
0.065
0.076
0.14
aAssuming a 35-yr exposure between ages 20 and 55 years.
bFrom incidence data for subcohort with interviews; invasive and in situ tumors (Steenland et al.. 20031.
°Linear model with square-root transformation of cumulative exposure as a continuous variable, with 15-yr lag (see Section 4.1.2.3 and SectionD.l of Appendix D)
dLinear model with cumulative exposure as a continuous variable, with 15-yr lag (see Section4.1.2.3 and SectionD.l of AppendixD).
eTwo-piece linear spline model with cumulative exposure as a continuous variable, with 15-yr lag (see Section 4.1.2.3 and Table D-10 of Appendix D for parameter
values and equations). Results for occupational exposures use both spline segments; knot at 5,750 ppm x days.
fConfidence limits for the "one-piece" linear RR models were estimated using a profile likelihood approach (Langholz and Richardson. 20101. which allows for
asymmetric CIs.
"Confidence limits for the two-piece linear spline model were approximated using a Wald approach (See Table D-10 of Appendix D for parameter values and equation)
because a formula for using the profile likelihood approach in the range of the second spline segment was not available for the life-table analysis.
hr = hour; MLE = maximum likelihood estimates; UCL = (one-sided) upper confidence limit.
-------
linear (l+(3*exp)
1+ P*sqrtexp
linear spline
• categorical
6000 8000
cumulative exposure (ppm x days)
10000
12000
Breast cancer models (see Section 4.1.2.3): linear two-piece spline model, with knot at 5,750 ppm x days; linear square-root
cumulative exposure model; (continuous exposure) linear model; categorical results (deciles). [Note that the various models
have different implicitly estimated baseline risks; thus, they are not strictly comparable to each other in terms of RR values
(i.e., along the _y-axis). They are, however, comparable in terms of general shape.]
Figure 4-10. RR estimates for breast cancer incidence from occupational EtO exposures (with 15-year lag).
-------
As shown in Figure 4-10, the slope of the (continuous exposure) linear model is too shallow
across the range of exposures of interest. This is consistent with the analysis presented in Section D. 1 of
Appendix D showing the strong influence of the upper tail of cumulative exposures on the results of the
log-linear cumulative exposure (standard Cox regression) model. The responses in the upper tail of
exposures are relatively dampened, such that when the highest 5% of exposures (exposures >
27,500 ppm x days, which are well in excess of the exposures corresponding to the occupational
exposure scenarios considered here) are excluded, the slope of the Cox regression model is substantially
increased (e.g., at 10,000 ppm x days, the RR estimate increases from about 1.1 to almost 1.5; see
Figure D-4 in Appendix D). This strong influence of the upper tail of exposures would similarly
attenuate the slope of the linear model, resulting in underestimation of the lower-exposure risks. The
two-piece linear spline model, on the other hand, is more flexible, and the influence of the upper tail of
exposures would be primarily on the upper spline segment; thus, the two-piece model is able to provide
a better fit to the lower-exposure data. The linear square root of cumulative exposure model gives
similar results as the two-piece linear model over the range of occupational exposures of interest.
Finally, MLEs and Wald UCLs of extra risk for the two-piece linear spline model with the knot
at 5,750 ppm x days and different lag periods (0, 5, 10, and 20 years) were calculated for the
occupational exposure scenarios, for comparison with the results with the 15-year lag (see Section
D.l.ll and Table D-18 of Appendix D). The MLEs ranged from about 40% less than (10-year lag) to
about 30% greater than (20-year lag) the estimates from the selected model (15-year lag). The 95%
(one-sided) upper-bound estimates ranged from about 25% less than (10-year lag) to about 20% greater
than (20-year lag) those from the selected model. The model with a 20-year lag had a slightly better fit
to the data, based on log likelihood and AIC, than the model with a 20-year lag, whereas, the models
with 0-, 5-, and 10-year lags had markedly worse fits (see Table D-12 in Appendix D).
4.7.3. Extra Risk Estimates for Total Cancer
For the total cancer risk combined across the two cancer types, the MLE can be obtained directly
by summing the MLEs for the individual cancer types. An upper bound can be approximated by
summing the 95% UCL estimates for the individual cancer types. Normally, this would tend to
overestimate the corresponding 95% UCL on total cancer risk (i.e., the 95% UCL on the sum of the
MLEs); however, as discussed above, because the lymphoid cancer extra risks are underestimated
compared to models with better global fits to the data (based on AIC) and because the Wald approach
used to approximate the upper-bound estimates in place of the more accurate profile-likelihood approach
4-108
-------
underestimates the upper bounds (especially for lymphoid cancer), the overestimation is mitigated. The
summed MLEs and upper-bound estimates are presented in Table 4-30.51
Table 4-30. Extra risk estimates for total cancer incidence for various
occupational exposure levelsa'b
8-hr TWA (ppm)
Maximum likelihood estimate
Upper-bound estimate
0.1
0.037
0.081
0.2
0.058
0.13
0.3
0.072
0.15
0.4
0.085
0.18
0.5
0.094
0.19
0.6
0.10
0.20
0.7
0.10
0.21
0.8
0.11
0.21
0.9
0.11
0.21
1.0
0.11
0.22
aAssuming a 35-yr exposure between ages 20 and 55 years.
bFrom combining results for lymphoid cancer incidence in both sexes and breast cancer incidence in females,
hr = hour.
Comparing results in Tables 4-28 to 4-30 shows that lymphoid cancer contributes about 2/3 of
the total risk at 0.1 ppm, and this contribution decreases to 1/3 by 1 ppm. In addition, one can calculate
a minimum bound for what the 95% UCL on the sum of the MLEs at each exposure level would be by
taking the maximum of the sum of the MLE for one cancer type plus the UCL for the other cancer type.
The sum of the 95% UCLs for the two cancer types (the upper bound shown in Table 4-30) exceeds this
minimum upper bound by at most 33%, indicating that the overestimation incurred by approximating the
95% UCL on the total cancer risk by summing the 95% UCLs for the individual cancer types is
minimal. Moreover, calculating a similar minimum upper bound using the log cumulative exposure Cox
regression model for the lymphoid cancer risk estimates indicates that using one of the log cumulative
exposure models with a better global fit to the lymphoid cancer data, discussed above, would alleviate
51More precisely, these sums reflect the total cancer risk to females and not a mixed-sex workforce because the breast cancer
risk estimates apply only to females. As a practical matter for regulatory purposes, females typically comprise a substantial
proportion of the sterilizer workforce and summing the extra risk estimates enables risk managers to evaluate the individual
risk for this substantial workforce group. In a situation in which the workforce of concern is comprised predominantly of
males, it might be appropriate to use a sex-weighted sum of the extra risks from the two cancer types. For the purposes of
estimating numbers of cancer cases attributable to specific exposure levels (e.g., for benefits analyses), it would be most
suitable to use the cancer-specific extra risk estimates, taking sex into account.
4-109
-------
all of the overestimation. In addition, the Wald approach used to approximate the upper-bound
estimates underestimates the more accurate profile likelihood upper bounds. Thus, the overall approach
is likely underestimating the upper bounds on total risk in the range of the occupational exposure
scenarios of interest.
Although there is model uncertainty, as discussed above, there is less overall uncertainty
associated with the extra risk estimates for occupational exposure scenarios than with the unit risk
estimates for environmental exposures, and the overall confidence in the extra risk estimates is high.
The extra risk estimates are derived for occupational exposure scenarios that yield cumulative exposures
well within the range of the exposures in the NIOSH study. Moreover, the NIOSH study is a study of
sterilizer workers who used EtO for the sterilization of medical supplies or spices (Steenland et al..
1991); thus, the results are directly applicable to workers in these occupations, and these are among the
occupations of primary concern to the EPA.
4.7.4. Calculation of Extra Risk Estimates for Other Occupational Exposure Scenarios
Some detailed guidance is provided here for calculating extra risk estimates outside of the range
of occupational scenarios considered above. Note that for 35-year exposures to exposure levels between
the exposure levels presented in Tables 4-28 and 4-29 (e.g., 0.15 ppm), one could interpolate between
the extra risk estimates presented for the closest exposure levels on either side.
4.7.4.1. For Occupational Exposures with Durations Other Than 35 Years
Extra risk estimates for a 45-year exposure to the same exposure levels were nearly identical to
those from the 35-year exposure for both lymphoid cancer in both sexes and breast cancer in females
(results not shown). With exposures beginning at 20 years of age and with the 15-year lag, the
assumption of an additional 10 years of exposure only negligibly affects the risks above age 70 and has
little impact on lifetime risk. For exposure scenarios of 35-45 years but with 8-hour TWAs falling
between those presented in the tables, the extra risk can be estimated by interpolation. For exposure
scenarios with durations of exposure less than 30-35 years, extra risks can be roughly estimated by
calculating the cumulative exposure and finding the extra risks for a similar cumulative exposure in
Tables 4-28 and 4-29. For a more precise estimation, or for exposure scenarios of much shorter duration
or for specific age groups, calculations using a life-table analysis should be done, as presented in
Appendix E but modified for the specific exposure scenarios.
4.7.4.2. For Occupational Exposures Below 0.1 ppm
For lymphoid cancer, the low-exposure continuation of the two-piece linear spline model
presented in Table 4-28 of the assessment is recommended. For 35-year exposures, the following
formulae would apply:
4-110
-------
95% UCL on extra risk for lymphoid cancer incidence = (8-h TWA occ exp [in
ppm]) x (0.056/0.1 ppm) = (8-h TWA occ exp [in ppm]) x (0.56/ppm)
MLE of extra risk for lymphoid cancer incidence = (8-h TWA occ exp [in
ppm]) x (0.024/0.1 ppm) = (8-h TWA occ exp [in ppm]) x (0.24/ppm)
For breast cancer, the low-exposure continuation of the two-piece linear spline model presented
in Table 4-29 of the assessment is recommended. For 35-year exposures, the following formulae would
apply:
95% UCL on extra risk for breast cancer incidence = (8-h TWA occ exp [in ppm]) x (0.025/0.1
ppm) = (8-h TWA occ exp [in ppm]) x (0.25/ppm)
MLE of extra risk for breast cancer incidence = (8-h TWA occ exp [in ppm]) x (0.013/0. lppm) =
(8-h TWA occ exp [in ppm]) x (0.13/ppm)
For total cancer risk, low-exposure linear extrapolation from the total cancer extra risk estimates
for the 0.1 ppm 8-hour TWA exposure level presented in Table 4-30 of the assessment is recommended.
Both of the underlying models are linear in the low-exposure range (e.g., at the 0.1 ppm TWA and
below); thus, their sum is also linear. For 35-year exposures, the following formulae would apply:
95% UCL on extra risk for total cancer incidence ~ (8-h TWA occ exp [in ppm]) x (0.081/0.1
ppm) = (8-h TWA occ exp [in ppm]) x (0.81/ppm)
MLE of extra risk for total cancer incidence = (8-h TWA occ exp [in
ppm]) x (0.037/0. lppm) = (8-h TWA occ exp [in ppm]) x (0.37/ppm)
4-111
-------
5. REFERENCES
Adam, B; Bardos. H; Adany. R. (2005). Increased genotoxic susceptibility of breast epithelial cells to
ethylene oxide. Mutat Res 585: 120-126. http://dx.doi.Org/10.1016/i.mrgentox.2005.04.009
Albertini. RJ. (2001). HPRT mutations in humans: Biomarkers for mechanistic studies [Review], Mutat
Res 489: 1-16. http://dx.doi.org/10.1016/S1383-5742(01 )00064-3
Ambroise. D; Moulin. JJ; Squinazi, F; Protois. JC; Fontana. JM; Wild, P. (2005). Cancer mortality
among municipal pest-control workers. Int Arch Occup Environ Health 78: 387-393.
http://dx.doi.org/10.1007/s00420-004-0599-\
Appelgren. L, -E; Eneroth. G; Grant. C; Lanstrom. -L; Tenghagen. K. (1978). Testing of ethylene oxide
for mutagenicity using the micronucleus test in mice and rats. Acta Pharmacol Toxicol 43: 69-
71. http://dx.doi.org/10.1 1 1 1/i. 1600-0773.1978.tb02235.x
Arias. E. (2007). United States life tables, 2004. Atlanta, GA: Centers of Disease Control and
Prevention; National Center for Health Statistics.
http://www.cdc,gov/nchs/data/nvsr/nvsr56/nvsr56 09.pdf
Bakhoum. SF; Kabeche. L; Murnane. JP; Zaki. Bl; Compton. DA. (2014). DNA-damage response
during mitosis induces whole-chromosome missegregation. Cancer Discovery 4: 1281-1289.
http://dx.doi.org/10.1 158/2159-8290.CD-14-0403
Bastlova. T; Andersson. B; Lambert. B; Kolman. A. (1993). Molecular analysis of ethylene oxide-
induced mutations at the HPRT locus in human diploid fibroblasts. Mutat Res 287: 283-292.
BEIR (Committee on the Biological Effects of Ionizing Radiation). (1988). Health risks of radon and
other internally deposited alpha-emitters. In Health Risks of Radon and Other Internally
Deposited Alpha-Emitters. Washington, DC: National Academy Press.
http://dx.doi.org/10.17226/1026
Benson. LO; Teta. MJ. (1993). Mortality due to pancreatic and lymphopoietic cancers in chlorohydrin
production workers. Br J Ind Med 50: 710-716. http://dx.doi.org/10.1 136/oem.50.8.710
Beranek. DT. (1990). Distribution of methyl and ethyl adducts following alkylation with
monofunctional alkylating agents [Review], Mutat Res 231: 11-30.
Bisanti. L; Maggini. M; Raschetti. R; Alegiani. SS: Ippolito. FM; Caffari. B; Segnan. N: Ponti. A.
(1993). Cancer mortality in ethylene oxide workers. Br J Ind Med 50: 317-324.
Boffetta. P; van der Hel, O; Norppa. H; Fabianova. E; Fucic. A; Gundy. S; Lazutka. J; Cebulska-
Wasilewska. A: Puskailerova. D; Znaor. A: Kelecsenvi. Z; Kurtinaitis. J: Rachtan. J: Forni. A:
Vermeulen. R; Bonassi. S. (2007). Chromosomal aberrations and cancer risk: Results of a cohort
study from Central Europe. Am J Epidemiol 165: 36-43. http://dx.doi.org/10.1093/aje/kwj367
Bolt. HM. (1996). Quantification of endogenous carcinogens. The ethylene oxide paradox [Review],
Biochem Pharmacol 52: 1-5. http://dx.doi.org/10.1016/0006-2952(96)00085-8
Bolt. HM. (2000). Carcinogenicity and genotoxicity of ethylene oxide: new aspects and recent advances
[Review], Crit Rev Toxicol 30: 595-608. http://dx.doi.org/10.1080/1040844000895 1121
Bonassi. S: Znaor. A: Ceppi. M; Lando. C: Chang. WP; Holland. N: Kirsch-Volders. M; Zeiger. E; Ban.
S; Barale. R; Bigatti. MP; Bolognesi. C; Cebulska-Wasilewska. A; Fabianova. E; Fucic. A;
Hagmar. L; Joksic. G; Martelli. A; Migliore. L; Mirkova. E; Scarfi. MR; Zijno. A; Norppa. H;
Fenech. M. (2007). An increased micronucleus frequency in peripheral blood lymphocytes
predicts the risk of cancer in humans. Carcinogenesis 28: 625-631.
http ://dx. doi. org/10.1093/carcin/b gl 177
R-l
-------
Boogaard. PJ. (2002). Use of haemoglobin adducts in exposure monitoring and risk assessment
[Review], J Chromatogr B Analyt Technol Biomed Life Sci 778: 309-322.
http://dx.doi.org/10.1016/S0378-4347(01)00445-5
Bovsen. G: Pachkowski. BF; Nakamura. J: Swenberg. JA. (2009). The formation and biological
significance of N7-guanine adducts [Review], Mutat Res 678: 76-94.
http://dx.doi.org/10.1016/i.mrgentox.2009.05.006
Brown. CD; Asgharian. B; Turner. MJ; Fennell. TR. (1998). Ethylene oxide dosimetry in the mouse.
Toxicol Appl Pharmacol 148: 215-222. http://dx.doi.org/10.1006/taap. 1997.8349
Brown. CD; Wong, BA; Fennell. TR. (1996). In vivo and in vitro kinetics of ethylene oxide metabolism
in rats and mice. Toxicol Appl Pharmacol 136: 8-19. http://dx.doi.org/10.1006/taap. 1996.0002
CDC (Centers for Disease Control and Prevention). (2015). Underlying cause of death 1999-2014 on
CDC WONDER online database (Version 1999-2014) [Database], Atlanta, GA. Retrieved from
http://vvonder.cdc.gov/ucd-icd 10.html
Chaganti. SR.; Chen. W; Parsa. N; Offit. K; Louie. DC; Dalla-Favera. R; Chaganti. RSK. (1998).
Involvement of BCL6 in chromosomal aberrations affecting band 3q27 in B-cell non-Hodgkin
lymphoma. Genes Chromosomes Cancer 23: 323-327. http://dx.doi.org/10.1002/(S1C1) 1098-
2264( 199812)23:4<323: :AlD-GCC7>3.Q.CO;2-3
Christiansen. DH; Andersen. MK; Desta. F; Pedersen-Bjergaard. J. (2005). Mutations of genes in the
receptor tyrosine kinase (RTK)/RAS-BRAF signal transduction pathway in therapy-related
myelodysplasia and acute myeloid leukemia. Leukemia 19: 2232-2240.
http://dx.doi.org/10.103 8/si. leu.2404009
Christiansen. DH; Andersen. MK; Pedersen-Bjergaard. J. (2001). Mutations with loss of heterozygosity
of p53 are common in therapy-related myelodysplasia and acute myeloid leukemia after
exposure to alkylating agents and significantly associated with deletion or loss of 5q, a complex
karyotype, and a poor prognosis. J Clin Oncol 19: 1405-1413.
http://dx.doi.org/10.1200/jco.2001.19.5.1405
Clare. MG; Dean. BJ; de Jong. G; van Sittert. NJ. (1985). Chromosome analysis of lymphocytes from
workers at an ethylene oxide plant. Mutat Res 156: 109-1 16. http://dx.doi.org/10.1016/0165-
1218(85)90013-8
Clevvell. HJ; Teeguarden. J; McDonald. T; Sarangapani. R; Lawrence. G; Covington. T; Gentry. R;
Shi pp. A. (2002). Review and evaluation of the potential impact of age- and gender-specific
pharmacokinetic differences on tissue dosimetry [Review], Crit Rev Toxicol 32: 329-389.
http://dx.doi.org/10.1080/20024091064264
Coggon. D; Harris. EC; Poole. J; Palmer. KT. (2004). Mortality of workers exposed to ethylene oxide:
Extended follow up of a British cohort. Occup Environ Med 61: 358-362.
http://dx.doi.org/10.1 136/oem.2003.008268
Crump. KS; Bussard. DA; Chen. C; Jinot. J; Subramaniam. R. (2014). The bottom-up approach does not
necessarily bound low-dose risk [Letter], Regul Toxicol Pharmacol 70: 735-736.
http://dx.doi.org/10.1016/j.yrtph.2014.10.008
Csanady. GA; Denk. B; Putz, C; Kreuzer. PE; Kessler. W; Baur. C; Gar gas. ML; Filser. JG. (2000). A
physiological toxicokinetic model for exogenous and endogenous ethylene and ethylene oxide in
rat, mouse, and human: Formation of 2-hydroxyethyl adducts with hemoglobin and DNA.
Toxicol Appl Pharmacol 165: 1-26. http://dx.doi.org/10.1006/taap.2000.8918
Dellarco. VL; Generoso. WM; Sega. GA; Fovvle. J. R.; Jacobson-Kram. D. (1990). Review of the
mutagenicity of ethylene oxide [Review], Environ Mol Mutagen 16: 85-103.
http://dx.doi.org/10.1002/em.2850160207
R-2
-------
Donner. EM; Wong. BA; James. RA; Preston. RJ. (2010). Reciprocal translocations in somatic and germ
cells of mice chronically exposed by inhalation to ethylene oxide: Implications for risk
assessment. Mutagenesis 25: 49-55. http://dx.doi.org/10.1093/mutage/gep042
Dunkelberg. H. (1982). Carcinogenicity of ethylene oxide and 1,2-propylene oxide upon intragastric
administration to rats. Br J Cancer 46: 924-933. http://dx.doi.org/10.1038/bjc. 1982.303
Ehrenberg. L; Hussain. S. (1981). Genetic toxicity of some important epoxides [Review], DNA Repair
86: 1-113. http://dx.doi.org/10.1016/0165-1 1 10(81 )90034-8
Embree. JW; Lyon. JP; Hine. CH. (1977). The mutagenic potential of ethylene oxide using the
dominant—lethal assay in rats. Toxicol Appl Pharmacol 40: 261-267.
http://dx.doi.Org/l 0.1016/0041 -008X(77)90096-5
EPIC (Ethylene Oxide Industry Council). (2001). Toxicological review of ethylene oxide in support of
summary information on the integrated risk information system. Arlington, VA.
Farooqi. Z; Tornqvist, M; Ehrenberg. L; Natarajan. AT. (1993). Genotoxic effects of ethylene oxide and
propylene oxide in mouse bone marrow cells. Mutat Res 288: 223-228.
http://dx.doi.org/10.1016/0027-5 107(93 )90088-W
Fennell. TR; Brown, CD. (2001). A physiologically based pharmacokinetic model for ethylene oxide in
mouse, rat, and human. Toxicol Appl Pharmacol 173: 161-175.
http://dx.doi.org/10.1006/taap.2Q01.9184
Filser. JG; Denk. B; Tornqvist. M; Kessler. W; Ehrenberg. L. (1992). Pharmacokinetics of ethylene in
man; body burden with ethylene oxide and hydroxyethylation of hemoglobin due to endogenous
and environmental ethylene. Arch Toxicol 66: 157-163.
Fost. U; Hallier. E; Ottenwalder. H; Bolt. HM; Peter. H. (1991). Distribution of ethylene oxide in human
blood and its implications for biomonitoring. Hum Exp Toxicol 10: 25-31.
Galloway. SM; Berry. PK; Nichols. WW: Wolman. SR.; Soper. KA; Stollev. PD; Archer. P. (1986).
Chromosome aberrations in individuals occupationally exposed to ethylene oxide, and in a large
control population. Mutat Res Genet Toxicol 170: 55-74. http://dx.doi.org/10.1016/0165-
1218(86)90082-0
Gardner. MJ; Coggon, D; Pannett. B; Harris. EC. (1989). Workers exposed to ethylene oxide: A follow
up study. Occup Environ Med 46: 860-865. http://dx.doi.org/10.1 136/oem.46.12.860
Garman. RH; Snellings. WM; Maronpot. RR. (1985). Brain tumors in F344 rats associated with chronic
inhalation exposure to ethylene oxide. Neurotoxicology 6: 117-137.
Garman. RH; Snel lings. WM; Maronpot. RR. (1986). Frequency, size and location of brain tumours in
F-344 rats chronically exposed to ethylene oxide. Food Chem Toxicol 24: 145-153.
http://dx.doi.org/10.1016/0278-6915(86)90349-2
Garry. VF; Hozier. J; Jacobs. D; Wade, RL; Gray. DG. (1979). Ethylene oxide: Evidence of human
chromosomal effects. Environ Mutagen 1: 375-382. http://dx.doi.org/10.1002/em.2860010410
Gelehrter. TP; Collins. FS; Ginsburg. D. (1990). Principles of medical genetics. In Principles of Medical
Genetics. Baltimore, MD: Williams & Wilkins.
Generoso. WM; Cain. KT; Cornett. CV; Cacheiro. NLA; Hughes. LA. (1990). Concentration-response
curves for ethylene-oxide-induced heritable translocations and dominant lethal mutations.
Environ Mol Mutagen 16: 126-131. http://dx.doi.org/10.1002/em.2850160209
Generoso. WM; Cain. KT; Hughes. LA; Sega. GA; Braden. PW; Gosslee. DG; Shelby. MD. (1986).
Ethylene oxide dose and dose-rate effects in the mouse dominant-lethal test. Environ Mol
Mutagen 8: 1-7. http://dx.doi.org/10.1002/em.2860080102
R-3
-------
Generoso. WM; Cain. KT; Krishna. M; Sheu. CW; Grvder. RM. (1980). Heritable translocation and
dominant-lethal mutation induction with ethylene oxide in mice. Mutat Res 73: 133-142.
http://dx.d0i.0rg/l 0.1016/0027-5107(80)90142-6
Generoso. WM: Gumming. RB; Bandy. JA; Cain. KT. (1983). Increased dominant-lethal effects due to
prolonged exposure of mice to inhaled ethylene oxide. Mutat Res 119: 377-379.
http://dx.doi.org/10.1016/0165-7992(83 )90188-4
Godderis. L; Aka. P; Matecuca. R; Kirsch-Volders. M; Li son. D; Veulemans. H. (2006). Dose-
dependent influence of genetic polymorphisms on DNA damage induced by styrene oxide,
ethylene oxide and gamma-radiation. Toxicology 219: 220-229.
http://dx.doi.org/10.1016/j.tox.2005.1 1.021
Golberg. L. (1986). Chemical and physical properties. In Hazard Assessment of Ethylene Oxide. Boca
Raton, FL: CRC Press.
Greenberg. HL; Ott. MG; Shore. RE. (1990). Men assigned to ethylene oxide production or other
ethylene oxide related chemical manufacturing: A mortality study. Br J Ind Med 47: 221-230.
http://dx.doi.org/10.1 136/oem.47.4.221
Greife. AL; Hornung. RW; Stayner. LG; Steenland. KN. (1988). Development of a model for use in
estimating exposure to ethylene oxide in a retrospective cohort mortality study. Scand J Work
Environ Health 1: 29-30.
Hagmar. L; Mikoczy. Z; Welinder. H. (1995). Cancer incidence in Svvedish sterilant workers exposed to
ethylene oxide. Occup Environ Med 52: 154-156. http://dx.doi.org/10.1 136/oem.52.3.154
Hagmar. L; Stromberg. U; Bonassi. S: Hansteen. IL; Knudsen. LE; Lindholm. C: Norppa. H. (2004).
Impact of types of lymphocyte chromosomal aberrations on human cancer risk: Results from
Nordic and Italian cohorts. Cancer Res 64: 2258-2263. http://dx.doi.org/10.1158/0008-
5472.CAN-03-3360
Hagmar. L; Welinder. H; Linden. K; Attewell. R; Osterman-Golkar. S; Tornqvist, M. (1991). An
epidemiological study of cancer risk among workers exposed to ethylene oxide using
hemoglobin adducts to validate environmental exposure assessments. Int Arch Occup Environ
Health 63: 271-277. http://dx.doi.org/10.1007/BF00386377
Hansen. JP; Allen. J: Brock. K; Falconer. J: Helms. MJ; Shaver. GC; Strohm. B. (1984). Normal sister
chromatid exchange levels in hospital sterilization employees exposed to ethylene oxide. J
Occup Med 26: 29-32.
Harada. H; Harada. Y; Tanaka. H; Kimura. A; Inaba. T. (2003). Implications of somatic mutations in the
AML1 gene in radiation-associated and therapy-related myelodysplastic syndrome/acute myeloid
leukemia. Blood 101: 673-680. http://dx.doi.org/10.1 182/blood-2002-04-1010
Harris. ML: Jaffe. ES; Diebold. J: Flandrin. G; Muller-Hermelink. HK; Vardiman. J: Lister. TA;
Bloomfield. CD. (1999). World Health Organization classification of neoplastic diseases of the
hematopoietic and lymphoid tissues: Report of the Clinical Advisory Committee meeting—Airlie
House, Virginia, November 1997. J Clin Oncol 17: 3835-3849.
Haufroid. V: Merz. B; Hofmann. A; T schopp. A; Li son. D; Hotz. P. (2007). Exposure to ethylene oxide
in hospitals: Biological monitoring and influence of glutathione S-transferase and epoxide
hydrolase polymorphisms. Cancer Epidemiol Biomarkers Prev 16: 796-802.
http://dx.doi.org/10.1 158/1055-9965.EP1-06-0915
Health Canada. (2001). Priority substances list assessment report: Ethylene oxide. (Cat. no. En40-
215/5 IE). Ottawa, Ontario: Environment Canada, Health Canada, http://www.hc-sc.gc.ca/evvh-
semt/pubs/contaminants/psl2-lsp2/ethylene oxide/index-eng.php
Hill. AB. (1965). The environment and disease: Association or causation? Proc R Soc Med 58: 295-300.
R-4
-------
Hogstedt. B; Gullberg. B; Hedner. K; Kolnig. A. -M; Mitel man. F; Skerfving. S; Widegren. B. (1983).
Chromosome aberrations and micronuclei in bone marrow cells and peripheral blood
lymphocytes in humans exposed to ethylene oxide. Hereditas 98: 105-113.
http://dx.doi.org/10.1111/i. 1601-5223.1983.tb00585.x
Hogstedt. C; Aringer. L; Gustavsson. A. (1986). Epidemiologic support for ethylene oxide as a cancer-
causing agent. JAMA255: 1575-1578. http://dx.doi.org/10.1001/jama. 1986.03370120053022
Hogstedt. C: Rohlen. O; Berndtsson. BS; Axel son. O; Ehrenberg. L. (1979). A cohort study of mortality
and cancer incidence in ethylene oxide production workers. Occup Environ Med 36: 276-280.
http://dx.doi.org/10.1 136/oem.36.4.276
Hogstedt. LC. (1988). Epidemiological studies on ethylene oxide and cancer: An updating. Lyon,
France.
Hong. H, -HL; Houle. CD; Ton, T, -VT; Sills. RC. (2007). K-ras mutations in lung tumors and tumors
from other organs are consistent with a common mechanism of ethylene oxide tumorigenesis in
the B6C3F1 mouse. Toxicol Pathol 35: 81-85. http://dx.doi.org/10.1080/01926230601Q63839
Hornung. RW; Greife. AL; Stayner. LT; Steenland. NK; Herrick. RF; Elliott. LJ; Ringenburg. VL;
Morawetz. J. (1994). Statistical model for prediction of retrospective exposure to ethylene oxide
in an occupational mortality study. Am J Ind Med 25: 825-836.
http://dx.doi.org/10.1002/ajim.4700250607
Houle. CD; Ton. T. - VT; Clayton. N: Huff. J: Hong. H. -HL: Sills. RC. (2006). Frequent p53 and H-ras
mutations in benzene- and ethylene oxide-induced mammary gland carcinomas from B6C3F1
mice. Toxicol Pathol 34: 752-762. http://dx.doi.org/10.1080/0192623060Q935912
Howlader. N; Noone. AM; Krapcho. M; Garshell. J; Miller. D; Altekruse. SF; Kosary. C, 1; Yu. M;
Ruhl. J; Tatalovich. Z; Mariotto. A; Lewis, DR.; Chen. HS; Feuer. EJ; Cronin. KA. (2014). SEER
cancer statistics review, 1975-2012. Bethesda, MD: National Cancer Institute.
http://seer.cancer.gov/archive/csr/1975 2012/
Hu, JJ; Smith. TR; Miller. MS; Lohman. K; Case. LP. (2002). Genetic regulation of ionizing radiation
sensitivity and breast cancer risk. Environ Mol Mutagen 39: 208-215.
http://dx.doi.org/10.1002/em. 10058
IARC (International Agency for Research on Cancer). (1994a). IARC monographs on the evaluation of
carcinogenic risks to humans: Ethylene (pp. 45-71). Lyon, France: International Agency for
Research on Cancer. http://monographs.iarc.fr/ENG/Monographs/vol60/
IARC (International Agency for Research on Cancer). (1994b). IARC monographs on the evaluation of
carcinogenic risks to humans: Ethylene oxide (pp. 73-159). Lyon, France: International Agency
for Research on Cancer. http://monographs.iarc.fr/ENG/Monographs/vol60/
IARC (International Agency for Research on Cancer). (2008). IARC monographs on the evaluation of
carcinogenic risks to humans: 1,3-butadiene, ethylene oxide and vinyl halides (vinyl fluoride,
vinyl chloride and vinyl bromide) [IARC Monograph], Lyon, France: International Agency for
Research on Cancer. http://monographs.iarc.fr/ENG/Monographs/vol97/
Ingvarsson. S. (1999). Molecular genetics of breast cancer progression [Review], Semin Cancer Biol 9:
277-288. http://dx.doi.org/10.1006/scbi. 1999.0124
Jenssen. D; Ramel. C. (1980). The micronucleus test as part of a short-term mutagenicity test program
for the predicition of carcinogenicity evaluated by 143 agents tested. Mutat Res 75: 191-202.
http://dx.doi.org/10.1016/0165-l 1 10(80)90014-7
J oh an son. G; Filser. JG. (1992). Experimental data from closed chamber gas uptake studies in rodents
suggest lower uptake rate of chemical than calculated from literature values on alveolar
ventilation. Arch Toxicol 66: 291-295. http://dx.doi.org/10.1007/BF02307176
R-5
-------
Kardos. L; Szeles. G; Gombkoto. G; Szeremi. M; Tompa. A; Adanv. R. (2003). Cancer deaths among
hospital staff potentially exposed to ethylene oxide: An epidemiological analysis. Environ Mol
Mutagen 42: 59-60. http://dx.doi.org/10.1002/em.10167
Karelova. J: Jablonicka. A: Vargova. M. (1987). Results of cytogenetic testing of workers exposed to
ethylene oxide. J Hyg Epidemiol Microbiol Immunol 31: 119-126.
Kiesselbach. N; Ulm. K; Lange, H. -J; Korallus. U. (1990). A multicentre mortality study of workers
exposed to ethylene oxide. Br J Ind Med 47: 182-188.
Kirman. CR; Sweeney, LM; Teta. Ml; Sielken. RL; Valdez-Flores. C; Albertini. RJ; Gar gas. ML.
(2004). Addressing nonlinearity in the exposure-response relationship for a genotoxic
carcinogen: Cancer potency estimates for ethylene oxide. Risk Anal 24: 1165-1183.
http://dx.doi.org/10.111 l/i.0272-4332.2004.005 17.x
Kligerman. AD; Erexson. GL; Phelps. ME; Wilmer. JL. (1983). Sister-chromatid exchange induction in
peripheral blood lymphocytes of rats exposed to ethylene oxide by inhalation. Mutat Res Lett
120: 37-44. http://dx.doi.org/10.1016/0165-7992(83)90071 -4
Kolman. A; Chovanec. M. (2000). Combined effects of gamma-radiation and ethylene oxide in human
diploid fibroblasts. Mutagenesis 15: 99-104. http://dx.doi.org/10.1093/mutage/15.2.99
Kolman. A; Chovanec. M; Osterman-Golkar. S. (2002). Genotoxic effects of ethylene oxide, propylene
oxide and epichlorohydrin in humans: Update review (1990-2001) [Review], Mutat Res 512:
173-194. http://dx.doi.org/10.1016/S1383-5742(02)00067-4
Krewski. D; Crump, KS; Farmer. J; Gaylor. DW; Howe, R; Portier. C; Salsburg. D; Sielken. RL; Van
Rvzin. J. (1983). A comparison of statistical methods for low dose extrapolation utilizing time-
to-tumour data. Fundam Appl Toxicol 3: 140-160. http://dx.doi.org/10.1016/S0272-
0590(83 )80075-X
Krishnan. K; Gar gas. ML; Fennel 1. TR; Andersen. ME. (1992). A physiologically based description of
ethylene oxide dosimetry in the rat. Toxicol Ind Health 8: 121-140.
http://dx.doi.org/10.1 177/074823379200800301
Lambert. B; Andersson. B; Bastlova. T; Hou. S. -M; Hellgren. D; Kolman. A. (1994). Mutations
induced in the hypoxanthine phosphoribosyl transferase gene by three urban air pollutants:
Acetaldehyde, benzo[a]pyrene diolepoxide, and ethylene oxide. Environ Health Perspect Suppl
102: 135-138. http://dx.doi.org/10.2307/343 1943
Langholz. B; Richardson. DB. (2010). Fitting general relative risk models for survival time and matched
case-control analysis. Am J Epidemiol 171: 377-383. http://dx.doi.org/10.1093/aje/kwp403
Laurent. CH; Frederic. J; Leonard. AY. (1984). Sister chromatid exchange frequency in workers
exposed to high levels of ethylene oxide, in a hospital sterilization service. Int Arch Occup
Environ Health 54: 33-43. http://dx.doi.onV 10.1007/BF00378726
Lerda. D; Rizzi. R. (1992). Cytogenetic study of persons occupationally exposed to ethylene oxide.
Mutat Res 281: 31-37. http://dx.doi.org/10.1016/0165-7992(92)90033-E
Lewis, SE; Barnett. LB; Felton. C; Johnson. FM; Skovv. LC; Cacheiro. N; Shelby. MP. (1986).
Dominant visible and electrophoretically expressed mutations induced in male mice exposed to
ethylene oxide by inhalation. Environ Mol Mutagen 8: 867-872.
http://dx.doi.org/10.1002/em.2860080609
Lieber. MR. (2010). The mechanism of double-strand DNA break repair by the nonhomologous DNA
end-joining pathway. Annu Rev Biochem 79: 181-211.
http://dx.doi.org/10.1 146/annurev.biochem.052308.093 13 1
R-6
-------
Liou. SH; Lung. JC; Chen. YH; Yang. T; Hsieh. LL; Chen. CJ; Wu, TN. (1999). Increased
chromosome-type chromosome aberration frequencies as biomarkers of cancer risk in a
blackfoot endemic area. Cancer Res 59: 1481-1484.
Lorenti Garcia. C; Darroudi. F; Tates. AD; Nataraian. AT. (2001). Induction and persistence of
micronuclei, sister-chromatid exchanges and chromosomal aberrations in splenocytes and bone-
marrow cells of rats exposed to ethylene oxide. Mutat Res Genet Toxicol Environ Mutagen 492:
59-67. http://dx.doi.org/10.1016/S1383-5718(01 )00149-8
Lossos. IS; Levy. R. (2000). Mutation analysis of the 5' noncoding regulatory region of the BCL-6 gene
in non-Hodgkin lymphoma: Evidence for recurrent mutations and intraclonal heterogeneity.
Blood 95: 1400-1405.
Lynch. DW; Lewis. TR; Moorman. WJ; Burg. JR.; Groth. DH; Khan. A; Ackerman. LJ; Cockrell. BY.
(1984a). Carcinogenic and toxicologic effects of inhaled ethylene oxide and propylene oxide in
F344 rats. Toxicol Appl Pharmacol 76: 69-84. http://dx.doi.org/10.1016/0041 -008X(84)90030-9
Lynch. DW; Lewis. TR; Moorman. WJ; Burg. JR.; Gulati. DK; Kaur. P; Sabharwal. PS. (1984b). Sister-
chromatid exchanges and chromosome aberrations in lymphocytes from monkeys exposed to
ethylene oxide and propylene oxide by inhalation. Toxicol Appl Pharmacol 76: 85-95.
http://dx.doi.org/10.1016/0041 -008X(84)90031 -0
Lynch. DW; Lewis. TR; Moorman. WJ; Burg. JR; Lai. JB; Setzer. JV; Groth. DH; Gulati. DK; Zavos.
PM; Sabharwal. PS; Ackerman. LJ; Cockrell. BY; Sprinz. H. (1984c). Effects on monkeys and
rats of long-term inhalation exposure to ethylene oxide: Major findings of the NIOSH study. In
Inhospital ethylene oxide sterilization: Current issues in ethylene oxide toxicity and occupational
exposure (pp. 7-10). (AAMI Technology Assessment Report No. 8-84). Arlington, VA:
Association for the Advancement of Medical Instrumentation.
Major. J; Jakab. MG; Tompa. A. (1996). Genotoxicological investigation of hospital nurses
occupationally exposed to ethylene-oxide: I. Chromosome aberrations, sister-chromatid
exchanges, cell cycle kinetics, and UV-induced DNA synthesis in peripheral blood lymphocytes.
Environ Mol Mutagen 27: 84-92. http://dx.doi.org/10.1002/(S1CD1098-
2280( 1996)27:2<84: :A1D-EM2>3.0.CQ;2-E
Major. J; Jakab. MG; Tompa. A. (2001). Genotoxicological investigation of hospital nurses
occupationally exposed to ethylene oxide. II. HPRT mutation frequencies. Central Eur J Occup
Env Med 7: 195-208.
Marsden. DA; Jones. DJ; Britton. RG; Ognibene. T; Ubick. E; Johnson. GE; Farmer. PB; Brown. K.
(2009). Dose-response relationships for N7-(2-hydroxyethyl)guanine induced by low-dose
[14C]ethylene oxide: Evidence for a novel mechanism of endogenous adduct formation. Cancer
Res 69: 3052-3059. http://dx.doi.org/10.1 158/0008-5472.CAN-08-4233
Marsden. DA; Jones. DJ; Lamb, JH; Tompkins. EM; Farmer, PB; Brown, K. (2007). Determination of
endogenous and exogenously derived N7-(2-hydroxyethyl)guanine adducts in ethylene oxide-
treated rats. Chem Res Toxicol 20: 290-299. http://dx.doi.org/10.102 l/tx600264t
Mayer. J; Warburton. D; Jeffrey. AM; Pero. R; Walles. S; Andrews, L; Toor. M; Latriano. L; Wazneh.
L; Tang. D; Tsai. WY; Kuroda. M; Perera. F. (1991). Biologic markers in ethylene oxide
exposed workers and controls. Mutat Res 248: 163-176. http://dx.doi.org/10.1016/0027-
5107(91)90098-9
Mazon. G; Philippin. G; Cadet. J; Gasparutto. D; Fuchs. RP. (2009). The alkyltransferase-like ybaZ
gene product enhances nucleotide excision repair of 0(6)-alkylguanine adducts in E. coli. DNA
Repair 8: 697-703. http://dx.doi.org/10.1016/j.dnarep.2009.01.022
R-7
-------
Memisoglu, A; Samson. L. (2000). Base excision repair in yeast and mammals [Review], Mutat Res
451: 39-51. http://dx.doi.org/10.1016/S0027-5107(00)00039-7
Mertes. I; Fleischmann. R; Glatt. H; Oesch. F. (1985). Interindividual variations in the activities of
cytosolic and microsomal epoxide hydrolase in human liver. Carcinogenesis 6: 219-223.
http://dx.doi.Org/10.1093/carcin/6.2.219
Mikoczy. Z; Tinnerberg, H; Bjork. J; Albin. M. (201 1). Cancer incidence and mortality in Swedish
sterilant workers exposed to ethylene oxide: Updated cohort study findings 1972-2006. Int J
Environ Res Public Health 8: 2009-2019. http://dx.doi.org/10.3390/iierph80620Q9
Mini no. AM; Arias. E; Kochanek. KD; Murphy, SL; Smith. BL. (2002). Deaths: Final data for 2000 (pp.
1-119). Hyattsville, MD: National Center for Health Statistics.
http ://www. cdc. gov/nchs/ data/nvsr/nvsr5 0/nvsr50 15.pdf
Murphy, SL; Kochanek. KD; Xu, J; Arias. E. (2015). Mortality in the United States, 2014. In NCHS
Data Brief, No 229. (NCHS Data Brief No. 229). Hyattsville, MD: U.S. Department of Health
and Human Services, National Center for Health Statistics.
http ://www. cdc. gov/nchs/ data/ databriefs/ db229.pdf
Natarajan. AT; Preston. RJ; Dellarco. V; Ehrenberg, L; Generoso. W; Lewis, S; Tates. AD. (1995).
Ethylene oxide: Evaluation of genotoxicity data and an exploratory assessment of genetic risk
[Review], Mutat Res 330: 55-70. http://dx.doi.org/10.1016/0027-5 107(95)00036-l
NCI (National Cancer Institute). (2004). Breast cancer screening (PDQ): Health professional version.
Washington, DC: U.S. Department of Health, Education, and Welfare, National Institutes of
Health. http://vvvvvv.cancer.gov/cancertopics/pdci/screening/breast/HealthProfessional/
Norman. SA; Berlin. JA; Soper, KA; Middendorf. BF; Stolley. PP. (1995). Cancer incidence in a group
of workers potentially exposed to ethylene oxide. Int J Epidemiol 24: 276-284.
http://dx.doi.org/10.1093/iie/24.2.276
NRC (National Research Council). (1983). Risk assessment in the federal government: Managing the
Process. In Risk Assessment in the Federal Government: Managing the Process. Washington,
DC: National Academy Press, http://www.nap.edu/read/366/chapter/1
NRC (National Research Council). (2011). Review of the Environmental Protection Agency's draft IRIS
assessment of formaldehyde. Washington, DC: The National Academies Press.
http://dx.doi.org/10.17226/13 142
NRC (National Research Council). (2014). Review of EPA's Integrated Risk Information System (IRIS)
process. In Review of EPA's Integrated Risk Information System (IRIS) Process. Washington,
DC: The National Academies Press, http://vvvvvv.nap.edu/catalog.php7record id= 18764
NTP (National Toxicology Program). (1987). Toxicology and carcinogenesis studies of ethylene oxide
(CAS no 75-21-8) in B6C3F1 mice (inhalation studies). Natl Toxicol Program Tech Rep Ser
326: 1-114.
01 sen. GW; Lacy. SE; Bodner. KM; Chau. M; Arceneaux. TG; Cartmill. JB; Ramlow. JM; Bosvvell. JM.
(1997). Mortality from pancreatic and lymphopoietic cancer among workers in ethylene and
propylene chlorohydrin production. Occup Environ Med 54: 592-598.
http://dx.doi.org/10.1 136/oem.54.8.592
Ong. T; Bi. H. -K; Xing. S; Stewart. J; Moorman. W. (1993). Induction of sister chromatid exchange in
spleen and bone marrow cells of rats exposed by inhalation to different dose rates of ethylene
oxide. Environ Mol Mutagen 22: 147-15 1. http://dx.doi.org/10.1002/em.2850220306
Pauvvels. W; Veulemans. H. (1998). Comparison of ethylene, propylene and styrene 7,8-oxide in vitro
adduct formation on N-terminal valine in human haemoglobin and on N-7-guanine in human
DNA. Mutat Res 418: 21-33. http://dx.doi.org/10.1016/S1383-5718(98)00106-5
R-8
-------
Paz-v-Mino. C; Perez. JC; Fiallo. BF; Leone. PE. (2002). A polymorphism in the hMSH2 gene
(gIVS12-6T>C) associated with non-Hodgkin lymphomas. Cancer Genet Cytogenet 133: 29-33.
http://dx.doi.oru/10.1016/SO165-4608(01 )00547-7
Pedersen-Biergaard. J: Christiansen. DH; Desta. F: Andersen. MK. (2006). Alternative genetic pathways
and cooperating genetic abnormalities in the pathogenesis of therapy-related myelodysplasia and
acute myeloid leukemia [Review], Leukemia 20: 1943-1949.
http://dx.doi.oru/10.1038/si.leu.2404381
Pero. RW; Widegren, B; Houstedt. B; Mitelman. F. (1981). In vivo and in vitro ethylene oxide exposure
of human lymphocytes assessed by chemical stimulation of unscheduled DNA synthesis. Mutat
Res 83: 271-289. http://dx.doi.oru/10.1016/0027-5 107(81 )9001 1 -7
Pfeiffer. P; Goedecke. W; Obe. G. (2000). Mechanisms of DNA double-strand break repair and their
potential to induce chromosomal aberrations [Review], Mutagenesis 15: 289-302.
http://dx.doi.oru/ 10.1093/mutage/15.4.289
Popp. W: Vahrenholz. C: Przvuoda. H; Brauksiepe. A: Goch. S: Muller. G: Schell. C: Norpoth. K.
(1994). DNA-protein cross-links and sister chromatid exchange frequencies in lymphocytes and
hydroxy ethyl mercapturic acid in urine of ethylene oxide-exposed hospital workers. Int Arch
Occup Environ Health 66: 325-332. http://dx.doi.oru/10.1007/BF00378365
Preston. RJ. (1999). Cytogenetic effects of ethylene oxide, with an emphasis on population monitoring
[Review], Crit Rev Toxicol 29: 263-282. http://dx.doi.oru/10.1080/10408449991349212
Preston. RJ; Abernethy. DJ. (1993). Studies on the induction of chromosomal aberration and si ster
chromatid exchange in rats exposed to styrene by inhalation. IARC Sci Publ 127: 225-233.
Preston. RJ; Fennell. TR; Leber. AP; Sielken. RL, Jr; Swenberu. JA. (1995). Reconsideration of the
genetic risk assessment for ethylene oxide exposures. Environ Mol Mutagen 26: 189-202.
http://dx.doi.oru/10.1002/em.2850260303
Preudhomme. C; Warot-Loze. D; Roumier. C; Grardel-Duflos. N; Garand. R; Lai. JL; Dastuuue. N;
Maclntyre. E; Denis. C; Bauters. F; Kerckaert. JP; Cosson. A; Fenaux. P. (2000). High incidence
of biallelic point mutations in the Runt domain of the AML1/PEBP2 alpha B gene in Mo acute
myeloid leukemia and in myeloid malignancies with acquired trisomy 21. Blood 96: 2862-2869.
Rapoport. IA. (1948). The effect of ethylene oxide, glycide and glycol on genetic mutations. Dokl
Biochem Biophys 60: 469-472.
Recio. L; Dormer. M; Abernethy. D; Pluta. L; Steen. A. -M; Wong. BA; James. A; Preston. RJ. (2004).
In vivo mutagenicity and mutation spectrum in the bone marrow and testes of B6C3F1 lacl
transgenic mice following inhalation exposure to ethylene oxide. Mutagenesis 19: 215-222.
http://dx.doi.oru/10.1093/mutaue/ uehO 17
Ribeiro. LR; Rabello-Gay. MN; Salvadori. DMF; Pereira. CAB; Becak. W. (1987). Cytogenetic effects
of inhaled ethylene oxide in somatic and germ cells of mice. Arch Toxicol 59: 332-335.
http://dx.doi.oru/10.1007/BF00295085
Ribeiro. LR; Salvadori. DM; Rios. AC; Costa. SL; Tates. AD; Tornqvist, M; Natarajan. AT. (1994).
Biological monitoring of workers occupationally exposed to ethylene oxide. Mutat Res 313: 81-
87. http://dx.doi.oru/10.1016/0165-1 161(94)90035-3
Richmond. GW; Abrahams. RH; Nemenzo. JH; Hine. CH. (1985). An evaluation of possible effects on
health following exposure to ethylene oxide. Arch Environ Occup Health 40: 20-25.
http://dx.doi.oru/10.1080/00039896.1985.10545884
Ries. LAG; Eisner. MP; Kosarv. CL; Hankev. BF; Miller. BA; Clegg. L; Mariotto. A; Feuer. EF;
Edwards, BK. (2004). SEER (Surveillance Epidemiology and End Results) cancer statistics
R-9
-------
review, 1975-2001. Bethesda, MD: National Cancer Institute, U.S. Department of Health,
Education, and Welfare, National Institutes of Health, http ://seer.cancer, gov/csr/1975 2001
Ries. LAG; Melbert. D; Krapcho. M; Mariotto. A; Miller. BA; Feuer. EJ; Clegg, L; Horner, MJ;
Howlader. N; Eisner. MP; M. R; BK. E. (2007). SEER (surveillance epidemiology and end
results) cancer statistics review, 19752004. Bethesda, MD: National Cancer Institute, U.S.
Department of Health, Education, and Welfare, National Institutes of Health,
http://seer.cancer.gov/csr/1975 2004
Rossner. P; Boffetta. P; Ceppi. M; Bonassi. S; Smerhovsky. Z; Landa. K; Juzova. D; Sram, RJ. (2005).
Chromosomal aberrations in lymphocytes of healthy subjects and risk of cancer. Environ Health
Perspect 113: 517-520. http://dx.doi.org/10.1289/ehp.6925
Rothman. KJ. (1986). Modern epidemiology. In Modern Epidemiology (1 ed.). Boston, MA: Little
,Brown & Co.
Russo. JFC; Russo. IH. (1999). Cellular basis of breast cancer susceptibility [Review], Oncol Res 11:
169-178.
Rusyn. I; Asakura. S; Li, Y; Kosyk. O; Koc. H; Nakamura. J; Upton. PB; Swenberg. J A. (2005). Effects
of ethylene oxide and ethylene inhalation on DNA adducts, apurinic/apyrimidinic sites and
expression of base excision DNA repair genes in rat brain, spleen, and liver. DNA Repair 4:
1099-1 110. http://dx.doi.org/10.1016/i.dnarep.2005.05.009
SAB (Science Advisory Board). (2007). Review of Office of Research and Development (ORD) draft
assessment entitled "Evaluation of the carcinogenicity of ethylene oxide". Washington, DC: U.S.
Environmental Protection Agency, Science Advisory Board.
http://yosemite.epa.gov/sab/sabproduct.nsf/368203f97a 15308a852574ba005bbd01/5D661BC1 18
B527A3 852573 B80068C97B/$File/EP A-SAB-08-004-unsigned.pdf
SAB (Science Advisory Board). (2015). Science Advisory Board Review of the EPAs evaluation of the
inhalation carcinogenicity of ethylene oxide: Revised external review draft - August 2014 [EPA
Report], (EPA-SAB-15-012). Washington, DC: U.S. Environmental Protection Agency, Science
Advisory Board.
https://yosemite.epa.gov/sab/sabproduct.nsf/fedrgstr activites/BD2B2DB4F84146A585257E9A
0070E655/$File/EPA-SAB-15-012+unsigned.pdf
Sarto. F; Clonfero. E; Bartolucci. GB; Franceschi. C; Chiricolo. M; Levis. AG. (1987). Sister chromatid
exchanges and DNA repair capability in sanitary workers exposed to ethylene oxide: Evaluation
of the dose-effect relationship. Am J Ind Med 12: 625-637.
http://dx.doi.org/10.1002/aiim.47001205 15
Sarto. F; Cominato. I; Pinton. AM; Brovedani. PG; Faccioli. CM; Bianchi. V; Levis. AG. (1984).
Cytogenetic damage in workers exposed to ethylene oxide. Mutat Res 138: 185-195.
http://dx.doi.org/10.1016/0165-1218(84)90043-0
Sarto. F; Tomanin. R; Giacomelli. L; Iannini. G; Cupiraggi. AR. (1990). The micronucleus assay in
human exfoliated cells of the nose and mouth: Application to occupational exposures to chromic
acid and ethylene oxide. Mutat Res 244: 345-35 1. http://dx.doi.org/10.1016/0165-
7992(90)90083-V
Sarto. F; Tornqvist. MA; Tomanin. R; Bartolucci. GB; Osterman-Golkar. SM; Ehrenberg. L. (1991).
Studies of biological and chemical monitoring of low-level exposure to ethylene oxide. Scand J
Work Environ Health 17: 60-64. http://dx.doi.org/10.5271/sjweh. 1733
Schulte. PA; Boeniger. M; Walker. IT; Schober. SE; Pereira. MA; Gulati. DK; Woiciechowski. JP;
Garza. A; Froelich. R; Strauss. G. (1992). Biologic markers in hospital workers exposed to low
R-10
-------
levels of ethylene oxide. Mutat Res 278: 237-251. http://dx.doi.org/10.1016/SQ165-
1218(10)80003-5
Schulte. PA; Walker. JT; Boeniger, MF; Tsuchiya. Y; Halperin. WE. (1995). Molecular, cytogenetic,
and hematologic effects of ethylene-oxide on female hospital workers. J Occup Environ Med 37:
313-320. http://dx.doi.org/10.1097/00043764-199503000-000Q8
Segerback. D. (1990). Reaction products in hemoglobin and DNA after in vitro treatment with ethylene
oxide andN-(2-hydroxyethyl)-N-nitrosourea. Carcinogenesis 11: 307-312.
Shi, Q; Wang, L, -E; Bondv. ML; Brewster, A; Singletary. SE; Wei. Q. (2004). Reduced DNA repair of
benzo[a]pyrene diol epoxide-induced adducts and common XPD polymorphisms in breast cancer
patients. Carcinogenesis 25: 1695-1700. http://dx.doi.org/10.1093/carcin/bgh 167
Sielken. RL, Jr; Valdez-Flores. C. (2009). Life-table calculations of excess risk for incidence versus
mortality: Ethylene oxide case study. Regul Toxicol Pharmacol 55: 82-89.
http://dx.doi.org/10.1016/i.yrtph.2009.06.003
Sisk. SC; Pluta. LJ; Mever. KG; Wong. BC; Recio. L. (1997). Assessment of the in vivo mutagenicity of
ethylene oxide in the tissues of B6C3F1 lacl transgenic mice following inhalation exposure.
Mutat Res 391: 153-164. http://dx.doi.org/10.1016/S1383-5718(97)00063-6
Snellings. WM; Weil. CS; Maronpot. RR. (1984). A two-year inhalation study of the carcinogenic
potential of ethylene oxide in Fischer 344 rats. Toxicol Appl Pharmacol 75: 105-117.
http://dx.doi.org/10.1016/0041 -008X(84)90081 -4
Solomon. JJ. (1999). Cyclic adducts and intermediates induced by simple epoxides. IARC Sci Publ 150:
123-135.
Starr. TB; Swenberg. J A. (2013). A novel bottom-up approach to bounding low-dose human cancer risks
from chemical exposures. Regul Toxicol Pharmacol 65: 311-315.
http://dx.doi.org/10.1016/i.vrtph.2013.01.004
Stayner. L; Steenland. K; Dosemeci. M; Hertz-Picciotto. I. (2003). Attenuation of exposure-response
curves in occupational cohort studies at high exposure levels. Scand J Work Environ Health 29:
3 17-324. http://dx.doi.org/10.5271 /siweh.737
Stayner. L; Steenland. K; Greife. A; Hornung. R; Hayes. RB; Nowlin. S; Morawetz. J; Ringenburg. V;
Elliot. L; Halperin. W. (1993). Exposure-response analysis of cancer mortality in a cohort of
workers exposed to ethylene oxide. Am J Epidemiol 138: 787-798.
Steenland. K; Deddens. J; Piacitelli. L. (2001). Risk assessment for 2,3,7,8-tetrachlorodibenzo-p-dioxin
(TCDD) based on an epidemiologic study. Am J Epidemiol 154: 451-458.
http://dx.doi.org/10.1093/aje/154.5.45 1
Steenland. K; Deddens. JA. (2004). A practical guide to dose-response analyses and risk assessment in
occupational epidemiology [Review], Epidemiology 15: 63-70.
http://dx.doi.org/10.1097/01 .ede.OOOO 100287.45004.e7
Steenland. K; Stayner. L; Deddens. J. (2004). Mortality analyses in a cohort of 18 235 ethylene oxide
exposed workers: Follow up extended from 1987 to 1998. Occup Environ Med 61: 2-7.
Steenland. K; Stayner. L; Greife. A; Halperin. W; Hayes. R; Hornung. R; Nowlin, S. (1991). Mortality
among workers exposed to ethylene oxide. N Engl J Med 324: 1402-1407.
http://dx.doi.org/10.1056/NEJM 199105 163242004
Steenland. K; Whelan. E; Deddens. J; Stayner. L; Ward. E. (2003). Ethylene oxide and breast cancer
incidence in a cohort study of 7576 women (United States). Cancer Causes Control 14: 531-539.
http://dx.doi.org/10.1023/A: 1024891529592
R-ll
-------
Stollev. PD; Soper. KA; Galloway. SM; Nichols. WW; Norman. SA; Wolman. SR. (1984). Sister-
chromatid exchanges in association with occupational exposure to ethylene oxide. Mutat Res
129: 89-102. http://dx.d0i.0rg/l0.1016/0027-5107(84)90127-1
Swaen. GMH; Burns. C: Teta. JM; Bodner. K; Keenan. D; Bodnar. CM. (2009). Mortality study update
of ethylene oxide workers in chemical manufacturing: A 15 year update. J Occup Environ Med
51: 714-723. http://dx.doi.org/10.1097/JQM.0b013e3181a2ca20
Swaen. GMH; Slangen. J MM; Ott. MG; Kusters. E; Van Den Langenbergh. G: Arends. JW; Zober. A.
(1996). Investigation of a cluster of ten cases of Hodgkin's disease in an occupational setting. Int
Arch Occup Environ Health 68: 224-228. http://dx.doi.org/10.1007/BF00381432
Swenberg. J A; Fedtke. N; Fennell. TR; Walker. VE. (1990). Relationships between carcinogen
exposure, DNA adducts and carcinogenesis. In DB Clayson; IC Munro; P Shubik; JA Swenberg
(Eds.), Progress in Predictive Toxicology (pp. 161-184). New York, NY: Elsevier.
Tates. AD; Boogaard. PJ; Darroudi. F; Natarajan. AT; Caubo. ME; van Sittert. NJ. (1995). Biological
effect monitoring in industrial workers following incidental exposure to high concentrations of
ethylene oxide. Mutat Res 329: 63-77. http://dx.doi.org/10.1016/0027-5 107(95 )00018-E
Tates. AD; Grummt, T; Tornqvist, M; Farmer. PB; van Dam, FJ; van Mossel. H; Schoemaker. HM;
Osterman-Golkar. S; Uebel. C; Tang. YS; Zwinderman. AH; Natarajan. AT; Ehrenberg. L.
(1991). Biological and chemical monitoring of occupational exposure to ethylene oxide. Mutat
Res 250: 483-497. http://dx.doi.org/10.1016/0027-5 107(91 )90205-3
Tates. AD; van Dam, FJ; Natarajan. AT; van Teylingen. CMM; de Zwart. FA; Zwinderman. AH; van
Sittert. NJ; Nil sen. A; Nil sen. OG; Zahlsen. K; Magnusson. A. -L; Tornqvist. M. (1999).
Measurement of HPRT mutations in splenic lymphocytes and haemoglobin adducts in
erythrocytes of Lewis rats exposed to ethylene oxide. Mutat Res 431: 397-415.
http://dx.doi.org/10.1016/S0027-5107(99)00182-7
Teta. MJ; Benson. LO; Vitale. JN. (1993). Mortality study of ethylene oxide workers in chemical
manufacturing: A 10 year update. Br J Ind Med 50: 704-709.
http://dx.doi.org/10.1 136/oem.50.8.704
Teta. MJ; Sielken. RL, Jr; Valdez-Flores. C. (1999). Ethylene oxide cancer risk assessment based on
epidemiological data: Application of revised regulatory guidelines. Risk Anal 19: 1135-1155.
http://dx.doi.org/10.1111/j. 1539-6924.1999.tb01 134.x
Tomkins. DJ; Haines. T; Lawrence. M; Rosa. N. (1993). A study of sister chromatid exchange and
somatic cell mutation in hospital workers exposed to ethylene oxide. Environ Health Perspect
101: 159-164. http://dx.doi.org/10.2307/343 1719
Tompa. A; Major. J; Jakab. MG. (1999). Is breast cancer cluster influenced by environmental and
occupational factors among hospital nurses in Hungary? Pathol Oncol Res 5: 117-121.
http://dx.doi.org/10.1053/paor. 1999.0182
Tompkins. EM; McLuckie. K1E; Jones. DJ; Farmer. PB; Brown. K. (2009). Mutagenicity of DNA
adducts derived from ethylene oxide exposure in the pSP189 shuttle vector replicated in human
Ad293 cells. Mutat Res 678: 129-137. http://dx.doi.org/10.1016/j.mrgentox.2009.05.01 1
Tornqvist. M. (1996). Ethylene oxide as a biological reactive intermediate of endogenous origin
[Review], Adv Exp Med Biol 387: 275-283. http://dx.doi.org/10.1007/978-1 -4757-9480-9 36
Tornqvist. MA; Almberg. JG; Bergmark. EN; Nilsson. S; Osterman-Golkar. SM. (1989). Ethylene oxide
doses in ethene-exposed fruit store workers. Scand J Work Environ Health 15: 436-438.
http://dx.doi.org/10.5271/siweh. 1829
U.S. EPA (U.S. Environmental Protection Agency). (1985). Health assessment document for ethylene
oxide: Final report. (EPA-600/8-84-009F). Research Triangle Park, NC: Environmental Criteria
R-12
-------
and Assessment Office, Office of Health and Environmental Assessement, Office of Research
and Development, U.S. Environmental Protection Agency.
https://nepis.epa.gov/Exe/ZvNET.exe/20009FCT.TXT?ZvActionD=ZyDocument&Client=EPA
&Index= 1981+Thru+1985&Docs=&Ouery=&Time=&EndTime=& SearchMethod= l&TocRestri
ct=n&Toc=&TocEntrv=&OField=&OFieldYear=&OFieldMonth=&OFieldDav=&lntOFieldOp
=0&ExtQFieldOp=0&XmlQuerv=&File=D%3A%5Czyfiles%5CIndex%20Data%5C81thru85%
5CTxt%5C00000002%5C20009FCT.txt&User=ANONYMOUS&Password=anonymous&Sort
Method=h%7C-
&MaximumDocuments=l&FuzzvDegree=0&ImageQualitv=r75g8/r75g8/xl50vl50gl6/i425&D
isplav=hpfr&DefSeekPage=x&SearchBack=ZvActionL&Back=ZyActionS&BackDesc=Results
0 o 2 0 p a ge & M a x i m u m P a ge s = 1 &ZyEntry= 1 & SeekPage=x&ZyPURL
U.S. EPA (U.S. Environmental Protection Agency). (1986). Guidelines for mutagenicity risk assessment
(pp. 1-17). (EPA/630/R-98/003). Washington, DC: U.S. Environmental Protection Agency, Risk
Assessment Forum, https://www.epa.gov/risk/guidelines-mutagenicitv-risk-assessment
U.S. EPA (U.S. Environmental Protection Agency). (1994). Methods for derivation of inhalation
reference concentrations and application of inhalation dosimetry [EPA Report] (pp. 1-409).
(EPA/600/8-90/066F). Research Triangle Park, NC: U.S. Environmental Protection Agency,
Office of Research and Development, Office of Health and Environmental Assessment,
Environmental Criteria and Assessment Office.
https://cfpub.epa.gov/ncea/risk/recordisplav.cfm?deid=71993&CFID=51174829&CFTOKEN=2
5006317
U.S. EPA (U.S. Environmental Protection Agency). (1997). Chemical and radiation leukemogenesis in
humans and rodents and the value of rodent models for assessing risks of lymphohematopoietic
cancers [EPA Report], (EPA/600/R-97/090). Washington, DC: Office of Research and
Development, National Center for Environmental Assessment.
http:// cfpub. epa. gov/ncea/cfm/recordisplay. cfm?deid= 12181
U.S. EPA (U.S. Environmental Protection Agency). (2000). Science policy council handbook: Risk
characterization (pp. 1-189). (EPA/100/B-00/002). Washington, D.C.: U.S. Environmental
Protection Agency, Science Policy Council, https://vvvvvv.epa.gov/risk/risk-characterization-
handbook
U.S. EPA (U.S. Environmental Protection Agency). (2005a). Guidelines for carcinogen risk assessment
[EPAReport] (pp. 1-166). (EPA/630/P-03/001F). Washington, DC: U.S. Environmental
Protection Agency, Risk Assessment Forum, http://vvvvvv2.epa.gov/osa/guidelines-carcinogen-
risk-assessment
U.S. EPA (U.S. Environmental Protection Agency). (2005b). Supplemental guidance for assessing
susceptibility from early-life exposure to carcinogens (pp. 1-125). (EPA/630/R-03/003F).
Washington, DC: U.S. Environmental Protection Agency, Risk Assessment Forum.
https://vvvvvv3.epa.gov/airtoxics/childrens supplement final.pdf
U.S. EPA (U.S. Environmental Protection Agency). (2006a). Evaluation of the carcinogenicity of
ethylene oxide: External review draft [EPA Report], (EPA/635/R-06/003). Washington, DC:
U.S. Environmental Protection Agency, Office of Research and Development, National Center
for Environmental Assessment. http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm7deidH57664
U.S. EPA (U.S. Environmental Protection Agency). (2006b). U.S. Environmental Protection Agency
peer review handbook 3rd edition (3 ed.). (EPA/100/B-06/002). Washington, DC: U.S.
Environmental Protection Agency, Science Policy Council.
R-13
-------
https://www.epa.gov/sites/production/files/2015-
09/documents/peer review handbook 2006 3rd edition.pdf
U.S. EPA (U.S. Environmental Protection Agency). (2012). Benchmark dose technical guidance.
(EPA/100/R-12/001). Washington, DC: U.S. Environmental Protection Agency, Risk
Assessment Forum, https://www.epa.gov/risk/benchmark-dose-technical-euidance
U.S. EPA (U.S. Environmental Protection Agency). (2013a). Evaluation of the inhalation
carcinogenicity of ethylene oxide - appendices (CASRN 75-21-8): In support of summary
information on the Integrated Risk Information System (IRIS) [EPA Report], (EPA/63 5/R-
13/128b). Washington, DC: U.S. Environmental Protection Agency, Office of Research and
Development, National Center for Environmental Assessment.
http s://nepis.epa. gov/Exe/ZyPURL. c gi? Dockey=P 100ICKB. txt
U.S. EPA (U.S. Environmental Protection Agency). (2013b). Evaluation of the inhalation
carcinogenicity of ethylene oxide (CASRN 75-21-8): In support of summary information on the
Integrated Risk Information System (IRIS) [EPA Report], (EPA/635/R-13/128a). Washington,
DC: U.S. Environmental Protection Agency, Office of Research and Development, National
Center for Environmental Assessment.
http s://nepis.epa. gov/Exe/ZyPURL. c gi? Dockev=P 1001CFH. txt
U.S. EPA (U.S. Environmental Protection Agency). (2014a). Evaluation of the inhalation
carcinogenicity of ethylene oxide (Revised Aug 2014 external review draft) [EPA Report],
(EPA/635/R-14/194A). Washington, DC: U.S. Environmental Protection Agency, Office of
Research and Development, National Center for Environmental Assessment.
https://cfpub.epa.gov/ncea/iris drafts/recordisplay.cfm?deid=282012
U.S. EPA (U.S. Environmental Protection Agency). (2014b). Evaluation of the inhalation
carcinogenicity of ethylene oxide (Revised Aug 2014 external review draft): Appendices [EPA
Report], (EPA/635/R-14/194B). Washington, DC: U.S. Environmental Protection Agency,
Office of Research and Development, National Center for Environmental Assessment.
https://cfpub.epa.gov/ncea/iris drafts/recordisplav.cfm?deid=282012
Valdez-F lores. C; Sielken. RL, Jr; Teta. MJ. (2010). Quantitative cancer risk assessment based on
NIOSH and UCC epidemiological data for workers exposed to ethylene oxide. Regul Toxicol
Pharmacol 56: 312-320. http://dx.doi.org/10.1016/j.yrtph.2009.10.001
van Delft. JHM; van Winden. MJM; Luiten-Schuite. A: Ribeiro. LR; Baan. RA. (1994). Comparison of
various immunochemical assays for the detection of ethylene oxide-DNA adducts with
monoclonal antibodies against imidazole ring-opened N7-(2-hydroxyethyl) guanosine:
Application in a biological monitoring study. Carcinogenesis 15: 1867-1873.
http://dx.doi.org/10.1093/carcin/15.9.1867
van Sittert. NJ; Beulink. GD; van Vliet. EWN; van der Waal, H. (1993). Monitoring occupational
exposure to ethylene oxide by the determination of hemoglobin adducts. Environ Health Perspect
99: 217-220. http://dx.doi.org/10.2307/343 1485
van Sittert. NJ; Boogaard, PJ; Natarajan. AT; Tates. AD; Ehrenberg. LG; Tornqvist, MA. (2000).
Formation of DNA adducts and induction of mutagenic effects in rats following 4 weeks
inhalation exposure to ethylene oxide as a basis for cancer risk assessment. Mutat Res-Fundam
Mol Mech Mutagen 447: 27-48. http://dx.doi.org/10.1016/S0027-5107(99)00208-0
Van Sittert. NJ; de Jong. G; Clare. MG; Davies. R; Dean. BJ; Wren. LJ; Wright. AS. (1985).
Cytogenetic, immunological, and haematological effects in workers in an ethylene oxide
manufacturing plant. Occup Environ Med 42: 19-26. http://dx.doi.org/10.1 136/oem.42.1.19
R-14
-------
van Wijngaarden. E; Hertz-Picciotto. I. (2004). A simple approach to performing quantitative cancer
risk assessment using published results from occupational epidemiology studies. Sci Total
Environ 332: 81-87. http://dx.doi.Org/10.1016/i.scitotenv.2004.04.005
Vergnes. JS: Pritts. IM. (1994). Effects of ethylene on micronucleus formation in the bone marrow of
rats and mice following four weeks of inhalation exposure. Mutat Res 324: 87-91.
http://dx.doi.org/10.1016/0165-7992(94)9005 1-5
Walker. VE; Fennell. TR; Boucheron. J A; Fedtke. N: Ciroussel. F; Swenberg. JA. (1990).
Macromolecular adducts of ethylene oxide: A literature review and a time-course study on the
formation of 7-(2-hydroxyethyl)guanine following exposures of rats by inhalation [Review],
Mutat Res 233: 151 -164. http://dx.doi.org/10.1016/0027-5 107(90)90159-2
Walker. VE; Fennell. TR; Upton. PB; MacNeela. JP; Swenberg. JA. (1993). Molecular dosimetry of
DNA and hemoglobin adducts in mice and rats exposed to ethylene oxide. Environ Health
Perspect 99: 1 1-17. http://dx.doi.org/10.2307/343 145 1
Walker. VE; Fennell. TR; Upton. PB; Skopek. TR; Prevost. V; Shuker. PEG; Swenberg. JA. (1992a).
Molecular dosimetry of ethylene oxide: Formation and persistence of 7-(2-hydroxyethyl)guanine
in DNA following repeated exposures of rats and mice. Cancer Res 52: 4328-4334.
Walker. VE; MacNeela. JP; Swenberg. JA; Turner. MJ. Jr; Fennell. TR. (1992b). Molecular dosimetry
of ethylene oxide: Formation and persistence of N-(2-hydroxyethyl)valine in hemoglobin
following repeated exposures of rats and mice. Cancer Res 52: 4320-4327.
Walker. VE; Sisk. SC; Upton. PB; Wong, BA; Recio. L. (1997). In vivo mutagenicity of ethylene oxide
at the hprt locus in T-lymphocytes of B6C3F1 lacl transgenic mice following inhalation
exposure. Mutat Res 392: 21 1-222. http://dx.doi.org/10.1016/S1383-5718(97)00062-4
Walker. VE; Skopek. TR. (1993). A mouse model for the study of in vivo mutational spectra: Sequence
specificity of ethylene oxide at the hprt locus. Mutat Res 288: 151-162.
http://dx.doi.org/10.1016/0027-5107(93)90216-3
Walker. VE; Wu, K. -Y; Upton. PB; Ranasinghe. A; Scheller. N; Cho. M. -H; Vergnes. JS; Skopek. TR;
Swenberg. J A. (2000). Biomarkers of exposure and effect as indicators of potential carcinogenic
risk arising from in vivo metabolism of ethylene to ethylene oxide. Carcinogenesis 21: 1661-
1669. http://dx.doi.org/10.1093/carcin/21.9.1661
Warwick. GP. (1963). The mechanism of action of alkylating agents [Review], Cancer Res 23: 13 15-
1333.
WHO (World Health Organization). (2003). Concise international chemical assessment document:
Ethylene oxide. In Concise International Chemical Assessment: Ethylene Oxide. Geneva,
Switzerland: World Health Organization, International Programme on Chemical Safety (IPCS).
http://www.who.int/ipcs/publications/cicad/en/cicad54.pdf
Wong, O; Trent. LS. (1993). An epidemiological study of workers potentially exposed to ethylene
oxide. Occup Environ Med 50: 308-3 16. http://dx.doi.org/10.2307/27727609
Wu, K, -Y; Ranasinghe. A; Upton. PB; Walker. VE; Swenberg. JA. (1999). Molecular dosimetry of
endogenous and ethylene oxide-induced N7-(2-hydroxyethyl) guanine formation in tissues of
rodents. Carcinogenesis 20: 1787-1792. http://dx.doi.org/10.1093/carcin/20.9.1787
Yager. JW. (1987). Effect of concentration-time parameters on sister-chromatid exchanges induced in
rabbit lymphocytes by ethylene oxide inhalation. DNA Repair 182: 343-352.
http://dx.doi.org/10.1016/0165-1 161 (87)90076-8
Yager. JW; Benz. RD. (1982). Sister chromatid exchanges induced in rabbit lymphocytes by ethylene
oxide after inhalation exposure. Environ Mutagen 4: 121-134.
http://dx.doi.org/10.1002/em.2860040204
R-15
-------
Yager. JW; Hines. CJ; Spear. RC. (1983). Exposure to ethylene oxide at work increases sister chromatid
exchanges in human peripheral lymphocytes. Science 219: 1221-1223.
http://dx.doi.oru/10.1126/science.682885 1
Yong. LC: Schulte. PA; Kao. C. -Y; Giese. RW: Boeniger. MF; Strauss. GHS; Petersen. MR: Wiencke.
JK. (2007). DNA adducts in granulocytes of hospital workers exposed to ethylene oxide. Am J
Ind Med 50: 293-302. http://dx.doi.org/10.1002/aiim.20443
Yong. LC: Schulte. PA; Wiencke. JK: Boeniger. MF: Connallv. LB; Walker. JT; Whelan. EA; Ward.
EM. (2001). Hemoglobin adducts and sister chromatid exchanges in hospital workers exposed to
ethylene oxide: Effects of glutathione S-transferase T1 and Ml genotypes. Cancer Epidemiol
Biomarkers Prev 10: 539-550.
Zeiger, E. (2010). Historical perspective on the development of the genetic toxicity test battery in the
United States [Review], Environ Mol Mutagen 51: 781-791. http://dx.doi.org/10.1002/em.20602
Zhang. F; Battels. MJ; LeBaron. MJ; Schisler. MR; Jeong. Y, -C; Gollapudi. BB; Moore. MP. (2015).
LC-MS/MS simultaneous quantitation of 2-hydroxyethylated, oxidative, and unmodified DNA
nucleosides in DNA isolated from tissues of mice after exposure to ethylene oxide. J Chromatogr
B Analyt Technol Biomed Life Sci 976-977: 33-48.
http://dx.doi.org/10.1016/i.ichromb.2014.10.042
Zharlyganova. D; Harada. H; Harada. Y; Shinkarev. S; Zhumadilov. Z; Zhunusova. A; Tchaizhunusova.
NJ; Apsalikov. KN; Kemaikin. V; Zhumadilov. K; Kawano. N; Kimura. A; Hoshi. M. (2008).
High frequency of AML1/RUNX1 point mutations in radiation-associated myelodysplastic
syndrome around Semipalatinsk nuclear test site. J Radiat Res (Tokyo) 49: 549-555.
http://dx.doi.org/10.1269/jrr.08040
R-16
------- |