United States
Environmental Protection
Agency
Office of Research and
Development
Washington DC 20460
EPA/600/P-93/001aF
April 1996
Urban Soil Lead Abatement
Demonstration Project
Volume I:
EPA Integrated Report
-------
-------
EPA/600/P-93/001aF
April 1996
Urban Soil Lead Abatement
Demonstration Project
Volume I: EPA Integrated Report
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Printed on Recycled Paper
-------
DISCLAIMER
This document has been revised 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
-------
TABLE OF CONTENTS
LIST OF TABLES
LIST OF FIGURES '
LIST OF AUTHORS AND STUDY PARTICIPANTS ....;.. xix
LIST OF INTEGRATED REPORT REVIEWERS xxiii
LIST OF ABBREVIATIONS, ACRONYMS, AND TERMS xxv
1. EXECUTIVE SUMMARY 1_1
1.1 BACKGROUND AND OVERVIEW 1-1
1.1.1 Comparison of Study Hypotheses 1-3
1.1.2 Study Design and Conduct 1-5
1.1.3 Intervention Procedures 1-8
1.2 SUMMARY OF INDIVIDUAL STUDY REPORTS ........ 1-12
1.2.1 Boston Study . ....... 1-12
1.2.2 Baltimore Study 1_13
1.2.3 Cincinnati Study 1-14
1.2.4 Individual Study Conclusions 1-15
1.3 SUMMARY OF EPA INTEGRATED ASSESSMENT RESULTS
AND FINDINGS 1_17
1.3.1 Quality of the Data . . . 1-17
1.3.2 Effectiveness and Persistency of Intervention 1-18
1.3.3 Comparison of EPA Integrated Report Results with
Individual Study Results . . 1-19
1.3.3.1 Boston Study . . . . 1-19
1.3.3.2 Baltimore Study 1_22
1.3.3.3 Cincinnati Study 1-25
1.3.3.4 Synthesis of Findings Across the
Three Studies 1_28
1.4 INTEGRATED PROJECT CONCLUSIONS 1-31
2. BACKGROUND AND OVERVIEW OF PROJECT 2-1
2.1 PROJECT BACKGROUND 2-1
2.1.1 The Urban Lead Problem 2-1
2.1.2 Legislative Background 2-1
2.1.3 Site Selection . 2-2
2.2 INTEGRATION OF THE THREE STUDIES . . 2-5
2.2.1 Study Hypotheses 2-5
2.2.2 General Study Design . . 2-6
2.2.3 Study Groups 2-8
2.2.4 Project Activity Schedule 2-11
2.2.5 Environmental and Biological Measurements
of Exposure . . . 2-11
2.2.5.1 Blood Lead ..' 2-14
2.2.5.2 Hand Lead 2-18
111
-------
TABLE OF CONTENTS (cont'd)
2.2.5.3 House Dust 2'18
2.2.6 Intervention Strategies • • 2~^
2 3 EXTERNAL FACTORS THAT COULD INFLUENCE
PROJECT RESULTS AND INTERPRETATION . . . • 2-21
2.3.1 Cycles and Trends in Environmental Lead
Concentrations 2~21
2.3.2 Unexplained and Unexpected Sources of Lead 2-24
2.3.3 Movement of Lead in Soil and Dust 2-25
2.3.4 Factors That Limit Interpretation of the Project
Results 2'26
3 METHODS INTERCOMPARISON AND QUALITY
ASSURANCE/QUALITY CONTROL 3-1
3 1 INTERCOMPARISON OF LABORATORY METHODS FOR
SOIL AND DUST MEASUREMENTS 3-2
3.1.1 Round Robin Intercalibration Exercise I 3-3
3.1.2 Quality Assurance/Quality Control Standards
and Audits 3-11
3.1.3 Round Robin Intercalibration Exercise II 3-12
3.1.4 Biweight Distribution and Final Interlaboratory
Calibration • • 3-14
315 Disposition of Audit Data 3-15
3.2 QUALITY ASSURANCE AND QUALITY CONTROL FOR
HAND DUST 3-19
3.3 QUALITY ASSURANCE AND QUALITY CONTROL FOR
BLOOD LEAD 3-19
3.4 DATABASE QUALITY . . . 3-20
4. INDIVIDUAL STUDIES 4-l
4.1 INDIVIDUAL STUDY INTERVENTION STRATEGIES
AND SAMPLING PLANS 4-1
4.1.1 Boston Study 4-l
4.1.2 Baltimore Study 4-4
4.1.3 Cincinnati Study • 4-6
4.2 DESCRIPTION OF THE DATA • • • • 4-8
4.2.1 Measures of Central Tendency for Property Level
Soil and Dust • • • 4'13
4.2.2 Adjustments and Corrections to the Data 4-16
4.2.2.1 Subjects Dropped from Study 4-16
4.2.2.2 Unit Conversion 4-16
4.3 STUDY DESIGNS 4-17
4.3.1 Design Differences 4-17
4.3.2 Strengths and Weaknesses of Study Designs 4-19
IV
-------
TABLE OF CONTENTS (cont'd)
4.3.3 Modifications of the Hypotheses 4-22
4.4 INDIVIDUAL STUDY CONCLUSIONS 4-24
5. RESULTS OF INTEGRATED ANALYSES 5-1
5.1 CONCEPTUAL APPROACHES TO EVALUATING
RESPONSE TO ABATEMENT 5-1
5.1.1 Expected Impact of Intervention 5-2
5.1.1.1 Expected Impact of Soil Abatement on
Exterior and Interior Dust . . 5-2
5.1.1.2 Expected Impact of Soil and Dust Abatement
on Hand Lead Loading 5-3
5.1.1.3 Expected Impact of Soil and Dust Abatement
on Blood Lead Concentration 5-4
5.1.2 Evaluation of Specific Statistical Approaches ....... 5-5
5.1.2.1 Cross-sectional Analysis of Variance 5-6
5.1.2.2 Cross-sectional Analysis of Covariance ..... 5-9
5.1.2.3 Cross-sectional Structural Equation Models ... 5-11
5.1.2.4 Longitudinal Analysis of Covariance . 5-14
5.1.2.5 Repeated Measures Analysis of Variance .... 5-16
5.1.2.6 Repeated Measures Analysis of Covariance ... 5-16
5.1.2.7 Longitudinal Structural Equation Models .... 5-18
5.1.3 Specific Problems with Statistical Methods . 5-20
5.2 DIFFERENCES IN GROUP MEANS 5-23
5.2.1 Changes in Mean Soil Lead Concentrations 5-25
5.2.2 Changes in Exterior Dust Concentrations and
Loadings 5-26
5.2.3 Changes hi Interior Dust Concentrations and
Loadings 5-32
5.2.4 Changes in Hand Dust Lead Loadings . 5-32
5.2.5 Changes in Blood Lead Concentrations 5-51
5.2.5.1 Baltimore Study Blood Lead Data 5-51
5.2.5.2 Boston Study Blood Lead Data 5-51
5.2.5.3 Cincinnati Study Blood Lead Data 5-51
5.3 PRE- AND POSTABATEMENT DIFFERENCES
IN INDIVIDUALS 5-51
5.3.1 Individual Changes in Blood Lead and Soil Lead 5-51
5.4 COMPARISON BY CROSS-SECTIONAL STRUCTURAL
EQUATION MODELS 5-60
5.4.1 General Issues hi Structural Equation Modeling 5-62
5.4.2 Boston Preabatement Cross-Sectional Structural
Equation Models 5-68
5.4.3 Cincinnati Cross-Sectional Structural Equations
Model 5-72
-------
TABLE OF CONTENTS (cont'd)
5.5 COMPARISON BY REPEATED MEASURES ANALYSIS ... ,.-, 5-78
5.5.1 Boston Repeated Measures Analysis of
Variance 5-79
5.5.2 Cincinnati Repeated Measures Analysis of ;,
Variance . . . 5-89
5.5.3 Baltimore Repeated Measures Analysis of
Variance • • • 5-96
5.5.4 Boston Repeated Measures Analyses of Covariance . . . 5-101
5.5.5 Cincinnati Repeated Measures Analyses of
Covariance 5-107
5.5.6 Baltimore Repeated Measures Analyses of Covariance . . 5-110
5.6 COMPARISON USING LONGITUDINAL STRUCTURAL
EQUATION MODELS . . . 5-111
5.6.1 Boston Study Longitudinal Structural
Equation Models 5-111
5.6.2 Cincinnati Study Longitudinal Structural
Equation Models 5-129
5.6.3 Calculating Effect Sizes from Longitudinal
Structural Equation Models 5-130
5.7 SUMMARY OF RESULTS OF STATISTICAL ANALYSES . . . 5-137
5.7.1 Synthesis of Results by Repeated Measures ANOVA
and Longitudinal Structural Equations Modeling ..... 5-137
5.7.2 Summary of Results by Study 5-143
5.7.2.1 Boston Study 5-143
5.7.2.2 Cincinnati Study 5-146
5.7.2.3 Baltimore Study 5-147
5.7.3 Summary of Results 5-148
5.7.4 Limitations of the Statistical Methods . 5-150
5.7.5 Comparison Across the Three Studies > 5-151
6. INTEGRATED SUMMARY AND CONCLUSIONS 6-1
6.1 PROJECT OVERVIEW 6-1
6.2 SUMMARY OF FINDINGS 6-3
6.2.1 Comparison of EPA Integrated Report Results with
Individual Study Results 6-3
6.2.1.1 Boston Study . . 6-3
6.2.1.2 Baltimore Study 6-6
6.2.1.3 Cincinnati Study 6-9
6.2.2 Synthesis of Findings Across the Three Studies 6-13
6.2.3 Application of Findings to Conceptual Framework
of Soil and Dust Lead Exposure Pathways 6-15
6.3 INTEGRATED PROJECT CONCLUSIONS 6-19
VI
-------
TABLE OF CONTENTS (coht'd)
7. REFERENCES 7.1
APPENDIX A: GROUP MEAN PARAMETERS FOR EACH STUDY
BY SAMPLE TYPE, TREATMENT GROUP,
AND ROUND . A-l
APPENDIX B: THE P'-VALUES FOR THE TABLES OF
CHAPTER 5 B-l
vu
-------
LIST OF TABLES
Number
1-1
1-2
1-3
1-4
1-5
2-1
2-2
3-1
3-2
3-3
3-4
3-5
3-6
3-7
4-1
4-2
4-3
Description of Study Groups and Types of Intervention
Number of Project Participants by Round
Soil Abatement Statistics for the Three Studies
Comparison of Phase 1 Effect Size Estimates Between the
Boston Study and This Report
Effect Size Estimates from the Baltimore Report Comparing
Blood Lead Reduction hi BAL SP Versus Controls
Treatment Group Nomenclature with Cross-Reference to
Individual Reports
Number of Projept Participants by Treatment Group
and Round
Wet Chemistry and Instrumental Methods Used for the First
Intercalibration Study
Analytical Results of the First Intercalibration Study:
Lead Concentration hi the Total and Fine Fractions of
10 Soils from Each Study
Soil and Dust Audit Program Results
Preliminary and Final Biweight Distributions for Soil and
Dust Audit Program
Results of the Final Intercalibration Study
Consensus Values and Correction Factors from the
Final Intercalibration Program
Quality Control Results for Centers for Disease Control
and Prevention Blind Pool Blood Lead Analyses
Soil Abatement Statistics for the Three Studies
Summary of Boston Study Data
Summary of Baltimore Study Data
Page
1-6
1-7
1-11
-20
1-24
-10
-12
3-4
3-5
3-13
-14
3-16
3-17
3-20
4-2
4-9
4-10
Vlll
-------
LIST OF TABLES (cont'd)
Number
4-4 Summary of Cincinnati Study Data
4-5 Design Differences Among the Three Studies . . .
5-1 Cross-Sectional Structural Equation Models for Boston Study .
5-2 Preabatement Cross-Sectional Structural Equation Models
for Boston Study: Blood Lead Truncated (9 to 22 /xg/dL) . . .
5-3 Preabatement Cross-Sectional Structural Equation Models
for Cincinnati Study: Dust Type Models .............
5-4 Preabatement Cross-Sectional Structural Equation Models
for Cincinnati Study: Floor Dust • . . .
5-5 Repeated Measures Analysis of Variance for Boston Study:
Effect of Age on Reduction in Blood Lead Between
Rounds 1 and 3 . . .
5-6 Repeated Measures Analysis of Variance for Boston Study:
Effect of Age on Reduction in Blood Lead Between
Rounds 3 and 4 ,
5-7 Repeated Measures Analysis of Variance for Boston Study:
Effect of Race or Sex .
5-8 Repeated Measures Analysis of Variance for Boston Study:
Effect of Truncation on Reduction in Blood Lead Between
Rounds 1 and 3
5-9 Repeated Measures Analysis of Variance for Boston Study:
Effect of Truncation on Reduction hi Blood Lead Between
Rounds 3 and 4
5-10 Repeated Measures Analysis of Variance for Cincinnati Study;
Effect of Age Between Rounds 1 and 4
5-11 Repeated Measures Analysis of Variance for Cincinnati Study:
Effect of Age Between Rounds 4 and 7
5-12 Repeated Measures Analysis of Variance for Cincinnati Study:
Effect of Truncation Between Rounds 1 and 4
4-11
4-18
5-70
5-73
5-74
5-76
5-80
5-80
5-85
5-87
5-88
5-90
5-92
5-94
IX
-------
LIST OF TABLES (cont'd)
Number 2sgg
5-13 Repeated Measures Analysis of Variance for Cincinnati Study:
Effect of Truncation Between Rounds 1 and 4 5-95
5-14 Repeated Measures Analysis of Variance for Cincinnati Study:
Effect of Truncation Between Rounds 4 and 7 5-97
5-15 Repeated Measures Analysis of Variance for Cincinnati Study:
Effect of Truncation Between Rounds 4 and 7 5-98
5-16 Repeated Measures Analysis of Variance for Baltimore Study:
Effect of Age 5-99
5-17 Repeated Measures Analysis of Variance for Baltimore Study:
Effect of Truncation Between Rounds 3 and 4 5-100
5-18 Repeated Measures Analysis of Covariance for Boston Study:
Effect of Age and Log Dust Lead Concentration Reduction on
Log Blood Lead Between Rounds 1 and 3 5-102
5-19 Repeated Measures Analysis of Covariance for Boston Study:
Effect of Age and Log Dust Lead and Soil Lead Concentration
on Reduction in Log Blood Lead Between Rounds 1 and 3 5-103
5-20 Repeated Measures Analysis of Covariance for Boston Study:
Effect of Age and Log Dust Lead Concentration on Reduction
in Log Blood Lead Between Rounds 3 and 4 5-104
5-21 Repeated Measures Analysis of Covariance for Boston Study:
Effect of Age and Log Dust Lead Loading on Reduction in
Blood Lead Between Rounds 3 and 4 5-104
5-22 Repeated Measures Analysis of Covariance for Boston Study:
Effect of Age and Log Dust Lead Concentration on Reduction
in Log Blood Lead Between Rounds 1 and 3 for Afro-American
Children 5-106
5-23 Repeated Measures Analysis of Covariance for Boston Study:
Effect of Age and Log Dust Lead Load on Reduction in Log
Blood Lead Between Rounds 1 and 3 for Afro-American
Children . . . 5-106
x
-------
LIST OF TABLES (cont'd)
Number
5-24 Repeated Measures Analysis of Covariance for Boston Study:
Effect of Age and Log Dust Lead Concentration, Soil Lead
Concentration on Reduction in Log Blood Lead Between
Rounds 1 and 3 for Afro-American Children
5-25 Repeated Measures Analysis of Covariance for Boston
Study: Effect of Age and Log Dust Lead Loading on
Reduction in Log Blood Lead Between Rounds 3 and 4
for Afro-American Children
5-26 Repeated Measures Analysis of Covariance for Boston
Study: Effect of Age and Log Dust Lead Concentration
on Reduction in Log Blood Lead Between Rounds 3 and 4
for Afro-American Children
5-27 Repeated Measures Analysis of Covariance for Cincinnati
Study: Reduction in Blood Lead Between Rounds 1 and 4 .
5-28 Models for Treatment Group Effects in Boston Longitudinal
Structural Equation Models , .
5-29 Longitudinal Structural Equation Models: Model Assessment
Statistics hi Boston Study Using Estimated Blood Lead
Persistence Factor
5-30 Longitudinal Structural Equation Models for Boston Study:
Regression Coefficients Using Estimated Blood Lead
Persistence Factor
5-31 Longitudinal Structural Equation Models for Boston Study:
Model Assessment Statistics Using Fixed Blood Lead
Persistence Factor
5-32 Longitudinal Structural Equation Models for Boston Study:
Regression Coefficients Using Fixed Blood Lead
Persistence Factor
5-33 Longitudinal Structural Equation Models for Boston Study:
Model Assessment Statistics Using Fixed Blood Lead
Persistence Factor for Males
5-34 Longitudinal Structural Equation Models for Boston Study:
Regression Coefficients Using Fixed Blood Lead
Persistence Factor for Males
5-107
5-108
5-108
5-109
5-112
5-113
5-114
5-115
5-116
5-117
5-118
XI
-------
LIST OF TABLES (cont'd)
Number
5-35 Longitudinal Structural Equation Models for Boston Study:
Model Assessment Statistics Using Fixed Blood Lead
Persistence Factor for Females
5-36 Longitudinal Structural Equation Models for Boston Study:
Regression Coefficients Using Fixed Blood Lead
Persistence Factor for Females .
5-37 Longitudinal Structural Equation Models Boston Study:
Model Assessment Statistics Using Fixed Blood Lead
Persistence Factor for Ages 18 to 41 Months
5-38 Longitudinal Structural Equation Models for Boston Study:
Regression Coefficients Using Fixed Blood
Persistence Factor for Ages 18 to 41 Months
5-39 Longitudinal Structural Equation Models for Cincinnati Study:
Model Assessment Statistics Using Fixed Blood Lead
Persistence Factor
5-40 Longitudinal Structural Equation Models for Cincinnati Study:
Regression Coefficients Using Fixed Blood Lead
Persistence Factor ,
5-41 Comparison of Statistical Methods for Boston Study:
Reduction in Blood Lead Between Rounds 1 and 3
5-42 Comparison of Statistical Methods for Cincinnati Study:
Reduction in Blood Lead Between Rounds 1 and 4
5-43 Calculation of Dust Lead Adjustment to Blood Lead
Reduction of the Boston and Cincinnati Studies from
Parameters of the Structural Equation Model in
Tables 5-32 and 5-40
6-1 Comparison of Phase 1 Effect Size Estimates Between the
Boston Study and This Report
6-2 Effect Size Estimates from the Baltimore Report Comparing
Blood Lead Reduction hi BAL SP Versus Controls
Page
5-119
5-120
5-121
5-122
5-130
5-131
5-133
5-135
5-136
6-5
6-9
XII
-------
LIST OF FIGURES
Number
1-1 Generalized concept of the sources and pathways of lead
exposure in humans 1-8
1-2 Typical pathways of childhood exposure to lead in dust . 1-9
2-1 Project activity schedule showing the round designations and
time periods for sampling and interviewing, and the time
periods for soil abatement 2-13
2-2 Generalized concept of the sources and pathways of lead
exposure in humans . . 2-14
2-3 Typical pathways of childhood exposure to lead hi dust
showing both the complexity of the routes of exposure and
the mobility of dust lead along these routes 2-15
2-4 Hypothetical representation of the expected decrease in
blood lead following abatement 2-17
f .
2-5 Literature values for seasonal patterns for childhood
blood lead (age 25 to 36 mo) hi Chicago . 2-22
2-6 Estimated seasonal variation based on residual blood-lead
levels hi Boston children after controlling for age and
date of birth effects 2-23
2-7 Predicted differences hi blood lead and hand lead during
early childhood, based on empirical data 2-24
3-1 Comparison of uncorrected data for two wet chemistry methods
of soil analysis showing the comparability of hot and cold
nitric acid for the Cincinnati laboratory 3-6
3-2 Comparison of uncorrected data for atomic absorption spectroscopic
analysis by two laboratories (Baltimore and Cincinnati) using
the hot nitric acid method of soil analysis 3-7
3-3 Interlaboratory comparison of uncorrected data for the X-ray
fluorescence method of soil analysis showing the comparability
of the Boston and Georgia Institute of Technology laboratories ... 3-8
xm
-------
LIST OF FIGURES (cont'd)
Number
3-4 Interlaboratory comparison of uncorrected data for soil
analysis showing the comparability of inductively coupled
plasma emission spectroscopy and atomic absorption spectroscopy
for the Baltimore and Cincinnati laboratories . :
3-5 Comparison of uncorrected data for soil analysis showing the
comparability of inductively coupled plasma emission spectroscopy
and atomic absorption spectroscopy within the Baltimore
laboratory
3-6 Interlaboratory comparison of uncorrected data for soil analysis
showing the comparability of X-ray fluorescence and atomic
absorption spectroscopy for the Cincinnati and Boston
laboratories
3-7 Interlaboratory comparison of uncorrected data for soil analysis
showing the comparability of X-ray fluorescence and atomic
absorption spectroscopy for the Baltimore and Boston
laboratories *. . .
3-8 Departures from consensus dust values for each of the
three studies
3-9 Departures from consensus soil values for each of the
three studies
4-1 Pathway intervention scheme for dust exposure (Boston Soil
Abatement Study) ,
4-2 Pathway intervention scheme for dust exposure (Baltimore Soil
Abatement Study)
4-3 Pathway intervention scheme for dust exposure (Cincinnati Soil
Abatement Study)
5-1 Analysis of variance
5-2 Analysis of covariance
5-3 Cross-sectional structural equations model .
5-4 Longitudinal analysis of covariance
5-5 Repeated measures analysis of variance
3-8
3-9
3-9
3-10
3-18
3-18
4-2
4-5
4-7
5-7
5-10
5-12
5-15
5-18
xiv
-------
LIST OF FIGURES (cont'd)
Number
5-6 Repeated measures analysis of covariance .
5-7 Longitudinal structural equation model
5-8 Schematic representation of expected outcomes for treatment and
control groups
5-9 Hypothetical representation of common statistical parameters for a
single group and a single round .
5-10 Boston soil lead concentrations by study group show the
effectiveness and persistency of soil abatement .
5-11 Cincinnati soil lead concentrations
5-12 Baltimore soil lead concentrations
5-13 Exterior dust lead concentrations from the street samples
in the Cincinnati study
5-14 Exterior dust lead concentrations from the sidewalk samples
in the Cincinnati study
5-15 Boston floor dust lead concentration
5-16 Boston floor dust load
5-17 Boston floor dust lead load
5-18 Boston window dust lead concentrations
5-19 Boston window dust load
5-20 Boston window dust lead load
5-21 Cincinnati floor dust lead concentrations
5-22 Cincinnati floor dust load
5-23 Cincinnati floor dust lead load
5-24 Cincinnati window dust lead concentration
Page
5-19
5-20
5-21
5-25
5-27
5-28
5-29
5-30
5-31
5-33
5-34
5-35
5-36
5-37
5-38
5-39
5-40
5-41
5-42
xv
-------
LIST OF FIGURES (cont'd)
Number
5-25 Cincinnati window dust load 5-43
5-26 Cincinnati window dust lead load 5-44
5-27 Cincinnati entry dust lead concentration 5-45
5-28 Cincinnati entry dust load 5-46
5-29 Cincinnati entry dust lead load • • • 5-47
5-30 The Boston hand lead load • • • 5"48
5-31 Baltimore hand lead load . . 5-49
5-32 Cincinnati hand lead load 5-50
5-33 Baltimore blood lead concentrations 5-52
5-34 Boston blood lead concentrations 5-53
5-35 Cincinnati blood lead concentrations 5-54
5-36 Line plots of blood lead for individual children hi Phase 1
(Rounds 1 through 3) of the Boston study for each treatment
group 5-56
5-37 Line plots of blood lead for individual children in Phase 2
(Rounds 3 and 4) of the Boston study for each treatment
C CO
group -3"-30
5-38 Line plots of blood lead for individual children hi Phase 1
(Rounds 1 through 4) of Cincinnati for each treatment group
and four age groups 5-59
5-39 Line plots of blood lead for individual children hi Phase 2
(Rounds 4 through 7) of Cincinnati for each treatment group
and four age groups 5-60
5-40 Line plots of blood lead for individual children hi Baltimore
for each treatment group and five age groups 5-61
5-41 Typical pathways of childhood exposure to lead in dust showing
both the complexity of the routes of exposure and the mobility
of dust lead along these routes 5-62
xvi
-------
LIST OF FIGURES (cont'd)
Number
5-42 Explanation of the terms and features of the structural
equation model . 5-63
5-43 Dust lead exposure pathway diagram, similar to Figure 5-41,
showing the assumed relationships for interior floor and
window dust that were modeled by cross-sectional SEM 5-64
5-44 Adaptation of the soil and dust pathway diagram (Figure 5-43)
that illustrates the general scheme for the cross-sectional
structural equation models, using the notation of Figure 5-42 .... 5-65
5-45 Pathway diagram of 12 different cross-sectional structural
equation models for Round 1 of the Boston study 5-69
5-46 Pathway diagram for Boston cross-sectional SEM Model 3,
with results as indicated from Table 5-1 5-71
5-47 Pathway diagram of 12 different cross-sectional structural
equation models for Round 1 of the Cincinnati study 5-77
5-48 Pathway diagram showing the results of the longitudinal
structural equation Model 17 for the Boston study,
from Table 5-32, using the standard terminology of
Figure 5-44 5-138
5-49 Pathway diagram showing the output from the longitudinal
structural equation Model 5 for the Cincinnati study,
from Table 5-40 5-140
5-50 Pathway diagram showing the output from longitudinal
structural equation Model J6 for the Cincinnati study 5-142
6-1 Total amounts of lead in various compartments of a child's
environment, using the assumptions for concentration
or lead loading 6-18
xvu
-------
-------
LIST OF AUTHORS AND STUDY PARTICIPANTS
U. S. EPA Integrated Report Authors
Dr. Robert Elias
National Center for Environmental
Assessment, RTF Office
U.S. EPA (MD-52)
Research Triangle Park, NC 27711
Dr. Lester D. Grant
Director, National Center for
Environmental Assessment, RTP Office
U.S. EPA (MD-52)
Research Triangle Park, NC 27711
Dr. Allan Marcus
National Center for Environmental
Assessment, RTP Office
U.S. EPA (MD-52)
Research Triangle Park, NC 27711
Individual City Study Participants
(Investigators, Assistants, Advisors, etc.)
Dr. Ann Aschengrau
Boston University School of Public Health
80 East Concord Street, T-355
Boston, MA 02118
Dr. Alexa Beiser
Boston University School of Public Health
80 East Concord Street, T-355
Boston, MA 02118
Dr. David Bellinger
Children's Hospital
Gardner House Room 455
300 Longwood Avenue
Boston, MA 02115
Dr. Robert Bornschein
University of Cincinnati
Department of Environmental Health
3223 Eden Avenue #56
Cincinnati, OH 45267-0056
Ms. Dawn Boyer
Inorganic Chemistry Department
Lockheed ESC
1050 East Flamingo
Las Vegas, NV 89119
Ms. Merrill Brophy
MDE/Lead & Soil Project
Maryland Department of the Environment
2500 Broening Highway
Baltimore, MD 21224
Dr. Richard Brunker
U.S. EPA - Region III
Site Support Section MD-3HW26
841 Chestnut Street
Philadelphia, PA 19107
Mr. Barry Chambers
Maryland Department of the Environment
2500 Broening Highway
Baltimore, MD 21224
xix
-------
LIST OF AUTHORS AND STUDY PARTICIPANTS (cont'd)
Dr. Rufus Chaney
U.S. Department of Agriculture
ARC, Building 318 BARC-East
Beltsville, MD 20705
Dr. Julian Chisolm
Kennedy Institute
707 N. Broadway
Baltimore, MD 21205
Dr. Scott Clark
University of Cincinnati
Department of Environmental Health
Mail Stop 56
3223 Eden Avenue
Cincinnati, OH 45267-0056
Ms. Linda Conway-Mundew
University of Cincinnati
Department of Environmental Health
1142 Main Street
Cincinnati, OH 45210
Dr. Donna Copenhafer
Boston University School of Public Health
80 East Concord Street, T-355
Boston, MA 02118
Dr. Robert Elias
U.S. EPA National Center
for Environmental Assessment
Mail Drop 52
Research Triangle Park, NC 27711
Dr. Katherine Farrell
Anne Arundell County Health Department
3 Harry S. Truman Parkway
Annapolis, MD 21401
Ms. Beverly Fletcher
U.S. EPA - Region I
Environmental Services Division
60 Westview Street
Lexington, MA 02173
Mrs. Barbara Gordon
Cincinnati Health Department
3101 Burnet Avenue, Room 309
Cincinnati, OH 45229
Ms. Jo Ann Grote
University of Cincinnati
Department of Environmental Health
1142 Main Street
Cincinnati, OH 45210
Mr. Bill Hanson
Cincinnati Health Department
UC Soil Project
1142 Main Street
Cincinnati, OH 45210
Mr. Reginald Harris
U.S. EPA - Region III
Site Support Section MD-3HW15
841 Chestnut Street
Philadelphia, PA 19107
Mr. Ronald Jones
Cleveland Department of Public Health
1925 East St. Claire Ave.
Cleveland, OH 44114
Dr. Boon Lim
Environmental Health Program
Maryland Department of the Environment
2500 Broening Highway
Baltimore, MD 21224
xx
-------
LIST OF AUTHORS AND STUDY PARTICIPANTS (cont'd)
Dr. Allan Marcus
U.S. EPA
National Center for Environmental
Assessment
Mail Drop 52
Research Triangle Park, NC 27711
Dr. Tom Matte
Public Health Service - Region II
26 Federal Plaza, Room 3337
New York, NY 10278
Ms. Lisa Matthews
U.S. EPA - OS 230
401 M Street, SW
Washington, DC 20460
Mr. Dave Mclntyre
U.S. EPA - Region I
Environmental Services; Division
60 Westview Street
Lexington, MA 02173
Mr. William Menrath
University of Cincinnati
Department of Environmental Health
1142 Main Street
Cincinnati, OH 45210
Dr. Winkey Pan
University of Cincinnati
Department of Environmental Health
3223 Eden Avenue
Cincinnati, OH 45267-0056
Dr. Dan Paschal
Centers for Disease Control
1600 Clifton Road, NE
Mail Stop F-18
Atlanta, GA 30333
Ms. Sandy Roda
University of Cincinnati
Department of Environmental Health
3223 Eden Avenue
Cincinnati, OH 45267-0056
Dr. Charles Rohde
Biostatistics
Johns Hopkins University
615 N. Wolfe Street
Baltimore, MD 21205
Ms. Penny Schmitgen
University of Cincinnati
Department of Environmental Health
3223 Eden Avenue
Cincinnati, OH 45210
Dr. James Simpson
Centers for Disease Control
CEHIC/EHHE, Mail Stop F28
1600 Clifton Road
Atlanta, GA 30333
Dr. Tom Spittler
U.S. EPA - Region I
Environmental Services Division
60 Westview Street
Lexington, MA 02173
Mr. Warren Strauss
MDE/Lead & Soil Project
Maryland Department of the Environment
2500 Broening Highway
Baltimore, MD 21224
Dr. Paul A. Succop
University of Cincinnati
Department of Environmental Health
3223 Eden Avenue
Cincinnati, OH 45267-0056
xxi
-------
LIST OF AUTHORS AND STUDY PARTICIPANTS (coni'd)
Dr. Pat Van Leeuwen
U.S. EPA - Region V
Technical Support Unit 5HR-11
230 S. Dearborn Street
Chicago, IL 60404
Dr. Harold A. Vincent
U.S. EPA
Quality Assurance Division
Environmental Monitoring Systems
Laboratory - Las Vegas
P.O. Box 93478
Las Vegas, NV 89193-3478
Dr. Michael Weitzman
Chief of Pediatrics
Rochester General Hospital
1425 Portland Avenue
Rochester, NY 14621
xxn
-------
LIST OF INTEGRATED REPORT REVIEWERS
External Expert Peer Reviewers1
Dr. Carol Angle
Department of Pediatrics
University of Nebraska Medical Center
600 S. 42nd Street
Omaha, NB 68198-6055
Mr. Dennis Cox
Department of Statistics
Rice University
P.O. Box 1892
Houston, TX 77251
Dr. Robert Goyer
Consultant
6405 Huntingtonridge Road
Chapel Hill, NC 27514
Dr. Vic Hasselblad
Center for Health Policy Res.
and Evaluation
Duke University
125 Old Chemistry Building
Durham, NC 27706
Dr. Paul Mushak
Principle, PB Associates
Couch Building, Suite G-3
714 Ninth Street
Durham, NC 27705
Dr. Janet Phoenix
National Safety Council
1019 19th Street, NW, Suite 401
Washington, DC 20036-5105
Dr. Michael Rabinowitz
Marine Biological Laboratory
MBL Street
Woods Hole, MA 02543
Dr. Jerry Sacks
National Institute of Statistical Science
P.O. Box 14162
Research Triangle Park, NC 27709
Dr. Alice Stark
Bureau of Env. Epi. & Occ. Health
New York State Department of Health
2 University Place, Room 130
Albany, NY 12237
Dr. Thomas B. Starr
7500 Rainwater Road
Raleigh, NC 27615-3700
Dr. Ian von Lindern
President, Terragraphics Environmental
121 South Jackstreet
Moscow, ID 83483
lfThe external (non-EPA) experts listed reviewed an external review draft of the present U.S. EPA Integrated
Report at an EPA-sponsored peer-review workshop held in September 1995. Most of these expert peer
reviewers and the following additional experts also reviewed an earlier Integrated Report draft in
August 1993: Dr. Alvaro Garza (San Francisco Dept. of Public Health), Dr. Michael Shannon (Children's
Hospital, Boston), and Dr. Micheal Symons (University of North Carolina, Chapel Hill). The internal EPA
reviewers for the September 1995 draft were Dr. Harlal Choudbury (NCEA/CIN), Dr. Lawrence Cox
(NERL/RTP), Dr. Sharon Harper (NERL/RTP), Dr. Karen Hogan (OPPTS/Washington, DC), Dr. Kathryn
Mahaffey (NCEA/CIN), Dr. Paul White (NCEA/Washington, DC), and Dr. Larry Zarogoza (OSWER/
Washington, DC). We gratefully acknowledge the advice and recommendations of these expert reviewers.
The final version of this Integrated Report is the responsibility of U.S. EPA and does not necessarily represent
the individual views of any single listed reviewer.
XX111
-------
-------
LIST OF ABBREVIATIONS, ACRONYMS, AND TERMS
AAS
ANCOVA
BALP1/BALP2
BALSP
BOS P-S
BOSPI-S
BOS SPI
CDC
CIN I-SE
CINNT
CIN SEI
dL
Double blind
Dust loading
ECAO/RTP
EPA
GLIM
Atomic absorption spectroscopy
Analysis of covariance
Baltimore Study Group with paint intervention
Baltimore Study Group with soil and paint intervention
Boston Study Group with paint intervention, followed by
soil abatement in second year
Boston Study Group with paint and interior dust
intervention, followed by soil abatement in second year
Boston Study Group with soil, paint, and interior dust
intervention
Centers for Disease Control and Prevention
Cincinnati Study Group with ulterior dust intervention,
followed by soil and exterior dust intervention (second
year)
Cincinnati Study Group with no treatment
Cincinnati Study Group with soil, exterior dust, and
interior dust intervention
Deciliter; used here as a measure of blood lead in
micrograms per deciliter
Analytical audit sample where analyst knows neither that
the sample is an audit sample nor the concentration
Mass of dust per unit area
Environmental Criteria and Assessment Office/Research
Triangle Park (now National Center for Environmental
Assessment/Research Triangle Park)
U.S. Environmental Protection Agency
Numerical Algorithms Group software package for a
general linear model
xxv
-------
LIST OF ABBREVIATIONS, ACRONYMS, AND TERMS (cbnt'd)
GLM
Hand dust
HEPA
ICP
Lead concentration
Lead loading
MGLH
NHANESH
ORD
OSWER
P-value
Pb
Project
P-XRF
QA/QC
Repeated measures analysis
Round
SARA
SAS procedure for general linear models approximately
equivalent to Systat MGLH
Sample taken by wiping the child's hand thoroughly;
a measure estimating the ingestion of lead
High-efficiency particle accumulator
Inductively coupled plasma emission spectroscopy
Mass of lead per mass of medium (soil, dust, water)
Mass of lead per unit area
Systat procedure for general linear models approximately
equivalent to SAS GLM
National Health Assessment and Nutrition Examination
Survey BE
Office of Research and Development
Office of Solid Waste and Emergency Response
Statistical term for the likelihood that an observed effect
differs from zero
Lead
In this report, "project" refers collectively to the three
individual studies that compose the Urban Soil Abatement
Demonstration Project.
Field or Portable XRF used hi this study for paint
measurements
Quality assurance/quality control
Statistical procedure for analyzing normally distributed
responses collected longitudinally
Period of sampling and data collection during study
Superfund Amendments and Reauthorization Act
xxvi
-------
LIST OF ABBREVIATIONS, ACRONYMS, AND TERMS (cont'd)
SAS
SES
Single blind
Study
SYSTAT
USLADP
XRF
Statistical software package
Socioeconomic status
Analytical audit sample where analyst knows sample is an
audit sample but doesn't know concentration (see Double
blind)
In this report, "study" refers to one of the three
individual soil abatement studies that compose the Urban
Soil Abatement Demonstration Project.
Statistical software package
Urban Soil Lead Abatement Demonstration Project
Laboratory scale X-ray fluorescence instrument used in
this study for soil and dust analysis (see P-XRF)
xxvu
-------
-------
1. EXECUTIVE SUMMARY
1.1 BACKGROUND AND OVERVIEW
During the past 25 years, concern for lead toxicity in children has steadily increased
with mounting evidence for the subtle but serious metabolic and developmental effects of lead
exposure levels previously thought to be safe. Childhood lead poisoning was formerly
considered a severe medical problem usually traced to swallowed chips of peeling lead-based
paint. Scientific evidence has systematically revealed deleterious effects of environmental
lead at lower levels of exposure. Federal agencies such as the U.S. Environmental
Protection Agency (EPA) and the Centers for Disease Control and Prevention (CDC) have
repeatedly lowered the level of concern for children's lead burden that recommends
environmental or clinical intervention—from a blood lead level of 30 /zg/dL established in
1978 by CDC to 25 jtg/dL hi 1985 (just prior to the start of this project), then to the present
level of 10 /ig/dL (defined hi October 1991 by CDC as a blood lead level that should trigger
community-wide prevention activities if found in many children).
The Urban Soil Lead Abatement Demonstration Project (USLADP), known also as the
"Three City Lead Study", was authorized in 1986 under Section lll(b)(6) of the Superfund
Amendments and Reauthorization Act (SARA), which mandated that EPA conduct soil lead
abatement projects hi up to three U.S. cities (SMSA's). The purpose of the project was to
determine whether abatement of lead in soil could reduce the lead hi the blood of inner city
children. It did not attempt to compare the relative effectiveness of alternative soil abatement
methods.
The project began hi December 1986 with the appointment of a U.S. EPA steering
committee to develop recommendations for implementing the SARA lead-in-soil abatement
demonstration project. A panel of experts was formed hi early 1987 to assist U.S. EPA hi
defining a set of criteria for selection of sites and the minimum requirements for a study at
each site. The panel also met in mid 1987 to discuss technical issues and study designs and
to evaluate technical criteria for selection of urban areas as potential soil-lead abatement
demonstration project sites, ultimately leading by the end of 1987 to the selection of Boston,
Baltimore, and Cincinnati as the participating cities.
1-1
-------
The individual studies in each city were designed around the concept of participating
families within a definable neighborhood. These families and their living units were part of
a study group, either a treatment group or a control group. Each study group was sampled
during preabatement and postabatement phases of the studies carried out in each city. Prior
to and after abatement, blood lead levels were ascertained and the environment of the child
was extensively evaluated through measurements of lead hi soil, dust, drinking water, and
paint, and through questionnaires about activity patterns, eating habits, family activities, and
socioeconomic status (SES). Because of the complex nature of this exposure assessment,
intermediate exposure indices, such as street dust, house dust, and hand dust were measured
in some study groups. The objective of the preabatement phase was to determine the
baseline exposure history and status (stability of the blood lead and environmental measures)
prior to abatement. During the postabatement phase, samples were taken to confirm the
effectiveness of abatement actions hi reducing lead in the abated media, to measure the
duration of the effect of soil abatement, and to detect possible recontamination. Blood lead
measurements were also obtained postabatement to ascertain abatement impacts at various
postabatement intervals.
Research teams in each city included state and/or local health department personnel,
academic researchers from local universities, and/or various other institutions (including in
Boston participation by U.S. EPA Region I Laboratory personnel). Protocols for the
environmental and blood lead measurements were developed by a Scientific Coordinating
Committee composed of representatives from each city's research team, three pertinent EPA
Regional Offices (I, HI, V), EPA/Office of Solid Waste and Emergency Response,
EPA/Office of Research and Development (ORD), and the CDC. Lead responsibility for
coordinating technical oversight for the project fell to EPA/ORD. This was accomplished
mainly via a series of workshops (2 to 3 per year) organized by ORD's Environmental
Criteria and Assessment Office in Research Triangle Park, NC (ECAO/RTP)1, at which
efforts were made to standardize measurement methods across the three individual city
studies, compare approaches to statistical analyses used by each research team, and,
^CAO/RTP, formerly within ORD's Office of Health and Environmental Assessment (OHEA), is now a unit
within ORD's National Center for Environmental Assessment (NCEA).
1-2
-------
ultimately, to obtain external peer review of the results of the studies contained in the
individual city reports2.
This report, then, is an integrated assessment of data from the above-noted coordinated
longitudinal studies of children in urban neighborhoods of three cities (Boston, Baltimore,
Cincinnati), where intervention into soil lead exposure pathways was expected to reduce the
children's blood lead. Many cross-sectional studies of childhood lead exposure have
previously shown that differences in soil lead exposure are associated with differences in
blood lead concentrations, but they did not evaluate the effectiveness of intervention steps in
terms of demonstrating that reductions hi external exposure to lead from soil result in
reductions in blood lead concentrations. Thus, a unique aspect of this project is that it
measures response to intervention, not to contamination. Because of the physiology of lead
mobilization in body tissues, there is a difference between the rate of change in a population
with increasing lead exposure and hi one with decreasing exposure. In other words, the
decrease in blood lead concentrations hi response to intervention was not expected to be at
the same rate as an increase in blood lead concentrations in response to increasing exposure.
The relationship between soil lead and blood lead is an indirect relationship hi the sense
that children most commonly do not eat soil directly, but rather they mainly ingest small
amounts of dust derived, in part, from soil. In the child's environment, dust is only one of
several sources of lead, that also include food, air, and drinking water. Likewise, the lead in
blood reflects not only recent exposure from these sources but also the lead from
accumulated body stores hi bone and other tissues, which is released to blood by biokinetic
processes that distribute and redistribute lead between blood and other body tissues.
1.1.1 Comparison of Study Hypotheses
The Scientific Coordinating Committee attempted to establish uniformity among the
three studies for major aspects of the project. This required a study plan from each city that
was discussed and reviewed at several early planning workshops. Although there were
differences hi form and content, each study plan contained
• a statement of the objectives of the study;
2Each of the three individual city reports prepared by the research teams from Boston, Baltimore, and Cincinnati,
are appended to this U.S. EPA Integrated Report.
1-3
-------
• a testable hypothesis that provided direction and focus to the study;
• protocols for collecting and analyzing the data;
• an array of treatment groups that addressed all features of the hypothesis;
• measures to be taken to ensure that all phases of the study would be conducted as
planned; and
• procedures by which the results of the study would be processed, analyzed, and
interpreted.
The objectives, protocols for sampling and analysis, quality assurance/quality control
(QA/QC) plans, and data processing procedures were nearly identical for all three studies.
Elements that differed among the three studies were the hypotheses and the varying array of
treatment groups. The hypotheses differed only slightly, as seen from the following
statements.
The overall central hypothesis of the USLADP is:
Reduction of lead in residential soil accessible to children will result
in a decrease in their blood lead concentration.
The formal statement of the Boston hypothesis is:
A significant reduction (equal to or greater than 1,000 ng/g) of lead
in soil accessible to children -will result in a mean decrease of at
least 3 pg/dL in the blood lead levels of children living in areas with
multiple possible sources of lead exposure and a high incidence of
lead poisoning.
The initial Baltimore hypothesis, stated in the null form, was:
A significant reduction of lead (> 1,000 pg/g) in residential soil
accessible to children -will not result in a significant decrease
(3 to 6 pg/dL) in their blood lead levels.
The Baltimore hypothesis, based on actual residential soil lead values averaging less
than 1,000 /*g/g, was later revised by U.S. EPA for statistical analyses purposes:
A one-time reduction of at least 500 ppm in the maximum lead
concentration in yard soil, even -when not accompanied by abatement
of household dust or lead paint inside the child's apartment.or
residence unit, will not result in a reduction of blood lead in children
living in housing in which exterior lead paint has been stabilized.
1-4
-------
The Cincinnati hypothesis was separated into two parts:
(1) A reduction of lead in residential soil accessible to children will result
in a decrease in their blood lead levels.
(2) Interior dust abatement, -when carried out in conjunction with exterior
dust and soil abatement, would result in a greater reduction in blood
lead than would be obtained with interior dust abatement alone, or
exterior dust and soil abatement alone.
Secondary hypotheses in the Cincinnati study are:
(3) A reduction of lead in residential soil accessible to children will result
in a decrease in their hand lead levels.
(4) Interior dust abatement, when carried out in conjunction with exterior
dust and soil abatement, would result in a greater reduction in hand
lead than would be obtained with interior dust abatement alone, or
exterior dust and soil abatement alone.
The array of treatment groups differed considerably among the three studies
(Table 1-1). In each study, the treatment groups had several features in common. The
groups were taken from demographically similar neighborhoods. All groups had some prior
evidence of elevated lead exposure, usually a greater than average number of public health
reports of lead poisoning. Three phases were employed in each study: a preabatement
baseline phase for 3 to 18 mo; an abatement or intervention (except for controls) phase; and
a postabatement foliow-up for 10 to 23 mo.
1.1.2 Study Design and Conduct
Table 1-1 describes the study groups and the forms of intervention employed in each of
the three cities. The Cincinnati study design used intervention on the neighborhood scale,
where the soil in parks, play areas and other common grounds were abated, and paved
surfaces in the neighborhood were cleaned of exterior dust. In Boston and Baltimore, only
soil on individual properties was abated. Table 1-2 shows the number of subjects
participating in different phases of the three studies hi relation to the respective participant
groups for each city. The general characteristics are that soil lead concentrations are
typically high in Boston, where it is also common to find lead hi both exterior and interior
1-5
-------
TABLE 1-1. DESCRIPTION OF STUDY GROUPS AND TYPES OF
INTERVENTION
Treatment Group
Namea
Cross-Reference to
Individual Study Report
Description of Treatment
BOS SPI
BOS PI-S
BOS P-S
BALSP
BAL Plb
BALP2b
CIN SEI (P)
CINI-SE
(B,D,F)C
CIN NT (G,M)C
Study Group
Control Group A
Control Group B
Study Area
Control Area
Study Area Not Abated
Area A
AreaB
AreaC
BOSTON
Soil and interior dust abatement, and interior paint
stabilization at beginning of first year, no further
treatment.
Interior dust abatement and interior paint stabilization at
beginning of first year. Soil abatement at beginning of
second year.
Interior paint stabilization at beginning of first year. Soil
abatement at beginning of second year.
BALTIMORE
Soil abatement and exterior paint stabilization at beginning
of first year, no further treatment.
Exterior paint stabilization at beginning of first year, no
further treatment.
Exterior paint stabilization at beginning of first year, no
further treatment because soil not above cut-off level.
CINCINNATI
Soil, exterior dust, and interior dust abatement at
beginning of first year, no further treatment. Includes
only the Pendleton neighborhood.
Interior dust abatement at beginning of first year, soil and
exterior dust abatement at beginning of second year, no
further treatment. Includes the Back St., Dandridge and
Findlay neighborhoods.
No treatment, soil and interior dust abatement following
last sampling round. Includes the Glencoe and Mohawk
neighborhoods. ^^
aThe treatment group designation indicates the location of the study (BOS = Boston, BAL = Baltimore,
CIN = Cincinnati), the type of treatment (S = soil abatement, E = exterior dust abatement, I = interior dust
abatement, P = loose paint stabilization, NT = no treatment).
'Treated as one group in the Baltimore report, analyzed separately in this report.
^Treated as one group for many of the analyses hi the Cincinnati report, analyzed as individual neighborhoods
in this report.
1-6
-------
TABLE 1-2. NUMBER OF PROJECT PARTICIPANTS BY ROUND*
Study
BOSTON
Middate
Children5
Famliesc
Properties'1
BALTIMORE
Middate
Children6
Families0
Properties'1
Round 1
- •• -
10/17/89
150
125
100
-^
Round 1
10/25/88
168
119
106
Round 2
4/9/90
146
121
96
Round 2
4/1/89
165
116
104
Rounds
9/12/90
147
122
97
Round 3
2/17/90
198
131
115
Round 4
7/20/91
92
77
67
Round 4 Round 5
1/27/91 6/7/91
190 186
126 122
112 105
Round 6
9/3/91
182
122
110
CINCINNATI
Middate
Children15
Families0
Properties'1
Round 1
7/6/89
201
.
129
215
Round 3
11/14/89
185
123
245
Round 4
7/1/90
219
122
245
Round 6
11/17/90
198
168
243
Round 7
6/16/91
169
142
245
aNumber shown is based on samples taken and does not include individuals enrolled but not sampled.
Intervention is shown by the vertical dashed lines.
bBased on number of children sampled for blood. Some children may not have been included in the statistical
analyses.
cBased on number of households sampled for dust.
dBased on number of properties (Boston, Baltimore) or soil parcels (Cincinnati) sampled.
paint, as well as in drinking water. In the Boston areas studied, housing is typically single
and multi-family units with relatively large lot sizes. In the Baltimore neighborhoods, the
houses were mixed single and multifamily, and the lots were smaller than Boston lots, with
typical yards less than 100 m2. Nearly every house had lead-based paint. Residential units
in Cincinnati were mostly multifamily with little or no soil on the residential parcel of land.
1-7
-------
1.1.3 Intervention Procedures
Figure 1-1 illustrates the generalized concept of human exposure to lead, showing the
pathways of lead from the several sources in the human environment to four compartments
immediately proximal to the individual. In the past decade, dramatic reductions in exposure
to lead in air and food have occurred as a result of regulatory and voluntary programs to
reduce lead in gasoline and canned food. Figure 1-2 expands the critical dust pathway to
show the complexity of the many routes of dust exposure for the typical child. The
strategies for intervention used in this project were designed to interrupt the movement of
lead along one or more of these dust pathways.
Figure 1-1. Generalized concept of the sources and pathways of lead exposure in
humans.
There were three forms of intervention hi this project: (1) soil abatement, (2) dust
removal, and (3) paint stabilization. Soil abatement was by excavation and removal,
followed by replacement with clean soil (<50 /*g/g). Dust intervention was by vacuuming,
wet mopping, and, in some cases, replacement of rugs and upholstered furniture. Cincinnati
1-8
-------
>,
Atmospheric
Particles I y
-^ y /
Soil
Exterior Paint
Dust
Local "\ \^
Fugitive >
Dust )
Exterior
Dust
Interior
Dust
/"Secondary^
^ Dust ^
Figure 1-2. Typical pathways of childhood exposure to lead hi dust.
and Boston performed interior dust abatement, and Cincinnati also removed neighborhood.
exterior dust with mechanical sweepers and hand tools. Dust intervention was not expected
to be permanent, because dust continually moves through the human environment. Instead,
the removal of dust with elevated lead concentrations was expected to expedite the impact of
soil abatement on the child's environment.
In the home, house dust is a mixture of street dust and soil dust, interior and exterior
paint dust, workplace dust carried home by adults, and dust generated from human activities
within the household. It is believed that most of the mass of the interior dust originates from
soil immediately exterior to the home, but this can vary greatly by the types of family
activities and by neighborhood characteristics. Nevertheless, in the absence of lead-based
paint inside the home, it would seem reasonable to assume that most of the lead in household
dust comes from soil and other sources immediately outside the home.
Many of the Boston and Baltimore households selected for the project had chipping and
peeling lead-based paint, both interior and exterior. In order to reduce the impact of this
paint, the walls and other surfaces were scraped and smoothed, then repainted. It is
important to note that this approach is not a full scale paint abatement and was not designed
1-9
-------
to permanently protect the child from lead-based paint. Paint stabilization was used on
interior surfaces in Boston, and on exterior surfaces in Baltimore. Paint stabilization was not
used in Cincinnati, because the lead-based paint was believed to have been removed from
these homes in the early 1970s as part of a housing rehabilitation project.
In order to accurately measure the effectiveness and persistency achieved by soil
abatement and the impact of this abatement on reducing lead exposure for children, the
sampling and analysis plans for soil and dust required robust quality control and quality
assurance objectives. Protocols were developed to define sampling schemes that characterize
the expected exposure to soil for children; collect, transfer, and store samples without
contamination; and analyze soil, dust, handwipe, and blood samples in a manner that would
maximize interlaboratory comparison. The original design focussed on sampling blood lead
during the late summer, as it was known that the seasonal blood lead cycle peaks during this
time. Where this schedule could not be adhered to, an effort was made to schedule the
follow-up blood lead sampling at a comparable time hi the cycle.
Information on area treated and volume of soil removed from each of the three cities
properties appears in Table 1-3. A total of 35 Boston properties were abated during the
study. In Baltimore, 63 properties in the BAL SP treatment group (see Table 1-3) were
abated between August and November 1990. An additional seven properties that did not
meet the requirements for abatement were transferred to a control group. Unpaved surfaces
were divided into areas on each property (usually front, back, and one side) and any area
with the maximum soil lead concentration above 500 /xg/g was abated entirely.
Within each of six neighborhoods, the Cincinnati study identified all sites with soil
cover as discrete study sites. The decision to abate was based on soil lead concentrations for
each parcel of land, and for the depth to which the lead had penetrated. Lead was measured
at two depths, the top 2 cm and from 13 to 15 cm. If the average concentration of the top
and bottom samples was greater than or equal to 500 /zg/g, the soil was removed and
replaced. If the average of the top samples exceeded 500 pig/g, but the average of the
bottom samples was less than 500 /tg/g, the soil was also abated. Ground cover was
reestablished on abated soils and some unabated soils according to protocols described in the
Cincinnati Report (appended).
1-10
-------
TABLE 1-3. SOIL ABATEMENT STATISTICS FOR THE THREE STUDIES
Number of properties3
Surface area (m2)
Volume soil removed (m3)
Surface area/property (m2)
Volume soil/property (m3)
Boston
35
7,198
1,212
200
34
Baltimore
63
4,100b
690
73
llb
Cincinnati
171
12,089
1,813
71
11
Includes only properties abated during the study. Properties abated at the end of the study, where no further
sampling was reported, are not included in this analysis, but are included in the individual study reports.
In Cincinnati, a property is the location of the soil abatement, not the location of the child's residence.
bSurface area not provided by Baltimore report. This was calculated using Boston volume-to-surface ratio,
which is equivalent to an average removal depth of 17 cm,
Exterior dust abatement was performed in the Cincinnati study only. The approach to
this abatement was to clean all types of hard surfaces where dust might collect, using vacuum
equipment that they tested and found to remove about 95% of the available dust on the area.
The dust surface categories were streets, alleys, sidewalks, parking lots, steps, and porches.
Dust measurements were made in a manner that determined the lead concentration
(micrograms of lead per gram of dust), the dust loading (milligrams of dust per square
meter), and the lead loading (micrograms of lead per square meter) for the surface measured.
This required that a dry vacuum sample be taken over a prescribed area, usually 0.25 to
0.50 m2. It is important to note that dust abatement is not expected to cause an immediate
change in the lead concentration on dust surfaces, only in the dust and lead loading.
Household dust was abated in the Boston and Cincinnati studies, but not in Baltimore.
The BOS SPI and CIN SEI groups (see Table 1-1) received interior dust abatement at the
same time as soil abatement, the BOS PI-S group received interior dust abatement without
soil abatement, and the three CIN I-SE neighborhoods received ulterior dust abatement hi the
first year, followed by soil and exterior dust abatement in the second year.
In Boston, interior dust abatement was performed after loose paint stabilization. Hard
surfaces (floors, woodwork, window wells, and some furniture) were vacuumed, as were soft
surfaces such as rugs and upholstered furniture. Hard surfaces were also wiped following
vacuuming. Common entries and stairways outside the apartment were not abated.
1-11
-------
The Cincinnati group performed interior dust abatement after exterior dust abatement.
Vacuuming was followed by wet wiping with a detergent. They vacuumed hard surfaces and
replaced one to three carpets and two items of upholstered furniture per housing unit. Their
previous studies had shown that these soft items could not be cleaned effectively with
vacuuming alone.
Most homes in the Cincinnati group had undergone extensive rehabilitation, which was
believed to have removed the lead-based paint 20 years prior to the project, but in Boston
and Baltimore lead-based paint occurred hi nearly every home. Because full paint abatement
was not within the scope of this project, the alternative was to retard the rate of movement of
lead from painted surfaces to household dust to the extent possible. The interior surfaces of
all Boston homes and the exterior surfaces of all Baltimore homes received loose paint
stabilization approximately one week before soil abatement.
In Boston, loose paint stabilization consisted of removing chipping and peeling paint
and washing the surfaces. Window wells were painted with a fresh coat of primer.
Baltimore homes were wet scraped over the chipping and peeling surfaces, followed by
vacuuming. The entire surface was pruned and painted with two coats of latex paint.
1.2 SUMMARY OF INDIVIDUAL STUDY REPORTS
Following the completion of data collection and analyses, the research teams in each
city prepared individual study reports characterizing hi detail the study design, procedures,
and results obtained hi their respective cities. Some of the more salient features of each
study and key findings reported by the individual city investigators are summarized next.
1.2.1 Boston Study
The Boston study retained 149 of the original 152 children enrolled, although
22 children moved to a new location while continuing in the study. Children with blood lead
concentrations below 7 ptg/dL or above 24 /Ag/dL had been excluded from the study and two
children were dropped from some aspects of the data analysis when they developed lead
poisoning, probably due to exposure to lead-based paint abatement debris at a location away
from then: home.
1-12
-------
Baseline characteristics (age, SES, soil lead, dust lead, drinking water lead, and paint
lead) were similar for the three study groups (BOS P-S, BOS PI-S, BOS SPI). The
preabatement average blood lead concentration was highest for BOS P-S. The proportion of
Hispanics was higher in BOS P-S than in BOS PI-S or BOS SPI, and the proportion of blacks
was lower. There was a larger proportion of male than female children in BOS P-S.
Data were analyzed by analysis of covariance (ANCOVA), which showed a significant
effect of intervention for both the BOS PI-S and BOS SPI groups. These results did not
change following adjustment for age, sex, SES, or any other variable except race and paint.
When the paint variable was controlled, the blood lead declines were diminished and the
results were borderline statistically significant. When the race variable was added, the blood
lead declines were also diminished, and the results were not statistically significant.
Participants were chosen to be representative of the population of urban preschool
children who are at risk of lead exposure. The Boston Childhood Lead Poisoning Prevention
Program identified potential participants from neighborhoods with the highest rates of lead
poisoning. Because study candidates with blood lead levels below 7 ptg/dL or in excess of
24 fig/dL at baseline were excluded from the study, no conclusion about the effect of abating
lead contaminated soil for children ovitside of this range can be made. Similarly, a different
effect might have been found for children having a greater blood lead contribution from soil,
such as in communities with smelters or other stationary sources where soil lead levels are
substantially higher than those seen in this study or where differences in soil properties result
in differences in bioavailability.
Follow-up blood lead measurements were made in Boston 11 months after intervention
and again at 23 months.
1.2.2 Baltimore Study
The Baltimore study recruited 472 children, of whom 185 completed the study; and of
those that completed the study, none were excluded from analysis. The recruited children
were from two neighborhoods, originally intended to be a treatment and a control group.
Because soil concentrations were lower than expected, some properties in the treatment group
did not receive soil abatement. The Baltimore report transferred these properties to the
1-13
-------
control group. In this report, the unabated properties in the treatment group are treated as a
separate control group.
Because of logistical problems, there was an extended delay between recruitment and
soil abatement that accounted for most of the attrition from the project. In their report, the
Baltimore group applied several statistical models to the two populations to evaluate the
potential bias from loss of participating children. These analyses showed that the two
populations remained virtually identical in demographic, biological and environmental
characteristics.
The Baltimore study provided limited information on the impact of house dust as a part
of the change in lead in the child's environment. The study design focused on changes in
biological parameters, hand dust and blood lead, over an extended period of tune. There
were no measurements of exterior dust, no ulterior paint stabilization, and no interior dust
abatement. Except for the abated properties, there were no follow-up measurements of soil
lead concentrations.
Including the prestudy screening measurements of hand dust and blood lead in the
original cohort of participants, the Baltimore study included six rounds of biological
measurements that spanned 20 months, including postabatement measurements made at 2, 7,
and 10 months following abatement.
1.2.3 Cincinnati Study *
The Cincinnati study recruited 307 children, including 16 children born to participating
families during the study, and an additional 50 children who were recruited after the
beginning of the study. In their primary data analysis, the Cincinnati group excluded these
66 children who were recruited after the start of the study, plus 31 children who were living
in nonrehabilitated housing suspected of having lead-based paint and four children (in two
families) who had become lead-poisoned from other causes. Thus, data for 206 children
were analyzed in the Cincinnati report, and results for these 206 children are included in this
integrated report along with 7 of the 31 children living in nonrehabilitated housing. The
remaining 24 were dropped because of insufficient follow-up data.
The Cincinnati study abated soil on 140 parcels of land scattered throughout six
neighborhoods. If soil were the only source of lead in the neighborhoods, exterior and
1-14
-------
interior dust should have responded to the reduction hi soil lead concentrations. However,
exterior dust lead loading decreased only slightly following both soil and dust abatement, and
returned to preabatement levels within one year. Corresponding changes in house dust, hand
lead, and blood lead paralleled changes in exterior dust. Interior dust returned to
preabatement levels about one year after abatement. Because blood lead concentrations also
decreased in the control area, the Cincinnati group concluded that there is no evidence for
the impact of soil and dust abatement on blood lead concentrations. However, this integrated
report concludes, through a detailed structural equation analysis, that there is a strong
relationship between entry dust and interior dust in this subset of the Cincinnati study, where
the impact of lead-based paint was minimized.
Postabatement measurements in Cincinnati were made at 2, 10, 14, and 21 months
following abatement in the first year, and at 3 and 10 months following abatement in the
second year.
1.2.4 Individual Study Conclusions
In their individual city report following the first phase of their study, the Boston group
stated their conclusions as follows:
• "...this intervention study suggests that an average l,856ppm reduction in soil lead
levels results in a 0.8-1.6 pg/dL reduction in the blood lead levels of urban children
with multiple potential sources of exposure to lead. "
During Phase II of the Boston study, soil abatement was conducted hi the two comparison
groups (BOS PI-S and BOS P-S) and follow-up was extended another year in order to assess
the generalizability and persistence of the blood lead decline observed in Phase I. Following
the second phase of the study, the Boston group concluded (Aschengrau et al., 1994):
• The blood lead "reduction in Phase II was somewhat greater than that in Phase I.
The combined results from both phases suggest that a soil lead reduction of
2,060 ppm is associated with a 2.2 to 2. 70 ng/dL decline in blood lead levels. "3
The basis for their conclusions consisted of an analysis of variance comparing mean
blood lead changes among the three intervention groups, paired t-tests for within group
3This value for soil, 2,060 ppm, cited in their published report, was not adjusted by the Boston group with
the Intel-laboratory correction factor of 1.037 in Table 3-6.
1-15
-------
effects, and analysis of covariance with one-at-a-time adjustment for age, SES, race, sex,
paint, water, and mouthing behavior. The analysis of covariance was performed using no
transformation of blood lead data, which appeared to be normally distributed.
The Baltimore group stated their conclusions as follows:
• "Statistical analysis of the data from the Baltimore Lead in Soil Project provides no
evidence that the soil abatement has a direct impact on the blood lead level of
children in the study."
• "In the presence of lead-based paint in the children's homes, abatement of soil lead
alone provides no direct impact on the blood lead levels of children."
The basis for these statements consisted of an adjusted and unadjusted analysis of
selected covariates. The natural log of the blood lead of children in the treatment group
showed no significant difference from the natural log of the blood lead of children in the
control group, even when adjustments were made for age, SES, hand lead, season, dust, soil,
sex, weak mouthing behavior, or strong mouthing behavior. These analyses were made on
two sets of data. The first set consisted of all children enrolled in Rounds one and six. The
second group consisted only of children enrolled hi all six rounds.
The Cincinnati conclusions can be paraphrased from their report as follows:
• Following ulterior and exterior dust and soil lead abatement, blood lead
concentrations decreased (hi Area A) from 8.9 to 7.0 (21%) but increased to
8.7 ^g/dL at 10 mo postabatement. Following ulterior dust abatement alone blood
lead concentrations decreased from 10.6 to 9.2 (13%) 4 mo postabatement and were
18% below preabatement 10 mo postabatement. In the two neighborhoods with no
abatement, blood lead levels decreased by 29 and 6% during these same tune
periods. Other comparisons also revealed no effects of the soil or dust abatement.
• There was no evidence that blood lead levels were reduced by soil lead or dust
abatement in Area A (with soil, exterior dust, ulterior dust abatement). There was
a slight reduction (net reduction over control area) of 0.6 jtig/dL hi Area B that
might be attributed to ulterior dust abatement. This difference is not statistically
significant.
The basis for the Cincinnati conclusions was a comparison of log transformed mean
blood lead concentrations in the three treatment groups between Rounds 1 and 4.
1-16
-------
1.3 SUMMARY OF EPA INTEGRATED ASSESSMENT RESULTS AND
FINDINGS
The original data sets for each of the three participating cities were submitted to EPA,
along with the individual study reports referred to above. Further analysis of the data were
conducted by EPA staff in ORD, especially in the Environmental Criteria and Assessment
Office/Research Triangle Park, NC (ECAO/RTP), now the National Center for
Environmental Assessment (NCEA-RTP). The present intergrated report summarizes
information on the additional EPA statistical analyses and their results.
From the perspective of the child's environment, changes in the soil lead concentration
are expected to bring about changes in the house dust concentration, the hand dust, and the
blood lead concentration. In each of the three studies, the soil lead concentrations were
reduced to approximately 25 to 200 jttg/g in the study area, and for many treatment groups,
there was a reduction of group mean blood leads, although not always statistically significant.
1.3.1 Quality of the Data
In the absence of certified standards for soil and dust, it was necessary to implement a
program that would ensure that chemical analyses performed by the three participating
laboratories would be internally accurate and externally consistent with similar analyses by
other researchers. This program consisted of identifying acceptable analytical and
mstrumental methods, establishing a set of soil and dust standards, and monitoring the
performance of the participating laboratories through an external audit program.
Chemical extraction of an estimated 75,000 soil and dust samples per study presented
a costly burden for the project both in terms of time and expense and there were advantages
associated with nondestructive analysis for a project of this nature. Because of these
considerations, the Scientific Coordinating Panel recommended the use of laboratory scale
X-ray fluoresence (XRF) for soil analysis, on the condition that a suitable set of common
standards could be prepared for a broad concentration range and that a rigorous audit
program be established to ensure continued analytical accuracy. Two groups, Boston and
Baltimore, elected to use laboratory XRF for ulterior dust analysis also, whereas Cincinnati
opted for hot nitric acid extraction with atomic absorption spectroscopy (AAS) for interior
dust and XRF for exterior dust. During the study, the Baltimore group recognized problems
1-17
-------
with analyzing dust by XRF when the sample size was small, less than 100 mg. They
reanalyzed the dust samples by AAS and reported both measurements. In Boston, this
problem was solved by compositing the floor dust samples for XRF analysis, reporting one
floor dust sample per housing unit.
During the project, there were two rounds of soil and dust interlaboratory calibration
exercises, one near the beginning and one at the completion of the soil and dust analyses.
These exercises, which involved the three participating laboratories and two additional
laboratories for each exercise, provided the basis for the evaluation of the performance of
each laboratory hi the audit sample program, and for the conversion factors used to compare
soil and dust data between laboratories.
Each study maintained rigorous standards for database quality. These included double
entry, 100% visual confirmation, and standard procedures for detecting outliers. Some
errors were found during the preparation of this report, and they were corrected in
consultation with the pertinent individual city investigators prior to use in this report. None
of these errors would have impacted the conclusions drawn by the individual study.
1.3.2 Effectiveness and Persistency of Intervention
Soil abatement reduced soil concentrations in all three studies, and there was no
evidence of soil recontamination in either Boston or Cincinnati. There were no follow-up
measures of soil in Baltimore that would detect recontamination. There was some evidence
for exterior dust recontamination hi Cincinnati. The Cincinnati group suggests that this
might be caused by chipping and peeling lead-based paint from the exterior surfaces of
nearby buildings not included hi the project.
Interior dust abatement was persistent hi both Boston and Cincinnati, even though
some recontamination occurred hi Cincinnati hi response to the exterior dust recontamination.
Paint stabilization appeared to have some impact on exposure, but there were no measures of
persistency.
1-18
-------
1.3.3 Comparison of EPA Integrated Report Results with Individual
Study Results
This integrated assessment looks at the three individual studies collectively to
determine if a broad overview can be taken of the project results when each study is placed
hi its correct perspective.
1.3.3.1 Boston Study
The key findings of this integrated assessment with regard to the Boston study are as
follows:
1. The median preabatement concentration of lead in soil was relatively high in
Boston, averaging about 2,400 jtig/g with few samples below 1,000
2. Abatement of the soil effectively reduced the median concentration of lead in the
soil to about 150 /*g/g (an average decrease of about 2,300
3. Soil was clearly a part of the exposure pathway to the child, contributing
significantly to house dust lead.
4. Other sources of lead, such as interior lead-based paint were minimized by
stabilization.
5. The reductions of lead hi both soil and house dust persisted for at least two years.
6. Blood lead levels were reduced by approximately 1.86 jug/dL at 10 mo after soil
lead abatement.
7. Additional reductions in blood lead of about 2.0 /xg/dL (relative to non-abated) were
observed at 22 mo postabatement for children in houses where the soil lead was
abated and the interior house dust lead was consequently reduced and remained low.
The Boston study used analysis of variance methods based on blood lead differences,
and analysis of covariance methods with the longitudinal aspect included by use of the
pre-abatement blood lead concentration (Round 1) as a covariate. The results of their
"crude" analysis (Table 15-10 hi the Boston study report) are virtually identical to the effect
size estimates calculated by U.S. EPA for the group as a whole using repeated measures
ANOVA and also using a longitudinal structural equations model. The results are shown in
Table 1-4. The effect size estimates are somewhat smaller in their "base" model, which the
1-19
-------
TABLE 1-4. COMPARISON OF PHASE 1 EFFECT SIZE ESTIMATES
BETWEEN THE BOSTON STUDY AND THIS REPORT
GROUP STUDY
ABATE VS CONTROL
BOS SPI BOS P-S
BOS SPI BOS PI-S
BOS PI-S BOS P-S
BOSTON
CRUDE
MODEL
1.92
1.53
0.39
REPORT1
BASE
MODEL
1.49
1.28
0.21
THIS REPORT1
RM
ANOVA
1.87
1.54
0.33
LSEM
MODEL 17
1.86
1.56
0.30
1 Units are /tg/dL reduction of Pb in blood.
longitudinal analysis of covariance model adjusted only for pre-abatement blood lead. In
view of the differences in methods and approaches, the overall conclusions are very similar.
The Boston investigators also studied the sensitivity of the effect size estimates to a
large number of other covariates, including environmental factors, family demographic
factors, behavioral factors, and biological covariates. None of these changed the estimated
effect of BOS SPI versus BOS P-S (soil abatement versus control) from their base model,
1.49 jtg/dL, by more than 0.22 /^g/dL. The factors were entered one at a time. The largest
decrease was by. inclusion of race as a factor, which reduced the effect to 1.27 /jg/dL, and
by inclusion of pre-abatement lead paint, which reduced the estimated effect to 1.34 jug/dL.
Five factors decreased the effect size, which nevertheless remained statistically significant:
water lead concentration, time away from home, time away from study area, playing or
sitting on inside floor, and ferritin level. The other 15 factors tested increased the estimated
effect size, particularly age (to 1.61 /Ag/dL) and hand washing before meals (to 1.63 /*g/dL),
as well as: gender, socioeconomic status, mouthing variables, chipping paint, yard play,
outdoor eating, hand washing after outdoor activity, pets that go outdoors, imported canned
food, lead-related occupations, lead-related hobbies, cigarette smoking, and owner
occupancy. Many of these factors are important in identifying individual exposure
components and lead risk factors and are worthy of additional scientific investigation.
However, none of these factors appear to have interacted so strongly with soil and dust
abatement as to have qualitatively affected the conclusions of the study, except for relatively
small effects related to age, race, and lead paint level. Much of the lead paint effect is
1-20
-------
mediated, both statistically and physically, by lead concentrations or loadings in house dust.
It is likely that the use of household dust as a covariate in the models of this report
effectively subsumed the lead paint effect, and that the dust abatement carried out in the
Boston study along with soil abatement may have affected some fraction of the blood lead
response that might have been otherwise attributed to lead-based paint. Even so, the overall
treatment group effect in the model that included lead paint was only slightly less significant
(P = 0.05) than the base model (P = 0.02). Conversely, including chipping paint in the
model increased the effect to 1.53 /xg/dL (P = 0.02 for the group model, P = 0.01 for the
BOS SPI versus BOS P-S effect). Additional studies involving the paint contribution to the
total lead exposure pathways, and assessment of the possible effects and interaction between
paint condition and paint lead loading on lead exposure, are needed to understand the
relatively small modifications of effect size attributable to lead paint.
Age and race effects are larger than the paint effects, and were evaluated in this report.
Larger effects were identified for children of ages 18 to 41 months, and for children of
Afro-American ancestry, than for the sample as a whole. The Afro-American children also
seemed to show larger responses to dust abatement than did the sample as a whole.
In summary, the abatement of soil in the Boston study resulted in a measureable,
statistically significant decline in blood lead concentrations in children, and this decline
continued for at least two years. It appears that the following conditions were present, and
perhaps necessary for this effect: (a) a notably elevated starting soil lead concentration (e.g.,
in excess of 1,000 to 2,000 jwg/g); (b) a marked reduction of more than 1,000 /^g/g in soil
lead consequent to soil abatement accompanied by (c) a parallel marked and persisting
decrease in house dust lead.
These conclusions are consistent with those reported by the Boston research team. This
integrated assessment found no basis for modifying then: conclusions, although we choose not
to express these findings as a broadly generalizeable linear relationship between soil and
blood, such as change in micrograms of lead per deciliter of blood per change in micrograms
of lead per gram of soil, because we believe that such a linear expression of abatement
effects is highly site specific for the soil-to-blood relationship. We found evidence that the
dust-to-blood relationship is more significant than the soil-to-blood relationship and therefore
the abatement effect also depends on soil-to-dust transfer, which may be very site-specific.
1-21
-------
1.3.3.2 Baltimore Study
With regard to analyses of Baltimore data conducted for this integrated assessment, the
participants in the abatement neighborhood that did not receive abatement were treated as a
separate control group, rather than being combined with the nonabatement neighborhood (as
the Baltimore research team did). The reason for this was to establish a control group not
influenced by differences between neighborhoods. This alternative approach used in this
integrated assessment had little impact on the statistical significance of soil abatement effects
as reported by the Baltimore research team.
The key findings of this integrated assessment for Baltimore are:
1. The preabatement concentrations of lead hi soil were notably lower (i.e., averaging
around 500 to 700 /xg/g, with few over 1,000 /ig/g) than in Boston.
2. The actual reduction of lead hi soil by abatement was small (a change of about
400 /tg/g), compared to the Boston study (a change of about 2,300
3. Measurements of blood lead were made for only ten months following abatement;
and no significant decreases in blood lead consequent to soil abatement were
observed compared to non-abatement control group children.
4. Except for exterior lead-based paint, there was no control of other sources of lead,
such as the stabilization of ulterior lead-based paint (as done in Boston) or
abatement of house dust (as done hi Boston and Cincinnati).
5. Follow-up measurements of soil (except immediately postabatement) were not made
to establish the persistency of soil abatement, and its possible effects on house dust.
The Baltimore report used a generalized linear regression model (GLIM). In its
simplest form, the regression model can be expressed as a linear model using log-
transformed variables. The Baltimore blood lead model 1 is a simple ANOVA model,
Log(Bqp = G
a
with only two treatment groups, Area 1 and Area 2. However, Area 1 includes some
non-abated residences as well the residences that received soil abatement, whereas Area 2
includes only non-abated residences. Therefore, the results in the Baltimore report cannot be
directly compared with the results reported here, where we have separated the abated and
1-22
-------
non-abated residences into two groups and used the non-abated residences in Area 1 as a
i ' .
second control group. Model 2 in the Baltimore report is a simple ANCOVA model,
i '
i .
logCBCy) .= Ggj + b2j Agey + b3j SES; +jb4j Seasony + b6j My log(Handy) + b7j (1-My) log(Handy) + ey.
i
i '
I " -• •
In this notation, Age is a semi-categorical variable, Season is included only for pre-abatement
rounds 1 and 2 that covered many ijnonths, and My is a dummy variable for low or high
mouthing behavior. While temporal comparisons are possible, no temporal correlation model
is assumed, and the Baltimore repoit notes that the lack of temporal modeling is a deficiency
hi the analyses. i
The Baltimore analyses were carried out for two distinct subgroups of children. The
first set of analyses used only thosej children who were present hi all six rounds. The second
set of analyses used all children who were present hi each round. Analyses for this EPA
Integrated Report used children who were present in Rounds 3,4, and 6. The set of children
who were present in all rounds is included in the EPA set, but does not include other
children in the EPA set (such as those children who were recruited at Round 3, especially
very young children). The second set of children in the Baltimore study is much closer to
the EPA children set in Rounds 4 and 6, but includes in Round 3 some additional children
who dropped out after Round 3. Therefore, the EPA effects size estimates are based on
somewhat different groups of specific children than hi the Baltimore report.
i •
Effecjt sizes were calculated in Table 1-5 as simple differences of treatment group
i
effects reported in Tables 7-7 and 7-8 of the Baltimore report. The effects were small and
i ">
probably not statistically significant! although the lack of correlation structure in the
Baltimore models makes any estimates of standard errors rather questionable. The
differences hi blood lead are negatiye between the treatment group (BAL SP) and the control
i
group (BAL PI and BAL P2). There is little reason to believe that major treatment group
differences would have been identified by other analyses of these data.
Other findings hi the Baltimore study are of interest. There were some indications of
significant differences associated with hand lead, with a modifying effect due to child
|
mouthing behaviors. There was also a strong effect of socioeconomic status on blood lead
1-23
-------
TABLE 1-5. EFFECT SIZE ESTIMATES FROM THE BALTIMORE REPORT
COMPARING BLOOD LEAD REDUCTION IN BAL SP VERSUS CONTROLS
ROUNDS
ROUNDS 3
AND 4
ROUNDS 3
AND 6
CHILD
GROUP
ALL 6
ROUNDS
EACH
ROUND
ALL 6
ROUNDS
EACH
ROUND
BALTIMORE
ANOVA
-0.55
-0.07
-0.92
-1.55
MODEL1'2
ANCOVA
0.12
-0.10
-0.71
-1.17
THIS REPORT1'3
BALSP
vs BAL PI
0.07
-0.54
BALSP
vs BAL P2
1.774
0.67
1 Units are /tg/dL reduction of Pb in blood.
2 Baltimore controls are BAL PI and BAL P2.
3 Children present in Rounds 3,4, and 6.
4 P=0.16; others, P>0.2.
and dust lead, and an age effect with maximum blood leads at ages 1 to 3 years (12 to
36 months), a general finding in these studies.
Thus, in Baltimore, where the difference between pre- and postabatement soil lead
concentrations was much less than hi Boston, and where the soil abatement criteria left some
properties only partially abated and no ulterior paint stabilization or dust abatement was
performed, no detectable effects of soil lead abatement on blood lead levels were found.
These conclusions are consistent with those reported by the Baltimore research group,
and are not inconsistent with those above for the Boston study. At soil concentrations much
lower than the Boston study, the Baltimore group would have likely been able to see only a
very modest change hi blood lead concentrations (perhaps less than 0.2 /xg/dL) assuming
similarity between the study groups in Boston and Baltimore and the same linear relationship
between change hi soil concentration and change in blood lead. Furthermore, the interior
paint stabilization and house dust abatement performed in Boston likely enhanced and
reinforced the impact of soil abatement on childhood blood lead, whereas in Baltimore, any
1-24
-------
possible small impact of soil abatement would have likely been swamped by the large
reservoir of lead in the interior paint and the large unabated amounts of lead in interior house
dust.
1.3.3.3 Cincinnati Study
As for the Cincinnati study, because of differences hi the neighborhoods, we found that
combining neighborhoods into treatment groups often obscures important effects; and so we
chose to analyze each of the six Cincinnati neighborhoods as separate treatment groups. One
neighborhood, Back Street, had an insufficient number of participants and was dropped from
some analyses; that group started with nine families, but by Round 5 there was only one
participating family in the study. We also found that the two control neighborhoods,
Glencoe and Mohawk, were substantially different, and that the three remaining treatment
groups (Pendleton, Dandridge, and Findlay) were more comparable, both demographically
and in geographic proximity, to Mohawk than to Glencoe.
The Cincinnati study used several different regression (ANCOVA) models, and cross-
sectional structural equation models. Their individual city report also included results of a
simple correlation analysis that did not allow for multiple covariate adjustments, and is not
further described. The response variables in the regression models included differences hi
blood lead between Round 1 and Round 4, hand lead differences, and differences in ulterior
floor dust loading and in exterior dust loading. The final regression model for the change in
blood lead involved only blood lead concentration (which we denote Blood), hand lead
loading (which we denote Hand), age of the child at the Round 4 blood lead measurement
(which we denote Bloodi4), and socioeconomic status (denoted SES). In our notation, their
model is:
Bloodi4 - Bloody = 8.52 + 0.038 (Handi4 - Handu) - 0.00079 Agei4*Handi4
- 0.17 SES- 0.43 Blood^
This model has one point of similarity to our Cincinnati longitudinal SEM models. By
transposing the Bloodn on the left side of the equation, we have a linear relation that is
expressed algebraically as Bloodi4 = 8.52 + ... other terms + 0.57 Blood^, which is close
1-25
-------
to the value of the blood lead persistence parameter A14 obtained for most of the Cincinnati
LSEM models, such as A14 = 0.58 hi Model J6 used hi the effects size comparisons.
Otherwise, blood lead is not predicted by neighborhood, nor by abatement group, nor by
environmental lead concentrations or loadings, but by another tune-variable and child-specific
variable, hand wipe lead loading, which tends to increase with the child's age. The
regression model for hand lead change also excludes treatment group or environmental
variables, except indirectly through Round 1 hand lead.
Their report also presents a structural equation model for blood lead and hand lead
differences, and for changes in ulterior and exterior dust lead. Their equations for blood and
hand lead are, in our notation:
Bloodi4 - Bloodu = 10.28 - 0.18 SES - 0.064 Agei5 - 0.46 Bloodn
Hand14 - Hand^ = 5.78 + 0.002 Handi5 - 0.62 Handu.
The two dust lead equations are totally unconnected to blood lead or hand lead.
The Cincinnati report also shows cross-sectional structural equation models for
Round 1, Round 3, and Round 4 respectively. The Round 1 SEM model shows large and
statistically significant age effects, and effects of mouthing behavior. Areas and
neighborhoods show no significant differences. The Cincinnati cross-sectional SEM model
uses no environmental covariates, but reports a significant regression of log(BloodRl) on
log(HandRl). The simultaneous equation for log(HandRl) depends strongly on age and not
at all on treatment group or neighborhood. Neither equation uses any of the environmental
covariates, but both include a significant fixed effects factor for "families", which is
analogous to the random effects term Hh(g) in our EPA repeated measures ANOVA and
ANCOVA models. However, their findings of no significant neighborhood differences or
environmental factors differs somewhat from some of the findings in our EPA cross-sectional
and longitudinal SEM models. Differences in model format and structure make direct
comparisons very difficult.
The Cincinnati investigators concluded that the Phase 1 changes in blood lead
concentrations and in hand lead loadings were not significantly different among the three
1-26
-------
abatement groups, using either multiple regression models or structural equation models.
They did not compare across different neighborhoods within treatment groups, which was an
additional source of variability in the study. We cannot therefore directly compare our effect
sizes or treatment differences across neighborhoods with their aggregated results. Since their
models are not directly comparable to our models without additional substantive analyses of
the role of hand wipe lead, we cannot directly compare effect sizes using longitudinal SEM.
The Cincinnati report giving a cross-sectional SEM for Round 4 (their Table 4-63)
presents a comprehensive and detailed SEM, which is in substantial qualitative agreement
with the EPA longitudinal SEM presented here for Cincinnati Round 4 blood lead and dust
lead. The use of hand lead in their model precludes direct comparisons with the longitudinal
SEM in Table 5-39. The use of log(HandR4) as a covariate that is only partially adjusted by
window and floor dust lead loadings, age, and SES permits the finding of large, statistically
significant, but negative relationships between log(BloodR4) and dust lead loadings on the
floor, ulterior entry, and exterior. Additional analyses of this model would be useful. The
model uses neighborhood or area as an adjustment covariate for hand-to-blood, dust-to-blood,
dust-to-hand, paint-to-dust, and exterior-to-floor pathways, with some significant differences.
While the application of this model does not allow comparison of effect sizes relative to
Round 1, there is a qualitative similarity between our EPA findings and those of the
Cincinnati investigators.
On this basis, we concluded that, hi most cases, the effect of soil abatement could not
be clearly determined, and offer the following explanation for this conclusion:
1. Most of the soil parcels hi each neighborhood were not adjacent to the living units,
and this soil was therefore not the primary source of lead in house dust. Evidence
for this statement includes the observation that street dust lead concentrations are
much higher than soil concentrations, indicating there is a large source of lead
contributing to street dust in addition to soil lead.
2. The preabatement median soil lead concentrations in the three treatment groups
were about 300 /ig/g in Pendleton, 700 jug/g in Findlay, and 800 jug/g in
Dandridge, and the postabatement soil concentrations were less than 100 /-tg/g, so
that the reduction of lead in soil was small, as in Baltimore.
Evidence for the impact of dust abatement or dust and soil abatement consists of a
statistically significant difference between changes in blood lead between Rounds 1 and 4,
approximately one year apart. Some Cincinnati neighborhoods showed decreased blood lead
1-27
-------
concentrations in response to dust abatement or dust and soil abatement. The two
neighborhoods that received only interior dust abatement in the first year, Dandridge and
Findlay, showed a small decrease hi blood lead concentrations, compared to large increases
in the nearest control group, Mohawk. The treatment group that received soil, exterior dust
and interior dust abatement, Pendleton, showed a smaller effect than did the Dandridge and
]?indlay neighborhoods. After consultation with the Cincinnati research team, we suspect
that there was recontamination of street dust hi Pendleton during the study, probably caused
by demolition of nearby buildings hi the neighborhood.
The consistent theme across the outcomes for all three studies is that soil abatement
must effectively reduce soil lead concentrations for an extended period of time and be
accompanied by a corresponding reduction hi house dust lead in order to result in any
detectable reduction of blood lead. The location of the soil relative to the exposure
environment of the child is important. In this project, the movement of lead from soil or
street dust into the home seems to be a key factor hi determining blood lead concentrations.
Although these USLADP results provide substantial evidence for the link between soil or
street dust and house dust lead, there is insufficient information by which to clearly quantify
this relationship hi terms of the lowest level of soil or street dust lead reduction that will
yield a measurable decrease of lead in blood.
1.3.3.4 Synthesis of Findings Across the Three Studies
While the USLADP was not intended to compare different methods for soil abatement,
the differences hi design and methodology among the three studies helped to identify
conditions for which soil abatement may be an effective intervention, and conditions under
which soil abatement is less likely to be effective. Abatement or intervention can be
effective if it can achieve one or both of the following goals:
1. Abatement or intervention produces an effective and persistent reduction in the
concentrations of lead hi soil and hi household dust.
1-28
-------
2. Abatement or intervention changes childhood lead exposure by reducing the
intake of lead-contaminated media, or effectively breaks the transport pathway from the lead-
contaminated source to the child's activity areas.
These are not mutually exclusive goals, but there are important distinctions between them.
The first goal, reducing lead concentrations, can be achieved without changing exposure or
transport. For example, removing bare lead-contaminated soil from a yard and replacing it
with bare soil that is not contaminated will not change the child's exposure nor the transport
of surface soil from the yard into the house. However, the child's intake of lead from soil
ingestion will immediately be reduced, and one would expect that over some period of time,
there will be a reduction of the child's intake of lead from household dust because the soil
component of household dust lead has been eliminated. All three studies achieved the
elimination of lead in yard soil. It is important to note the requirement that the soil not be
recontaminated by unreinediated sources such as exterior paint and by transport of lead from
unremediated areas. Even in the Boston study, a few soil-abated yards became substantially
recontaminated. However, most of the sampled locations in the Boston and Cincinnati did
not suffer significantly recontaminated soil after abatement. The Baltimore sites were not
followed up over a similar period of time.
Both Boston and Cincinnati residences received interior dust abatement. The Boston
residences showed slight evidence of recontamination, whereas most the residences in the
Cincinnati areas that received interior dust abatement (with or without soil abatement) during
Phase 1 of the study showed significant recontamination. The floor dust lead concentrations
showed a significant association with window lead and mat lead, suggesting exterior sources
of recontamination. Long-term changes in dust lead were not followed up hi the Baltimore
study. Significant blood lead reduction was detected only in the Boston study, where
persistent reduction of dust lead occurred in most residences that received soil lead and
interior dust abatement. The effect was even greater in Phase 2 in the group BOS PI-S that
received both Phase 1 dust abatement and both soil and dust abatement in Phase 2.
The second goal, reduction of exposure, requires reducing the amount of potentially
lead-contaminated media consumed by the child. The Boston study shows some indication
that this also may have occurred, whereas the Cincinnati provides little indication of
1-29
-------
exposure reduction. Soil abatement can reduce exposure by covering soil with sod or other
barriers that reduce the child's access to surface soil particles. The reduction in exposure is
distinct from reducing the lead concentration in the soil to which the child is exposed.
Likewise, frequent and effective washing or vacuuming of household dust can reduce the
amount of dust (dust loading) that is accessible to the child, however much lead is in the
dust. Changes hi behavior, such as more frequent hand washing or greater parental
attention, can also reduce contact with dust and soil. Since all of these studies may have
initiated behavioral changes from the moment of recruitment simply by informing parents and
caretakers of potential lead hazards, such changes cannot be detected with this study design.
The second goal can also be achieved by any process that reduces transport of the
contaminant from the source to the areas hi which the child may come into contact with it.
Covering bare soil with sod, concrete, or other barriers will clearly prevent contamination of
house dust and outside play areas, as the encapsulation of paint will prevent paint chips from
contaminating dust, so long as the barrier remains intact. Removing the source of lead
contamination was shown to be effective in Boston, but in addition to this, there is also some
possibility that the post-abatement pathway regression coefficient from soil to dust may have
been changed. However, there may also have been a serious attenuation of the apparent
pathway in the Round 1 data set, possibly attributable to the blood lead truncation of the
study. Additional studies on the effects of soil abatement on environmental lead pathway
kinetics would be useful. In general, any method that attempts to estimate post-intervention
or post-abatement blood lead concentrations (for example, EPA's IEUBK Model or "slope
factor" models) should take into account not only the changes in environmental lead
concentrations that may occur as a results of abatement or intervention, but also the changes
in the pathways to childhood exposure that may occur following abatement or intervention.
Finally, one should recognize that any environmental lead abatement or intervention
may be limited hi its ability to reduce blood lead concentrations hi currently lead-burdened
children. It appears that in the first year after abatement, at most 40 to 50 percent of the
child's existing blood lead burden may be removable by soil abatement or any other
combination of abatements and interventions apart from medical treatment by chelation.
There may be a much greater effect of lead abatement hi preventing lead exposure for future
1-30
-------
residents. Long-term benefits of lead abatement should therefore be considered in assessing
abatement effectiveness, as well as short-term benefits.
1.4 INTEGRATED PROJECT CONCLUSIONS
The main conclusions of this Integrated Report report are two-fold:
(1) When soil is a significant source of lead in the child's environment, under certain
conditions, the abatement of that soil will result in a reduction in exposure that will
cause a reduction in childhood blood lead concentrations.
(2) Although these conditions for a reduction in blood are not fully understood, it is
likely that five factors are important in determining the magnitude of any possible
reduction: (1) the past history of exposure of the child to lead, as reflected in the
preabatement blood lead; (2) the initial soil lead concentration and the magnitude
of the reduction in soil lead concentrations; (3) the initial interior house dust lead
loading and the magnitude of reduction in house dust lead loading; (4) the
magnitude of other sources of lead exposure, relative to soil; and (5) the strength
of the exposure pathway between soil and the child relative to other lead exposure
pathways in the child's environment.
The basis for the first conclusion is: in Boston, where the soil lead concentrations were
high (mostly > 1000 to 2000 /tg/g) and the contribution from lead-based paint was reduced
by paint stabilization, there was a measurable reduction of blood lead concentrations. This
reduction continued to increase for two years following abatement in Boston.
Conversely, in Baltimore and Cincinnati, where soil was not a significant source of lead
relative to other sources, there was no measurable reduction of blood lead except in cases
where other sources were also removed or abated. In Baltimore, these sources may have
been interior lead-based paint that was not stabilized, or house dust that was not abated.
In Cincinnati, the principal source of lead seemed to be neighborhood dust that may have
been contaminated with lead-based paint.
The basis for the second conclusion is: firstly, in those cases where all important
elements of the exposure pathway were available for assessment, the structural equation
model analyses showed that preabatement blood lead concentration was an important
predictor of postabatement blood lead, suggesting that the remobilization of bone lead is an
important component of the measured blood lead. Secondly, all other factors being equal,
the measurable reduction in blood lead was observed only at higher concentrations of soil
1-31
-------
lead. In the absence of information about other sources of lead, no clear statement can be
made about the possibility of smaller reductions in blood lead at lower soil lead
concentrations.
In spite of the recent successes in reducing exposure to lead by removing lead from
gasoline and canned food, lead exposure remains a complex issue. This integrated
assessment attempts to assess exposure to lead hi soil and house dust. Lead in soil and
lead-based paint are closely linked hi the child's environment. If there is exterior lead-based
paint, then soil lead is likely to be elevated with a consequent elevation in house dust lead.
If there is interior lead-based paint, then efforts to reduce the impact of soil lead on house
dust will be only partially effective. The maximum reduction in lead exposure will not be
achieved unless both paint and soil abatement are implemented.
There is evidence from all three studies that lead moves through the child's
environment. This means that lead hi soil contributes to lead hi street or playground dust,
lead in exterior paint contributes to lead in soil, and lead in street dust contributes to lead in
house dust. A more detailed analysis of the data may show the relative contribution from
two or more sources, but the present analyses imply that this transfer takes place.
The analysis of the data from the three studies showed evidence that blood lead
responds to changes hi house dust lead. There is also evidence for the continued impact of
other, independent sources following abatement of one source. This means that abatement of
soil or exterior paint does not necessarily reduce the contribution of lead from other sources
such as ulterior lead-based paint.
The conclusions of this report suggest that soil abatement can have a measurable effect
on reducing exposure to lead if there is a substantial amount of lead in soil and if this soil
lead is the primary source of lead hi house dust. In such cases, both soil abatement and
interior dust removal should be performed to be fully effective. Likewise, soil abatement
should be considered hi conjunction with paint abatement when it is likely that soil will
otherwise continue to contaminate house dust after a paint abatement is completed.
From one perspective, decisions about soil abatement should be made on an individual
home basis. This report shows that, on an individual house basis, soil abatement may reduce
the movement of lead into the home and its incorporation into house dust. The magnitude of
this reduction depends on the concentration of lead in the soil, the amount of soil-derived
1-32
-------
dust that moves into the home, the frequency of cleaning hi the home and the cleanability of
the home. The number and ages of cljiildren and the presence of indoor/outdoor pets are
factors known to increase this rate of dust movement, whereas frequent cleaning with an
effective vacuum cleaner, use of entry! dust mats, and removing shoes at the door serve to
reduce the impact of soil lead on house dust.
From another perspective, soil abatement at the neighborhood level poses problems not
pertinent to individual homes. Playground, vacant lot, and other plots of soil may pose an
immediate problem if they are accessible to children and there is a direct pathway for dust
I;
generated by this soil to enter the noire. Likewise, sources of lead other than soil may
contribute more to exterior dust than soil itself. The evidence in this report suggests that the
key to reducing lead exposure at the neighborhood level is to abate significant sources of lead
contributing to exterior dust, in additic n to the soil and paint abatement that would be
performed on an individual property.
1-33
-------
-------
2. BACKGROUND AND OVERVIEW OF PROJECT
2.1 PROJECT BACKGROUND
2.1.1 The Urban Lead Problem
Children are exposed to lead through complex pathways from multiple sources. In the
mid 1980s, when there was a dramatic increase in public concern for childhood lead
exposure, attention focused on urban environments with high concentrations of lead hi soil,
where there was an apparent correlation with the incidence of high blood lead concentrations.
At that time, there were several other sources of exposure that could potentially account for
unusually high blood lead in a population of urban children. Among these were lead hi the
air (primarily from automobile emissions), lead hi food (primarily from canned foods with
lead soldered side seams), lead hi drinking water (primarily from lead pipes or newly
soldered copper pipes), and lead hi paint. The lead hi the soil was believed to be a mixture
of lead from the atmosphere and lead from exterior paint. Regulations were hi place that
would largely remove lead from gasoline by the end of 1986, and there was a voluntary
program among food processors to phase out cans with lead soldered side seams. Renewed
public interest in paint abatement emerged in the late 1980's.
Prior to the start of this project, soil abatement had been performed in many nonurban
residential areas with elevated soil lead. The decision to abate soil was usually based in part
on the distribution of blood lead within the population of children. There was limited
experience on the effectiveness of this abatement and little or no opportunity for follow-up
studies of the results. There were little data from controlled evaluations because the intent of
abatement was remediation, not experimentation.
2.1.2 Legislative Background
In the mid 1980s, the scientific evidence for a correlation between soil lead and blood
lead was sufficient to warrant concern for the health of children, but not strong enough to
support a large scale program for soil lead abatement. Consequently, the Urban Soil Lead
Abatement Demonstration Project (USLADP), known also as the Three City Study, was
2-1
-------
authorized in 1986 under Section lll(b)(6) of the Superfund Amendments and
Reauthorization Act (SARA). SARA called for EPA to conduct a "pilot program for the
removal, decontamination, or other actions with respect to lead-contaminated soil in one to
three different metropolitan areas."
Although not specified in the amendment, the legislative history focused on lead-based
paint as the source of lead in soil in urban residential areas. In response to the Superfund
mandate, USLADP was designed to evaluate the effectiveness of removal of lead-
contaminated soil hi urban residential areas as a means to reduce blood lead levels of young,
preschool children residing hi abated residences or neighborhoods. The project was not
designed to evaluate the relative effectiveness of different soil abatement technologies per se,
but rather to focus on determining the extent to which the blood lead levels of children less
than six years old ( as a key risk group for lead health effects) could be reduced by
intervention to decrease soil lead concentrations.
The EPA's Office of Solid Waste and Emergency Response (OSWER) had the principal
responsibility for overall implementation of the project, as a Superfund-mandated activity.
Administrative and financial management responsibilities, it was decided, were to be
delegated to EPA regional offices for the geographic areas containing those cities selected for
inclusion in the project. EPA's Office of Research and Development was asked to provide
technical oversight and coordination assistance to help integrate scientific activities across the
cities selected. An EPA Steering Committee was set up to oversee site selection and
initiation of the project.
In 1987, EPA convened a set of experts to advise on the design of the project and to
develop selection criteria for study sites. Six cities submitted proposals, and Boston,
Baltimore, and Cincinnati were chosen by the following site selection process.
2.1.3 Site Selection
The three cities were selected based on an evaluation of each proposal in relationship to
the following site selection criteria, as recommended by the experts.
A. To be considered for selection, a metropolitan area must have:
1. Agreement by the appropriate EPA regional office to provide general project
oversight, and to disburse the funds.
2-2
-------
2. An established entity, preferably the state, documented as willing to be responsible
for removing and disposing of lead contaminated soil. This included identification
of an appropriate facility within the state for disposal of the soil, facilitation of
permits, community relations and education, and any other activities necessary to
expeditiously provide for safe disposal.
3. The administrative infrastructure to carry out a large scale project. This included a
key government department with appropriate authority to coordinate the project,
and generally included active participation by the state, by community groups, and
by all the different metropolitan departments with some responsibility for the
project.
4. Access to scientific and medical expertise to ensure that sampling and analysis were
properly conducted, and access to medical care needed for any children found to
have lead toxicity.
5. Evidence that there are children with elevated blood lead levels (25 /xg/dL as
defined by the CDC in its 1985 childhood lead screening guidelines), and soil hi
residential areas with lead levels of 1,500 /*g/g or greater.1 It would be desirable
for lead-based paint to be established as a major contributor to the soil lead levels.
B. To be considered for selection, a metropolitan area should have:
6. A documented high incidence of children with elevated blood lead levels hi the
proposed study areas. This meant that the municipality supported an active
childhood lead screening program.
7. A pattern of high density population in study areas. The number of children
available for evaluation as part of the project was important to the statistical
validity of the study.
8. Availability of other sources of funding for portions of the project not funded by
SARA. Such items might include de-leading the outside of houses, or intensive
interior vacuuming to remove residual leaded dust.
The Steering Committee reviewed proposals from six metropolitan areas: Boston,
Baltimore, Cincinnati, Minneapolis, Detroit, and East St. Louis. These were reviewed on
December 3 and 4, 1987, by the Steering Committee and the set of expert consultants.
Boston, Baltimore, and Cincinnati were selected based on the following key points:
1 Note that the stipulated soil value of 1,500 jig/g was interpreted as a significant number of soil parcels in
which at least one soil measurement, exceeded this value. Reports in this document of means or median values
below 1,500 pg/g for individual soil parcels or entire treatment groups should not be misinterpreted as failure
to meet the original selection criteria.
2-3
-------
1. The Boston investigators proposed to select three groups of families randomly from
several neighborhoods known to have soil lead concentrations hi the range of
2000 to 5000 jitg/g. One of these groups would receive only paint stabilization; a
second group would receive paint stabilization and dust abatement, and the third
group would receive soil abatement, dust abatement, and paint stabilization.
2. The Boston proposal involved collaboration among Boston City Hospital, Boston
University, and the EPA Region I Laboratory (for conduct of analysis of lead in
soil, dust, etc.). This collaborative group also had demonstrated experience with
collection, analysis, and assessment of soil and blood lead data in inner city
neighborhoods of Boston. .
3. Cincinnati proposed a neighborhood level abatement study where housing units had
been previously gutted and rehabilitated approximately 20 years ago, and most of
the lead-based paint was either removed or encapsulated. The Cincinnati sites
contained soil lead from 220 to 900 /*g/g, exterior surface dust (primarily from
paved areas) averaging 2,000 to 5,000 ptg/g, and a number of children with blood
lead concentrations above 25 /Ltg/dL.
4. The Cincinnati proposal was prepared by the University of Cincinnati and
demonstrated a high degree of organizational infrastructure, with commitments
from the City of Cincinnati. There was an established infrastructure of
neighborhood associations that was perceived to be a plus for the project.
5. The Baltimore project proposed individual housing units with soil lead
concentrations averaging hi excess of 1,000 /xg/g. Lead-based paint had been
abated hi some, but not all houses.
6. The Baltimore proposal was prepared by the State of Maryland and showed a
satisfactory level of organizational infrastructure and local scientific expertise;
problems with the proposed statistical approach were resolved by consultation with
the Steering Committee.
With the selection of Boston, Cincinnati, and Baltimore, a Scientific Coordinating
Committee was established to provide scientific and technical support for the three studies
and to coordinate the exchange of scientific information. The Steering Committee also
recognized that there would be much value hi standardizing and coordinating methods for
media sampling and analysis. The Scientific Coordinating Committee was composed of
representatives from the research teams of each of the three cities, the three EPA regional
offices (Regions I, HI, and V), the Office of Solid Waste and Emergency Response, the
Environmental Criteria and Assessment Office/Research Triangle Park, NC (ECAO-RTP)
(now the National Center for Environmental Assessment/RTF), and the Centers for Disease
2-4
-------
Control and Prevention. The task of organizing, scheduling, and conduct of meetings of the
Scientific Coordinating Committee was assigned to ECAO/RTP. Major policy decisions
remained with the Steering Committee.
The funding mechanisms were set into place individually through the respective EPA
regional offices (Regions I, III, and V). Each of these regional offices set up an independent
funding mechanism and oversight plan. The regional project officer became the liaison to
the-Steering Committee and to the Scientific Coordinating Committee. Each city submitted a
work plan, which included the project description, organization, operation plan, and
reporting mechanisms, and the Quality Assurance (QA) plan. These work plans required
more than one year to complete and obtain Regional approval. In the meantime, the projects
were staffed and made operational. Community relations programs were initiated that began
the process of recruiting the study participants. Coordination between the three cities was
accomplished through a series of workshops, organized and convened by ECAO/RTP,
approximately three per year.
This integrated assessment includes a review of the hypotheses and study designs of the
individual studies (Chapter 2), a report of the methods intercomparison and quality
assurance/quality control program (Chapter 3), a summary of the individual study results and
conclusions reported by the three cities (Chapter 4), a description and explanation of the
statistical procedures performed as part of this EPA integrated assessment and the results of
these procedures (Chapter 5), and a summary of key findings and conclusions derived from
this assessment (Chapter 6).
2.2 INTEGRATION OF THE THREE STUDIES
2.2.1 Study Hypotheses
To place this project in perspective, it is helpful to look at the similarities and
differences among the three studies. They are similar hi that their hypotheses and study
designs were derived to evaluate the same general hypothesis, namely, that removing lead
from soil will reduce lead exposure of young children.
2-5
-------
The central hypothesis of the USLADP is
A reduction of lead in residential soil accessible to children will
result in a decrease in their blood lead levels.
Each study chose to develop a specific hypothesis that could be tested by data and
observations from their own study design. The formal statement of the Boston hypothesis is
A significant reduction (equal to or greater than 1,000 pg/g) of lead
in soil accessible to children mil result in a mean decrease of at
least 3 ng/dL in the blood lead levels of children living in areas with
multiple possible sources of lead exposure and a high incidence of
lead poisoning.
The Baltimore hypothesis, stated in the null form, is
A significant reduction of lead (^ 1,000 pg/g) in residential soil
accessible to children will not result in a significant decrease (3 to
6 v-gldL) in their blood lead levels.
The Cincinnati hypothesis, separated into two parts, is
(1) A reduction of lead in residential soil accessible to children will
result in a decrease in their blood lead levels.
(2) Interior dust abatement, when carried out in conjunction with
exterior dust and soil abatement, would result in a greater
reduction in blood lead than would be obtained with interior dust
abatement alone, or exterior dust and soil abatement alone.
Secondary hypotheses in the Cincinnati study are
(3) A reduction of lead in residential soil accessible to children will
result in a decrease in their hand lead levels.
(4) Interior dust abatement, when carried out in conjunction with
exterior dust and soil abatement, would result in a greater
reduction in hand lead than would be obtained with interior dust
abatement alone, or exterior dust and soil abatement alone.
2.2.2 General Study Design
The project objective was to measure the relationship between soil lead and blood lead.
This is an indirect relationship in the sense that children most commonly do not eat soil
2-6
-------
directly but usually ingest small amounts of dust derived, in part, from this soil. Likewise,
the lead in blood reflects not only recent exposure from all environmental sources, but the
remobilization of lead from bone tissue.
Each study was designed around the concept of participating families within a definable
neighborhood. There were a total of twelve neighborhoods hi the project, six in Cincinnati,
four in Boston, and two in Baltimore. Except in Boston, these neighborhoods constituted the
treatment and control groups hi the study. In Boston, families hi the treatment group were
randomly assigned from volunteers from each of the four neighborhoods, as were families in
the control group. For each treatment group, there was a preabatement, abatement, and
postabatement phase. The immediate residential environment of the child was extensively
evaluated prior to and after abatement, through measurements of lead in soil, dust, drinking
water, and paint, and through interviews about activity patterns, eating habits, family
activities, and socioeconomic status. Parallel environmental and biological measurements, as
well as interviews, were taken in the control groups, but without abatement. The objective
of the preabatement phase was to achieve a clear understanding of the exposure history and
status (stability of the blood lead and environmental measures) prior to abatement. During
the abatement phase, attention was given to preventing any possible exposure that might
result from the abatement activities. During the postabatement phase, the project was
designed to determine the duration of the effect of soil abatement and to detect possible
recontamination.
The array of treatment groups differed considerably among the three studies. Each
treatment group, however, had several features in common. All groups were taken from
demographically similar neighborhoods with some prior evidence of elevated lead exposure,
usually a greater than average number of public health reports of childhood lead poisoning.
Each group received the same pattern of treatment: baseline phase for 3 to 18 months,
intervention (except for controls), and follow-up for 12 to 24 months.
In each treatment group, even the controls, there was an attempt to minimize the impact
of chipping and peeling lead-based paint. In Boston, this was done by paint stabilization of
interior paint. In Baltimore, only exterior paint was stabilized. Therefore, in these two
studies, the effects of soil abatement should be evaluated in the context of some intervention
for lead-based paint. In Cincinnati, most of the living units were abated of lead-based paint
2-7
-------
more than 20 years before the start of the study. In the case of those that had lead-based
paint, the lead-based paint was measured but not treated prior to the study.
The Boston and Baltimore studies used a parallel intervention scheme, compared to the
staggered scheme used in Cincinnati. In other words, intervention in Boston (and Baltimore)
took place at the same time for all treatment groups, and the follow-up period was of the
same duration. But hi Cincinnati, the soil and exterior dust intervention was delayed for
three neighborhoods, such that follow-up varied between 12 and 24 months. Throughout all
phases of each study, the tuning of the blood lead measurements was planned according to a
seasonal cycle of blood lead levels that peaks hi the late summer and according to an
age-related pattern that peaks at 18 to 24 months.
The complex nature of this project required measurement of exposure indices, such as
street dust, house dust, and hand dust, that are in the pathway between soil and blood. New
sampling and analysis protocols for these measurements, not generally available in the
scientific literature, were developed during the initial coordinating workshops.
The studies differ hi several respects. The two pathways, (1) soil -* exterior dust and ,
(2) paint -> house dust, differ slightly among the studies, as do the intervention strategies to
interrupt the flow of lead along these pathways. Collectively, these differences hi study
design broaden the scope of the project to cover aspects of lead exposure intervention not
possible through the study of a single neighborhood or even a single city.
2.2.3 Study Groups
Variations in the nature and form of intervention were included hi the study designs to
take advantage of the unique characteristics of the cities and their housing types. For
example, soil lead concentrations are typically high hi Boston, where it is also common to
find elevated concentrations of lead hi drinking water and hi both exterior and ulterior paint.
In the areas studied, housing is typically multi-unit with some single family units with
relatively large soil cover in accompanying yards. In the Baltimore neighborhoods, nearly
every house had lead-based paint, the houses were mixed single and multifamily, and the soil
areas were smaller, typically less than one hundred square meters. On the other hand,
houses hi Cincinnati were selected because they were thought to be relatively free of interior
lead-based paint that might obscure the contribution of soil lead to house dust lead. As it
2-8
-------
happened, these neighborhoods were mostly multifamily housing with little or no soil on the
residential parcel of land. The Cincinnati study design therefore focused on intervention at
the neighborhood scale, where the soil in parks, play areas, and other common grounds was
abated, and exterior dust on paved surfaces in the neighborhood removed.
Detailed information on study design and methods of analysis can be found hi the
appended individual reports for each city. Table 2-1 summarizes the study design
characteristics for each of the three studies and their respective neighborhood groups. The
nomenclature for these groups has been standardized for this report. With the exception of
the Cincinnati control group (CIN NT), all groups received some,form of intervention during
the study.
For the purposes of consistency, certain descriptive terms that are used differently hi
the three individual study reports are standardized here and described hi the glossary of this
document. One example is the use of the terms "study" and "project". In order to avoid
confusion, the term "study" refers to one of the three separate community studies, and the
term "project" is used hi reference to the three studies collectively. Similarly, the collective
term for "treatment group" or "control group" in this report is "study group".
The names that identify the individual study groups have been modified hi this report to
assist the reader hi remembering the type of intervention performed on each group.
Table 2-1 lists these names, with a brief description and the corresponding term in the report
of each separate study. This nomenclature identifies location of the study and the nature of
the intervention. For example, BOS SPI refers to the Boston group that received Soil, Paint,
and Interior dust intervention., A hyphen is used to indicate intervention in two different
rounds, as in CIN I-SE, where ulterior dust abatement took place about one year before soil
and exterior dust abatement. The reader should become familiar with this nomenclature for
the ten study groups in the project, as the data and results will be presented using these
designations without further explanation. One further note: The BOS PI, BOS P, and
CIN NT groups each received soil abatement at the end of the study. Because no data were
reported following this intervention, the designation "-S" was not used.
Other departures here from the terminology of the respective individual study reports
are conversion to a common system of units (metric where possible) and standard terms for
phases, stages, or rounds of the project. The term "round" refers to a distinct period of
2-9
-------
TABLE 2-1. TREATMENT GROUP NOMENCLATURE WITH
CROSS-REFERENCE TO INDIVIDUAL REPORTS
Treatment Group
Name3
Cross-Reference to
Individual Study
Report
Description of Treatment
BOSTON
BOS SPI
BOS PI-S
BOS P-S
BALTIMORE
BALSP
BALPlb
BALP2b
CINCINNATI
CIN SEI (P)
CIN NT (G,M)
Study Group
Control Group A
Control Group B
Study Area
Control Area
Study Area Not
Abated
Area A
CIN I-SE (B,D,F)C Area B
AreaC
Soil and interior dust abatement, and interior
paint stabilization at beginning of first year, no
further treatment.
Interior dust abatement and interior paint
stabilization at beginning of first year. Soil
abatement at beginning of second year.
Interior paint stabilization at beginning of first
year. Soil abatement at beginning of second
year.
Soil abatement and exterior paint stabilization
at beginning of first year, no further treatment.
Exterior paint stabilization at beginning of first
year; no further treatment.
Exterior paint stabilization at beginning of first
year, no further treatment because soil not
above cut-off level.
Soil, exterior dust, and interior dust abatement
at beginning of first year, no further treatment.
Includes only the Pendleton neighborhood.
Interior dust abatement at beginning of first
year, soil and exterior dust abatement at
beginning of second year, no further treatment.
Includes the Back St., Dandridge, and Findlay
neighborhoods.
No treatment; soil and interior dust abatement
folio whig last sampling round. Includes the
Glencoe and Mohawk neighborhoods.
aThe treatment group designation indicates the location of the study (BOS = Boston, BAL = Baltimore,
CIN = Cincinnati), the type of treatment (S = soil abatement, E = exterior dust abatement, I = interior dust
abatement, P = loose paint stabilization, NT = no treatment).
'Treated as one group hi the Baltimore report, analyzed separately in this report.
Treated as one group for many analyses in the Cincinnati report, analyzed as individual neighborhoods in this
report.
2-10
-------
time when one or more measurements were made. Other activities, such as soil abatement,
occurred between rounds. There is no consistent pattern for when abatement occurred (i.e.,
after Round 1, Round 3, etc.) for the different individual cities.
The numbers of participating children, families, and properties appear hi Table 2-2.
Because of attrition and recruitment in Baltimore and Cincinnati, these numbers do not
accurately represent the number of participants present for the duration of the study. In this
report, subsets of these participants were statistically analyzed for specific purposes and to
meet specific statistical requirements, and these subsets may not be the same subsets used by
the individual study teams hi then: statistical analysis described hi then: respective individual
city reports.
2.2.4 Project Activity Schedule
The project activity schedule, shown in Figure 2-1, illustrates the major intervention
and measurement activities of the individual studies and the sequence and duration of these
activities. The frequency and tuning of sampling relative to abatement and seasonal cycles
are important issues hi the study design. These time lines are the actual occurrence of these
events and they differ somewhat from the planned schedule. The original design focused on
sampling blood lead during the late summer, as it was known that the seasonal cycle for
blood lead reaches a peak during this period.
2.2.5 Environmental and Biological Measurements of Exposure
Figure 2-2 illustrates the generalized concept of the pathways and sources of human
exposure to lead, showing the routes of lead from the several sources in the human
environment to the four compartments (inhaled air, dust, food, drinking water) immediately
proximal to the individual child. One of these proximal sources, dust, is the primary route
of concern hi this project. Figure 2-3 expands this dust route to show both the complexity of
the many routes of dust exposure for the typical child and the mobility of dust lead along
these routes. Both of these concepts were poorly understood hi the late 1980's. The
intervention strategies used in this project were designed to interrupt the movement of lead
along one or more of these pathways.
2-11
-------
TABLE 2-2. NUMBER OF PROJECT PARTICIPANTS BY TREATMENT
GROUP AND ROUNDa
BOSTON
Middate of round
Children15
Families0
Properties'1
Treatment Group
BOS SPI
BOS PI-S
BOS P-S
BOS SPI
BOS PI-S
BOS P-S
BOS SPI
BOS PI-S
BOS P-S
Rl
(PRE)
10/17/89
52
51
47
150
43
43
39
125
34
36
30
100
R2
(POST 1)
4/9/90
52
48
46
146
43
40
38
121
34
33
29
96
R3
(POST 2)
9/12/90
52
49
46
147
43
41
38
122
34
34
29
97
R4
(Phase 2)
7/20/91
33
33
26
92
28
27
22
77
24
24
19
87
BALTIMORE
Middate of round
Childrenb
Families'3
Properties'5
CINCINNATI
Middate of round
Children15
Families6
Parcelsd
BALSP
BALP1
BALP2
BALSP
BALP1
BALP2
BALSP
BALP1
BALP2
CIN SEI (P)
CIN I-SE (B,D,F)
CIN NT (G,M)
CIN SEI (P)
CIN I-SE (B,D,F)
CIN NT (G,M)
CIN SEI (P)
CIN I-SE (B,D,F)
CIN NT (G,M)
Rl R2
10/25/88 4/1/89
88 85
73 73
7 7
168 165
63 60
50 50
6 6
119 116
55 53
45 45
6 6
106 104
Rl
(P01)
7/6/89
54
86e
_6L
201
31
58e
40
129
55
R3
(P03)
11/14/89
52
81e
52
185
30
56e
37
123
39
74e 121e
86
85
215 245
R3
2/17/90
110
80
8
198
72
52
*7
131
61
47
7
115
R4
(P05)
7/1/90
46f
92f
81f
219
31
56
35
122
39
121
85
R4
1/27/91
103
79
8
190
68
51
7
126
59
46
7
112
R6
(P07)
11/17/90
37
87
74
198
31
74
63
168
40
119
84
245 243
R5
6/7/91
99
80
7
186
65
51
6
122
53
46
6
105
R7
(P09)
6/16/91
31
77
61
169
30
60
52
142
40
121
84
245
R6
9/3/91
95
79
8
182
64
51
7
122
57
46
7
110
* Round designations (Rl, R2, etc.) are not the same as used in the Boston and Cincinnati study reports. Their round designations are
shown in parentheses. Some rounds are omitted from this table because blood lead data were not collected. Intervention, shown by the
dashed lines, occurred between Rl and R2 in the first year and R3 and R4 in the second year in Boston, R3 and R4 in Baltimore, Rl and
R3 in the first year of the Cincinnati study, and R4 and R6 in the second year. Middates are the mean blood sampling dates.
b Based on number of children sampled for blood.
e Based on number of households sampled for dust.
d Based on number of soil areas sampled.
e Dandridge was added to the Cincinnati study after the soil sampling for Rl, but before the completion of all other Rl sampling. Thus,
the number of Dandridge children and families are included in Rl for CIN I-SE, but the number of parcels are not included until R3.
f These numbers reflect additional children recruited from participating families in July, 1990. The Cincinnati Teport does not include these
children.
2-12
-------
I IMHNMMIIIIIIIIIIIIIWtmill 1111111111111II !!llllllllllllllllll"|'lilllll 1111
f f
o °
QQ
CO
o
CO
I
-?Sli8
-d o o 3 to 2 "2L
< CO CO Q X CO ^
2-13
-------
Auto I C Industrial ^
Emissions I I Emissions I
\ /
C Crustal |
I Weathering I
Surface and
Ground Water
I Solder,
Lead Glazes
Figure 2-2. Generalized concept of the sources and pathways of lead exposure in
humans.
Exposure is the amount of a substance that comes into contact with an absorbing
surface over a specific period of time. In the case of lead, the absorbing surface can be the
gastrointestinal tract or the lungs. Exposure is measured in micrograms of lead per day.
Thus, an exposure of 10 pig/day represents a total ingestion and inhalation of 10 micrograms
of lead from all sources; a fraction of this 10 micrograms would be absorbed into the body.
2.2.5.1 Blood Lead
In this project, blood lead was used as an indicator of exposure, and reductions in blood
lead concentrations were expected as a result of any combination of the interventions
described above. The units for blood are micrograms of lead per deciliter of blood Og/dL)
and they are not compatible with the normal units of exposure', micrograms of lead per day.
2-14
-------
[Atmospheric
I Particles
Soil
Exterior Paint
Dust
Interior Paint
Dust
Local
Fugitive
Dust
Exterior
Dust
Interior
Dust
^Secondary]
^—I Occupational!
I Dust J
Hand
Dust
Food
Child
Figure 2-3. Typical pathways of childhood exposure to lead in dust showing both the
complexity of the routes of exposure and the mobility of dust lead along
these routes.
The fraction of ingested lead that is actually absorbed in the gastrointestinal tract
depends in part on the bioavailability of the particular form of lead. The amount of absorbed
lead that reaches specific body tissues depends on the biokinetics of lead in the human body.
Blood tissue is in dynamic equilibrium with all other body tissues, including bone tissue,
where the lead is stored for longer periods of time.
The relationship between blood lead concentration and the onset of health effects of
lead, depends largely on the distribution of lead to the target tissues, including the red blood
cells themselves. Blood lead, then, is a convenient indicator of both exposure and potential
health risk to the child. This situation becomes important when measuring the rate at which
blood lead concentrations might decline following abatement. For a child with lead stored in
bone tissue following a long history of high lead exposure, the decline in blood lead might be
expected to be slower than for a child with low previous exposure. Even if lead-burdened
children were moved to an environment completely free of lead exposure, a significant
2-15
-------
amount of lead would still be present in the child's blood due to the slow release of lead
from the large amounts stored in bone and other body tissues.
Autopsy data show that as much as 60 to 70% of the lead in a child's body is stored in
the skeletal system, especially hi the hard (or cortical) part of long bones such as the femur
and the tibia (Barry, 1981). In adults this percentage is even larger, 90 or 95%. Lead is
retained in cortical bone for many years, and even though bone remodeling in young children
is very rapid, these large body burdens contained hi the bone constitute a significant internal
source of lead exposure for several years after exposure has stopped. The long-term stability
of blood lead levels hi a stationary exposure environment has been noted by a number of
authors (David et al., 1982; Rabinowitz, 1987).
Persistence of elevated blood lead after abatement has both biological and
environmental components. The biological component is the resorption of skeletal lead.
In adults, recent stable lead isotope studies (Smith et al. 1995, 1996; Gulson et al., 1995)
suggest that 30 to 65% of the circulating lead hi adults is due to skeletal lead, which is
consistent with other estimates. Similar studies have not been reported for children.
Although a somewhat lower percentage may be appropriate for children rather than adults, it
is clear that even in children a substantial fraction of blood lead has a skeletal origin.
The environmental component of persistence is the child's remaining exposure to other
nonremediated lead media, such as lead hi diet, drinking water, or air. This is illustrated hi
Figure 2-4, which shows a blood lead profile (for an individual, or possibly as a population
mean) before and after a hypothetical lead abatement. The steady-state blood lead
concentrations are shown as flat curves, although hi reality there may be substantial age-
dependent changes during the course of abatement even when environmental lead
concentrations remain constant. Assuming that environmental concentrations remain constant
after abatement, the child's blood lead would eventually reach a new steady-state
concentration at a much lower level. At any given time after abatement, the child's blood
lead is a mixture of three components, denoted "A", "B", and "C" in Figure 2-4.
Component A shows the relatively rapid decrease hi blood lead from elimination of
preabatement lead deposits hi blood and soft tissues. Component B shows the contribution of
preabatement skeletal lead to post-abatement blood lead, which is much slower because the
large skeletal burden in cortical bone is eliminated on a tune scale of several years. Almost
2-16
-------
Combined Preabatement
Steady-state Blood Lead
Combined Postabatement
Blood Level - A + B + C
-o
o
_o
CO
Lead in Blood
from Bone
A = Rapid Postabatement
Elimination of Stored
Lead from Soft Tissue
_g
CD
c
1
.3
•V, /
V /
\ '•'•'•'!
SIB!
B = Slow Postabatement
Elimination of Stored
Lead from Bone
p. Postabatement Steady-state
.,.
Abatement
••••• ••MM mmmmmmmmmmmamm m m m mm St*mjf «"
..... '"" V\ ^--^
C = Buildup to
Postabatement Steady-state
"'" --------- from Postabatement Exposure
Year 2
Yearl
Time
Figure 2-4. Hypothetical representation of the expected decrease in blood lead (solid
curved line) following abatement. This rate of decrease is less than might
be expected from exposure reduction alone. This is because blood also
contains lead recently released from storage in bone and soft tissue.
all of the stored lead may eventually be eliminated. However, the contribution of
preabatement deposits of lead now stored as an internal source of exposure may be
quantitatively significant compared to remaining postremediation environmental exposure
media.
The combination of persistent internal exposure and persistent baseline external
exposure amounts to a post-abatement blood lead contribution of about 50 or 60% of the
preabatement blood lead starting value at 8 to 12 months after abatement. This suggests that
under optimum conditions any environmental abatement or intervention can likely only
achieve a 40 to 50% reduction in child blood lead concentrations within a year after
abatement (see Figure 2-4).
2-17
-------
Several authors have reported differences in persistency of elevated blood lead
concentrations between smelter and non-smelter communities (Angle et al., 1984; Gallacher
et al., 1984a,b; Roels et al., 1980, 1976; Angle and Mclntire, 1979; Yankel et al., 1977).
In general, blood lead concentrations in non-smelter children tend to decrease at ages beyond
four years, whereas smelter children usually retain childhood pattern of elevated blood lead
into their teens. This difference has been attributed by Mushak (1993) to the nature and
disposal of smelter emissions. In general, the hypothesis is that urban children older than
four years should show lower blood lead concentrations than they did at age 2 to 3 years.
This hypothesis can be tested with the data from the present studies, but the hypothesis that
smelter children differ from urban children cannot.
2.2.5.2 Hand Lead
Because blood lead reflects exposure to lead from all environmental sources, a second
exposure indicator, hand lead, was used to focus directly on the immediate pathway of dust
to the child. The units of measure are micrograms of lead per pair of hands, and like blood
lead, this measure does not reflect the rate at which lead moves into the body hi units of
micrograms of lead per day. Instead, this hand dust is a measure of lead loading on the
hand. It is a measure of the "dirtiness" of the hand hi the same sense that dust loading is a
measure of the dirtiness of the floor. Hand dust loading could possibly be converted to
micrograms of lead per day if there were a measure of the area of the hand mouthed by the
child, the frequency of hand to mouth activity, and the frequency of hand washing during
each day.
2.2.5.3 House Dust
House dust is a mixture of lead from many sources, including soil, street dust, interior
paint, and occupational dusts carried home by family workers. The units of measurement are
jig Pb/g (lead concentration), /ig Pb/m2 (lead loading), and mg dust/m2 (dust loading).
When expressed as micrograms of lead per gram, the measurement can be converted to an
exposure measurement by assuming a specific amount of dust ingested per day, usually about
100 mg/day for preschool children. Exposure to household dust then becomes micrograms
per day:
2-18
-------
Pb Concentration x Ingestion = Exposure
gdust =
(2-1)
gdust
day day
In a similar manner, exposure to food, drinking water, and inhaled air can be expressed
as /ig/day, and in 1990 these three sources normally accounted for about 5, 1, and
0.1 /Ag Pb/day respectively (U.S. Environmental Protection Agency, 1994). If the lead
concentration hi household dust is 200 /ig/g and dust ingestion is 0.1 g/day, the exposure is
20 jwg/day or much more than the other sources combined.
By a different calculation, childhood lead exposure may be expressed as a function of
dust lead loading. In mis case, the ingestion parameter is in units of m2/day:
Pb Loading x Ingestion = Exposure
x
m _ tig
(2-2)
day day
The ingestion parameter estimates the effective contact area for the child's hands (assuming
all dust is ingested by hand-to-mouth activity). Literature reports of childhood lead exposure
based on contact area are not known.
2.2.6 Intervention Strategies
Intervention is defined here as the interruption of the flow of lead along an exposure
pathway. Soil abatement is one form of intervention. If done correctly, this abatement
should establish an effective and persistent barrier to the movement of lead through the
child's exposure pathways. Other forms of intervention used in this project were exterior
dust abatement, interior dust abatement, and paint stabilization. Because dust is a very
mobile constituent of the human environment, exterior and ulterior dust abatement would not
be expected to form a permanent barrier to lead unless other sources of lead, such as soil,
were also abated. Likewise, the form of paint stabilization used hi Boston and Baltimore,
where chipping and peeling paint was removed and the walls repainted, was not intended to
be permanent lead-based paint abatement.
2-19
-------
The strategy for soil abatement was to remove all soil with concentrations above a
specific level (500 jig/g for Baltimore and Cincinnati, 1,000 jttg/g for Boston), and replace
this soil with clean soil in the range of 25 to 100 /tg/g lead concentration. This method
(excavation, removal, and replacement) was used in all three studies. In some cases, repair
and maintenance of ground cover was used where the soil concentrations did not warrant
excavation, removal, and replacement. To further interrupt the flow of lead along the
exposure pathways, entire neighborhoods hi Cincinnati were cleaned of exterior dust using
street cleaning vacuum equipment and hand tools.
Interior house dust is believed to be a major direct lead exposure pathway for children.
Because household dust typically contains a mixture of lead from several sources (e.g., soil,
interior/exterior paint, air, etc.), abating house dust temporarily separates such sources from
the child's environment. Their recontamination of house dust and consequent impact on the
child's lead exposure can be evaluated by comprehensive measurements of the household dust
that include changes hi lead concentration, lead loading, and dust loading. Understanding the
expected impact of abatement on these three parameters is critical to interpreting the
observed changes in blood lead concentrations. Following dust abatement, there should be
an immediate decrease in the dust loading, with no change hi the lead concentration for those
groups that did not receive soil, exterior dust, or paint intervention. The rate at which this
dust loading returns to preabatement levels reflects the rate of movement of dust from other
sources into the home, the frequency of cleaning, and the "cleanability" of the home. (Many
inner city homes have surfaces that are cracked, pitted, or hi disrepair and are difficult to
clean effectively.)
The effectiveness of both paint stabilization and soil and dust abatement can be
observed by changes hi the lead concentrations of house dust. In the presence of lead-based
paint, the concentration of lead hi house dust is expected to be greater than 1,500 to
2,000 /tg/g, whereas without the influence of lead-based paint, the house dust is expected to
be comparable to external dust and soil (U.S. Environmental Protection Agency, 1986).
House dust is a mixture of dusts from many sources within and outside the home.
In the absence of lead-based paint inside the home, it would seem reasonable to assume that
most of the lead hi household dust comes from soil and other sources external to the home.
2-20
-------
Therefore, to enhance the impact of soil abatement, interior dust abatement was carried out
for some treatment groups in Boston and Cincinnati.
Many of the Boston and Baltimore households selected for the project had chipping and
peeling paint, both interior and exterior. In order to reduce the impact of lead-based paint,
the walls and other surfaces were scraped and smoothed, then repainted. It is important to
note that no attempt was made to remove all lead-based paint, nor to isolate intact paint from
the child. Paint stabilization was used on interior surfaces hi Boston and on exterior surfaces
in Baltimore. Paint stabilization was not used in Cincinnati because most of the lead-based
paint was believed to have been removed from most of these homes hi the early 1970s.
2.3 EXTERNAL FACTORS THAT COULD INFLUENCE PROJECT
RESULTS AND INTERPRETATION
The Scientific Coordinating Panel recognized that several extraneous factors might
influence the outcome of the project and that these factors were generally beyond the control
of the investigators. Among these are seasonal cycles and tune trends of childhood blood
lead concentrations, unexplained or unexpected sources of lead hi the children's homes or
neighborhoods, changes in public perception and avoidance of lead exposure hazards, and
movement of lead in soil either down the soil column or laterally with surface runoff or as
fugitive dust.
2.3.1 Cycles and Trends in Environmental Lead Concentrations
Figure 2-5 illustrates a pattern of childhood blood lead concentrations for Chicago
during the 1970s, shov/ing a seasonal cycle and a downward trend throughout the decade.
The National Health Assessment and Nutrition Examination Survey II (NHANES II) data for
the entire country and all age groups reported a similar seasonal cycle and downward trend
during the last half of that decade (Annest et al., 1983). (Seasonal patterns from the
NHANES III data of 1988 through 1991 are not yet available.)
Investigators have known about this seasonal pattern for some time (Figure 2-6). Most
epidemiological studies are planned so that measurements can be taken at the peak of this
cycle, generally during the late summer. Studies of large numbers of children show a
2-21
-------
50
40
I
2 so
m
§
o
-------
5.0 -
0.5 -I
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Sampling Month
Figure 2-6. Estimated seasonal variation based on residual blood-lead levels in Boston
children after controlling for age and date of birth effects. (Bars represent
95% confidence bounds for blood-lead residuals.)
Source: United States Environmental Protection Agency (1995).
Although this project was designed to maximize the measurements of blood lead during
the late summer for each of the three studies, measurements were made during other tunes of
the year in order to observe changes immediately after abatement. For most statistical
analyses hi this report, comparisons were made from measurements taken approximately
twelve months apart in order to minimize the impact of the seasonal effect. A more detailed
description of this treatment appears in Chapter 5.
Two other patterns, long-term time trends and early childhood patterns dependent on
age, are applicable to this project. Little is known about age related patterns, but one study
hi Cincinnati, prior to the project, showed a pattern of blood lead changes during early
childhood growth patterns (Figure 2-7).
Long-term downward trends were documented for child blood lead concentrations
during the 1970s and 1980s and have been attributed to decreasing concentrations of lead in
2-23
-------
10
•65 8
I 6
o
nr
•o
1 4
DQ
PbH
PbB
u 0 1 2 3 4 5 6
Age (years)
Figure 2-7. Predicted differences in blood lead (PbB) and hand lead (PbH) during early
childhood, based on empirical data. The peak for blood lead at age 2 may
be due to activity patterns related to dust ingestion for toddlers and young
children. The steady increase in hand lead could be due to the increase in
hand size as well as activity patterns favoring play outside the home.
Source: Bornschein et al. (1985).
food and air. The QA/QC measures reported in detail in Chapter 4 rule out the possibility of
this trend being caused by a measurement artifact such as analytical drift.
2.3.2 Unexplained and Unexpected Sources of Lead
Occasionally, measurements of environmental lead are higher than expected and
difficult to explain. Atmospheric deposition can be a reasonable explanation, because this
route can change much more abruptly than soil, dust, food or drinking water. This section
discusses the possibility that the observed fluctuation in street dust and house dust can be
attributed to changes in air concentration alone. Because this project began after the national
phasedown of lead in gasoline, the air concentrations of lead in these cities had decreased to
2-24
-------
about 0.1 /tg/m3 by the start the project.2 The following is a theoretical calculation of the
amount of lead that could be transferred to soil or dust at this concentration and from this
source alone.
Atmospheric deposition during the project was assumed to be typical for air
concentrations that averaged 0.1 jig/m3 (1.0 x 10"7 /^g/cm3). At a deposition rate of
0.2 cm/s, this would accumulate 0.6 jtg/cm2-year at the soil surface. Assuming that this lead
would be retained hi the upper 1 cm of soil surface (therefore 1 cm2 of soil surface equals
1 cm3 of soil), then the annual increment would be 0.6 /ig/cm3. Because 1 cm3 of soil
weighs about 2 g, the ajtinual incremental increase in lead concentration would be
0.3 fjig Pb/g soil, an insignificant annual contribution for soils that average several hundred
micrograms per gram. The calculation for annual deposition to a surface is
1 x 10
-7
cm
_ X 0.2 — x 3.15 x 107 _fL_
3 s year
= 0.6
Pb
cm year
(2-3)
For the accumulation of dust on hard surfaces, however, the same calculation indicates
a potentially greater influence of atmospheric lead. Converting to units of lead loading, the
0.6 pig/cm2-year becomes 6,000 /xg/m2-year, or 16 jtg/m2-day. Therefore, 0.1 ptg/m3 in air
concentration could account for a change of 16 jug Pb/m2 per day in the dust lead loading to
a surface. An accumulation of 160 pig/m2 over 10 days is hi, the range of the observed
changes in surface dust loading in this project.
2.3.3 Movement of Lead in Soil and Dust
There are several reasons why localized soil lead fluctuations might occur. Changes hi
soil lead concentration independent of intervention that might increase lead concentration are:
atmospheric deposition (relatively minor as discussed above), exterior paint chipping and
chalking, and human activity such as household waste dumping (motor oil, etc). Soil lead
concentrations might decrease if lead leaches downward into the lower soil horizon, or if
surface soil shifts by dust reentrainment. The downward leaching of lead through the soil
2 The 1989 maximum quarterly average air lead concentration for the metropolitan statistical areas of Boston,
Baltimore, and Cincinnati were 0.08, 0.11, and 0.11 jig/m3, respectively (U.S. Environmental Protection
Agency, 1991a).
2-25
-------
profile mass occurs at a very slow rate, approximately a few millimeters per decade (Grant
et al., 1990). The reentraiiunent of dust at the soil surface is usually hi equilibrium with the
local environment, such that inputs would equal outputs by this pathway. This would not be
the case if there is flaking or peeling lead-based paint within the neighborhood or an
industrial source of fugitive dust in the vicinity of the neighborhood. A limited effort was
made to monitor and control the impact of lead-based paint on soil concentrations.
In Baltimore, buildings with exterior lead-based paint were stabilized by removal of the
chipping and peeling paint, done in a manner to avoid contaminating the soil. In Boston,
homes were selected with less then 30% exterior chipping and peeling paint, by area.
In Cincinnati, neighborhoods with mostly rehabilitated houses were selected. There were no
attempts hi any of the studies to control the introduction of lead to the soil by human activity
such as household waste dumping.
Lead hi household dust is a mixture of dust brought into the house from outside and
dust generated from within the home. Studies have shown that as much as 85% of the mass
of dust comes from outside the home and much of this is apparently brought in on the feet of
children and pets (Roberts et al., 1991). Household dust lead concentrations are usually
similar to the soil concentration hi the immediate vicinity of the house, unless there are
internal sources of lead, such as lead-based paint. Thus, changes in soil concentrations are
likely to be reflected by changes hi household dust concentrations within a few days and
probably reach equilibrium within a few months, depending on the relative contribution from
soil and other sources, the frequency and efficiency of house cleaning, and the cleanability of
the house.
2.3.4 Factors That Limit Interpretation of the Project Results
In the following chapters, this report discusses several issues that identify possible
limitations of the studies. This detailed assessment: (1) examines measurement methods used
and related QA/QC data to ascertain that adequate measures were taken to produce data of
good quality that can be compared across the three studies; (2) examines the study designs to
determine if the individual study groups are comparable within each study and if comparisons
are possible across the three studies; and (3) performs rigorous statistical analyses that
2-26
-------
attempt to quantify differences between study groups and identify specific exposure factors
that may be responsible for the differences.
With respect to the QA/QC data, it should be noted that there are no estimates of
sampling reproducibility for any of the environmental or biological measurements. This
would have required collecting duplicate samples for a specified percentage of the samples.
In retrospect, the following observations are worth noting:
1. Duplicate soil samples would not have been informative unless the entire soil parcel
was sampled in duplicate. In this report, the reproducible number is the arithmetic
mean of all soil samples from the parcel;
2. Duplicate sampling of house dust would have identified reproducibility of lead
concentration, but probably not lead loading, which changes on a daily basis.
Duplicate sampling of house dust may also have impacted the child's environment if
a substantial amount of the targeted play areas were sampled.
Nevertheless, this report recognizes the limitations of statistical analysis due to the
absence of an estimate of sampling error.
There are several exposure-related factors other than those measured by environmental
sampling that must be taken into account during the statistical analyses. Among these are
seasonal patterns in weather (especially rainfall as it affects dust loading and mobility),
activity patterns (which affect indoor/outdoor play patterns), and possible physiological
growth cycles (which affect remobilization of lead from bone tissue). Age of the child may
also impact exposure by differences in activity patterns, body size, and parental supervision.
For the most part, this report is only able to assume that all groups within a study were
impacted equally by these and other confounding factors during the study.
2-27
-------
-------
3. METHODS INTERCOMPARISON AND QUALITY
ASSURANCE/QUALITY CONTROL
Specific details on measurement methodology employed in each study can be found in
the appended individual city reports. This chapter describes the initial evaluation of several
methods for soil, dust, hand wipe, and blood sampling and analysis that were considered by
the Scientific Coordinating Committee, and the basis for selection of these methods by the
participating research teams.
Soil sampling procedures were defined based on agreement that five 2-cm soil cores
would be taken according to a prescribed pattern about a randomly selected point, and that a
prescribed number of these points would be selected based on the size and shape of the plot
of soil. These procedures are described hi the individual reports, and no further assessment
is made here of the representativeness of this sampling procedure.
Interior dust sampling methods were determined based on the desirability of obtaining
dust loading information. This required that a dry sample be taken (as opposed to a wet
wipe) hi order to determine the mass of dust collected as a function of area (dust load).
Although the sampling devices differed, the basic protocol called for a vacuum pump that
collected the dust sample on a filter pad at a prescribed flow rate and using a prescribed
pattern of moving the pump nozzle over the sample area. No further attempt was made to
calibrate the collection devices between the individual studies.
Hand wipe samples were taken according to procedures developed by the Cincinnati
group hi previous studies. Field blanks and lot blanks were determined by each group.
There were some differences hi the tuning of the hand wipe sample (home visit versus clinic
visit) as reported by the individual study teams.
Blood samples were taken according to methods prescribed by CDC in their blood lead
certification program. The analysis of blood for health indicators (FEP, TIBC, etc.) other
than lead differed among the three groups. Only the blood lead concentration data were used
hi this integrated assessment.
The procedures and results of interlaboratory comparisons of analytical methodology
and the results of the QA/QC plan for the individual studies are described hi the following
3-1
-------
sections. These procedures and their results were reviewed and evaluated throughout the
project at the scheduled workshops and during monthly teleconference calls.
The research team for each study prepared a sampling and analysis plan that included
rigorous QA/QC objectives. These plans included protocols that: defined sampling schemes
designed to characterize the expected exposure to soil for children; described how to collect,
transfer, and store samples without contamination; and described how to analyze samples
with the maximum degree of accuracy and precision. Sampling protocols for soil,
handwipes, and blood lead were nearly identical. Dust sampling protocols differed with
regard to the vacuum device used, location of sample within the residence, and procedures
for pooling samples prior to analyses. These differences may, hi retrospect, have affected
the comparability of both the dust load and dust lead concentration data. During the course
of the project, several intercalibration exercises were performed to ensure that the analytical
results for measurements of soil, dust, handwipes, and blood would be accurate and that the
data would be as comparable as possible.
3.1 INTERCOMPARISON OF LABORATORY METHODS FOR SOIL
AND DUST MEASUREMENTS
The objective of the laboratory intercomparison and QA/QC program was to ensure that
the three studies could achieve high-quality analyses of soil and dust samples, and that each
of the three laboratories would be expected to get similar results when analyzing the same
soil sample. The participating cities recognized the need for standardizing the sampling and
analytical protocols so that data from each study could be compared. This standardization
was accomplished for soil and dust by measuring the analytical difference between each of
the three labs. Common standards were prepared and a program for assuring data quality
was put into place. A three step program was agreed to that involved: (1) a round robin
calibration exercise of soil samples to measure differences between laboratories and
differences between analytical methods and instrumentation; (2) a double blind audit system
for soil and dust to monitor the performance of each laboratory during the project; and
(3) a second round robin calibration exercise to determine the arithmetic correction factor
that would normalize dust and soil data to a common project basis. This program ensured
3-2
-------
that analyses performed by each of the three participating laboratories would be internally
accurate and externally consistent with similar analyses by other research laboratories.
Intercalibration Exercise I was conducted prior to the beginning of each study using soil
and dust samples collected from representative neighborhoods in each city. Intercalibration
Exercise II was conducted near the end of the sampling phase of the project using aliquots of
soil and dust samples collected at the beginning of the sampling phase, some of which were
used for QA/QC monitoring during the project. In each calibration exercise, two additional
laboratories were invited to participate in order to determine some measure of comparability
with other studies reported in the scientific literature. All laboratories reported their results
independently. In the time period between these two calibration exercises, the effectiveness
of the individual QA/QC programs was also monitored by inserting double blind audit
samples into the sample stream of each study to measure the consistency of analytical
precision throughout the study and to monitor any analytical drift.
3.1.1 Round Robin Intercalibration Exercise I
At the beginning of this project, the methods proposed by each study for soil and dust
analysis were reviewed by the Scientific Coordinating Panel. The preferred method, hot
nitric acid digestion followed by atomic absorption spectroscopy (AAS), was time consuming
and expensive. The number of samples was expected to exceed 75,000 per study, so more
rapid and less expensive methods were evaluated. Laboratory scale X-ray fluorescence
(XRF) spectroscopy and inductively coupled plasma (ICP) emission spectroscopy were
proposed, and a cold nitric acid extraction method for AAS was also considered.
In May 1988, prior to the beginning of each study, each of the three laboratories
collected ten soil samples from areas similar to those that would be included in their study.
One of the samples from Cincinnati was a street dust sample of very high lead concentration.
The other 29 samples were selected from soils with lead concentrations expected to range
from 250 to 8,000 /ig/g. The samples were dried and sieved according to the study
protocols. Approximately 200 g of each sample were sent to the other two laboratories and
to an outside laboratory at Georgia Tech Research Institute (GTRI). Table 3-1 shows the
instrumentation and method of analysis used by each laboratory. In making these analyses,
each laboratory used its own internal standards for instrumental calibration and shared a
3-3
-------
TABLE 3-1. WET CHEMISTRY AND INSTRUMENTAL METHODS USED FOR
THE FIRST INTERCALIBRATION STUDY
Participating Laboratories
Method3 Boston
Hot HNO3/AAS '
Cold HN03/AAS
Hot HNCyiCP
XRF X
Baltimore
X
X
Cincinnati
X
X
GTRIb USDAC
X
X
aHNO3 = Nitric acid; AAS = Atomic absorption spectroscopy; ICP = Inductively coupled plasma emission
spectroscopy; XRF = X-ray fluorescence.
bGTRI = Georgia Tech Research Institute.
°USDA « U.S. Department of Agriculture.
common set of five standards provided by Dr. Rufus Chaney at the U.S. Department of
Agriculture. The intercalibration exercise successfully established a baseline for cross study
comparison of soil and dust results.
In summary, the test conditions were that each laboratory would be provided with
instructions for preparing the samples (drying, sieving, and chemical extraction) but would
use their own internal standards and instrumental settings. They would have access to a set
of external standards (from U.S. Department of Agriculture) with known values from which
they could make corrections if necessary.
Each of the three study laboratories sent aliquots of 10 samples to the other two
participating laboratories and to two external laboratories. The samples were subdivided by
sieving during preparation to a "total" and "fine" fraction. Thus there were 30 samples,
each with two size fractions analyzed by each of five laboratories using either one or two
analytical methods, as indicated in Table 3-1. The results of the analyses appear in «'.
Table 3-2.
The cold nitric acid extraction method was found to be essentially equivalent to the hot
nitric acid extraction method for soils with lead concentrations up to 8,000 /*g/g (Figure 3-1)
for the samples analyzed in this study. The AAS method used by Cincinnati and Baltimore
was also equivalent (Figure 3-2), showing a high degree of comparability between these two
laboratories under these test conditions.
3-4
-------
TABLE 3-2. ANALYTICAL RESULTS OF THE FIRST
INTERCALDiRATION STUDY: LEAD CONCENTRATION G*g/g)
IN THE TOTAL AND FINE FRACTIONS OF 10 SOILS FROM EACH STUDY
Sample
Fraction6
IT
2T
3T
4T
5T
6T
7T
8T
9T
10T
11T
12T
1ST
14T
1ST
16T
17T
18T
19T
20T
21T
22T
23T
24T
26T
27T
28T
29T
30T
IF
2F
3F
4F
5F
6F
7F
8F
9F
10F
11F
12F
13F
14F
15F
16F
17F
18F
19F
20F
21F
22F
Boston
XRF
1,200
1,750
400
550
1,100
1,450
1,000
500
550
1,450
250
800
100
700
550
220
220
75
50
4,800
500
950
1,700
2,400
2,800
3,800
5,200
4,000
6,500
1,500
2,650
500
1,600
1,700
2,400
1,200
600
650
2,200
220
1,800
100
800
620
300
100
100
50
5,100
550
1,100
Baltimore
Hot HNO3
AAS
1,418
2,893
492
619
1,058
2,323
1,359
683
608
1,649
484
1,069
2,200
1,754
264
126
106
9
15,792
496
850
1,559
2,260
2,484
3,846
5,092
5,097
7,995
1,545
3,540
625
1,814
1,793
3,137
1,344
723
686
2,398
356
2,707
96
100
796
3,200
118
142
7,866
606
1,118
Hot HNO3
ICP
1,324
2,544
389
462
882
1,955
1,098
535
485
1,330
365
878
53
1,701
1,410
200
62
48
7
12,030
372
698
1,298
1,880
2,119
3,440
4,667
4,510
6,560
1,421
2,921
507
1,554
1,475
2,387
1,105
598
558
1,946
244
2,220
68
779
616
236
73
85
10
6,000
506
916
Cincinnati
Hot HNO3
AAS
1,552
2,868
387
423
964
1,876
1,383
491
455
1,679
316
1,850
63
2,068
747
253
59
74
2
14,593
387
837
1,567
2,284
2,754
4,337
5,454
5,586
8,467
1,560
3,335
478
1,678
1,689
2,835
1,306
595
593
1,808
267
2,683
68
926
635
237
73
91
. 3
8,109
480
1,069
Cold HNO3
AAS
1,215
2,211
466
415
854
1,722
990
725
417
1,228
348
1,103
45
1,713
785
295
58
61
3
8,147
378
739
1,368
2,003
2,401
3,835
4,747
4,700
7,502
1,404
3,127
508
1,595
1,971
2,009
1,184
298
601
1,116
277
2,683
64
818
642
239
66
87
2
7,432
467
944
GTRP
XRF
1,174
1,912
400
500
980
1,524
651
400
261
1,660
180
900
100
652
505
187
30
100
20
4,817
383
717
1,390
2,021
2,331
3,500
4,460
3,280
4,704
1,223
2,263
440
1234
1,290
2,134 .
815
490
375
1,980
180
1,680
100
693
600
236
100
100
30
4,780
505
980
USDA"
Cold HNO3
AAS
1,338
2,695
417
464
988
1,808
1,473
726
605
1,764
304
1,944
73
1,710
825
286
83
111
13
14,733
1,120
1,761
2,561
2,472
4,983
3,184
6,473
10,042
1,569
3,273
515
1,824
1,683
2,682
1,297
672
630
280
2,610
89
895
664
242
80
92
20
8,451
470
904
3-5
-------
TABLE 3-2 (cont'd). ANALYTICAL RESULTS OF THE FIRST
EMTERCALIBRATION STUDY: LEAD CONCENTRATION (jig/g)
IN THE TOTAL AND FINE FRACTIONS OF 10 SOILS FROM EACH STUDY
Sample
Fraction0
23F
24F
25F
26F
27F
28F
29F
30F
Boston
,
XRF
1,700
2,200
2,200
2,800
4,000
3,100
4,500
8,000
Baltimore
Hot HN03
AAS
1,679
2,331
2,372
2,899
4,833
3,087
5,896
8,555
Hot HN03
ICP
1,424
2,014
2,000
2,402
3,969
2,616
4,717
7,443
Cincinnati
GTRJP
Hot HNO3 Cold HNO3
AAS
1,710
2,328
1,665
2,946
4,531
3,073
5,606
8,679
AAS
1,431
2,010
2,089
2,568
4,130
2,720
4,869
7,789
XRF
1,320
1,940
2,005
2,249
3,739
2,445
4,240
6,015
USDAb
Cold HN03
AAS
1,640
2,492
3,156
4,979
6,194
6,680
9,754
*GTRI = Georgia Tech Research Institute.
bUSDA = U.S. Department of Agriculture.
°T — Total fraction, F = Fine fraction.
O
10 15
Cincinnati Hot HNO3 (ng/g)
Thousands
20
Figure 3-1. Comparison of uncorrected data for two wet chemistry methods of soil
analysis showing the comparability of hot and cold nitric acid for the
Cincinnati laboratory. The straight line indicates a slope of 1.
3-6
-------
I
£
o
CQ
5 10 15
Cincinnati AAS Hot HNO3 (jig/g)
Thousands
20
Figure 3-2. Comparison of unconnected data for atomic absorption spectroscopic
analysis by two laboratories (Baltimore and Cincinnati) using the hot nitric
acid method of soil analysis. The straight line indicates a slope of 1.
The interlaboratory comparison of XRF between the Boston and GTRI Laboratories
showed the method was acceptable, although not fully linear above 5,000 /xg/g. There were
no soil standards available above 2,000 /*g/g, so the analysts had some difficulty calibrating
their XRF instruments above this level. The data shown hi Figure 3-3 suggest a systematic
difference between the two laboratories that could be corrected with a more uniform
calibration. Both interlaboratory (Cincinnati and Baltimore in Figure 3-4) and intralaboratory
(Baltimore in Figure 3-5) comparisons of AAS versus ICP demonstrated equivalency between
these two instrumental methods. These comparisons showed that there is likewise a
systematic difference that can be statistically corrected.
Finally, the interlaboratory comparison of XRF versus AAS (Boston and Cincinnati hi
Figure 3-6, and Boston and Baltimore in Figure 3-7) led to the conclusion that, if suitable
soil standards at higher concentrations could be made available, XRF would be an acceptable
alternative method to AAS for soil analysis.
3-7
-------
•I
123456789
Boston XRF (ng/g)
Thousands
figure 3-3. Intel-laboratory comparison of uncorrected data for the X-ray fluorescence
method of soil analysis showing the comparability of the Boston and
Georgia Institute of Technology laboratories. The straight line indicates a
slope of 1.
10
Cincinnati AAS
Thousands
15
20
Figure 3-4. Interlaboratory comparison of uncorrected data for soil analysis showing
the comparability of inductively coupled plasma emission spectroscopy and
atomic absorption spectroscopy for the Baltimore and Cincinnati
laboratories. The straight line indicates a slope of 1.
3-8
-------
5 10
Baltimore AAS Hot HNO3
Thousands
15
20
Figure 3-5. Comparison of uncorrected data for soil analysis showing the comparability
of inductively coupled plasma emission spectroscopy and atomic absorption
spectroscopy within the Baltimore laboratory. The straight line indicates a
slope of 1.
10
Cincinnati Hot HNOa
Thousands
AAS(ng/g)
Figure 3-6. Interlaboratory comparison of uncorrected data for soil analysis showing
the comparability of X-ray fluorescence and atomic absorption spectroscopy
for the Cincinnati and Boston laboratories. The straight line indicates a
slope of 1.
3-9 .
-------
5 10 15
Baltimore Hot HNO3 AAS (jjig/g)
Thousands
20
Figure 3-7. Intel-laboratory comparison of uncorrected data for soil analysis showing
the comparability of X-ray fluorescence and atomic absorption spectroscopy
for the Baltimore and Boston laboratories. The straight line indicates a
slope of 1.
The Scientific Coordinating Panel recommended the use of XKF for soil analysis on the
condition that a suitable set of common standards could be prepared for a broader
concentration range and that a rigorous audit program be established to ensure continued
analytical accuracy. This recommendation was based on the interlaboratory comparison
study, the awareness that chemical extraction of a large number of soil samples presented a
costly burden on the project both hi terms of time and expense, and the value of
nondestructive analysis in preserving the samples for reanalysis. The Round Robin I
calibration exercise also revealed the need for a broader scale calibration exercise to
determine the arithmetic correction factor for converting the data to a common basis.
Two groups, Boston and Baltimore, also elected to use XRF for interior dust analysis,
whereas Cincinnati opted for hot nitric extraction with AAS for interior dust and XRF for
exterior dust. During the study, Baltimore recognized problems with analyzing dust by XRF
when the sample size was small (less than 100 mg). They reanalyzed the dust samples by
3-10
-------
AAS and reported both measurements. In Boston, this problem was solved by compositing
the floor dust samples for XKF analysis, reporting one floor dust sample per housing unit.
3.1.2 Quality Assurance/Quality Control Standards and Audits
After the first intercalibration exercise, a set of nine soil and six dust interlaboratory
standards was prepared to monitor the QA/QC performance of soil and dust analysis
throughout the project. These were prepared from three soil and two dust samples from each
of the three studies, collected in bulk (about 30 kg), in a range thought to be high, medium,
and low for that area. Seven of the soil samples and five of the dust samples were dried,
sieved, and analyzed at the EPA Environmental Monitoring Systems Laboratory hi
Las Vegas, NV (EMSL/LV). Following homogenization, approximately 50 aliquots of each
of the samples were analyzed by laboratory scale XRF at the EMSL/LV laboratory to
estimate the acceptable range for a single laboratory. Three of the 15 were distributed to the
participating cities for use as interlaboratory reference standards. The remaining 12 were
used as double blind external audits for soil and dust.
Each city appointed a QA/QC officer who was not directly involved with the analysis
of the soil samples, but who had access to the soil sample preparation stream on a daily
basis. This person mailed prelabeled soil sample containers with typical sample numbers to
the EMSL/LV laboratory. Approximately 20 g samples from one of the six external audit
materials typical for each city were placed in the sample containers fully disguised as field
soil samples and returned to the QA/QC officer in lots of 20 to 30. The identification
numbers and soil concentration values were monitored by the project QA/QC officer at
ECAO/RTP. Each city's QA/QC officer inserted the double blind samples into the sample
stream on a random basis at a frequency that would ensure about four QA/QC samples per
analytical day. These were occasionally placed as duplicates hi the same batch to provide
information about replication within the batch.
The preliminary acceptance range for the double blind audit samples was established
using the original 50 XRF analyses by the Las Vegas laboratory discussed above. As the
analytical results were reviewed by the study QA/QC officer, the audit sample results were
sent to the project QA/QC officer at ECAO/RTP. If the audit samples were outside the
acceptable range, the study QA/QC officer was informed and could recommend either
3-11
-------
reanalysis or flagging the data for that entire batch. The initial acceptable range for the six
audit samples was based on analyses by a single laboratory (EMSL/LV). This range was
adjusted for interlaboratory variation after the Intercalibration Exercise II. Final decisions on
the disposition of the audit sample anomalies were deferred until the completion of the
second intercalibration exercise near the end of the study.
The results of the double-blind audit program are given in Table 3-3 based on the final
biweight distributions shown in Table 3-4. The preliminary biweight distributions, shown
also hi Table 3-4, contained no measure of interlaboratory variability because the preliminary
analyses were performed by only the EMSL-LV laboratory. These values could only be used
in a preliminary assessment of the audit program to identify and flag batches of soil samples
that might need to be reanalyzed pending the determination of the final biweight
distributions.
The laboratories were found to be systematically low or high. This was not of major
concern, as these discrepancies could be resolved by a more detailed intercalibration exercise
and statistical correction at the end of the study. The Cincinnati group elected to make a
midcourse change hi instrumental parameters that reduced this difference, and they described
this procedure in their report. Occasionally, the measured audit sample was sporadically
high or low, hi which case the laboratory investigated the problem and resolved it. Most of
these discrepancies occurred for dust samples where the sample size for XRF analysis was
below 200 mg. The Boston group found, but did not report hi detail, that a calibration curve
for XRF analysis using standards that were also less than 200 mg would provide a suitable
correction to the original data. They elected, however, to composite their floor dust
samples.
3.1.3 Round Robin Intercalibration Exercise II
Near the end of the project, aliquots of the nine soil and six dust audit samples used
during the project were redistributed to the three study laboratories for single blind analysis.
The analyst was aware that the samples were audit samples, but did not know their
concentrations. These measurements were the basis for establishing the final range of
acceptability for the audit samples and for adjusting the soil and dust measurements hi each
study to values common to the project.
3-12
-------
TABLE 3-3. SOIL AND DUST AUDIT PROGRAM RESULTS
Study/Audit Sample
BOSTON DUST (XRF)
BAL03
CIN 01
CIN02
BOSTON SOIL (XRF)
BOSM
BALH
CIN L
CINH
BALTIMORE DUST (XRF)
BAL 02
CIN 01
BOS 01
Number of
Samples
N/Ab
N/A
N/A
N/A
N/A
N/A
N/A
8
10
10
Mean
G*g/g)
1,232
2,671
331
6,786
1,044
399
14,074
218
3,280
14,444
Range
0*g/g)
980-1,441
2,075-3,228
115-461
6,015-7,549
747-1,244
207-570
11,407-16,592
159-281
800-3,660
14,080-14,920
Percent Within
Final Biweight
Distribution4
92
100
65
100
73
61
50
100
90
N/A
BALTIMORE SOIL (XRF)
BOSM
BALH
CINL
CINH
CINCINNATI DUST (AAS)
BAL 03
BOS 01
CIN 01
CIN 02
CINCINNATI SOIL (XRF)
BOSM
BALH
CINL
CINH
15
15
15
15
34
35
38
26
32
49
130
31
5,046
838
286
11,290
1,727
24,104
2,683
259
5,580
885
263
12,304
4,800-5,200
433-916
266-307
10,100-12,500
1,322-2,687
20,266-27,962
2,070-3,163
200-393
4,759-6,107
822-1,012
244-310
9,838-13,632
100
60
100
53
N/A
N/A
100
N/A
100
N/A
100
N/A
"These percentages include audit samples for which analyses were outside the biweight distribution range and
for which the action required by the QA/QC plan, such as reanalysis of the entire batch, was implemented.
bN/A = Not available.
3-13
-------
TABLE 3-4. PRELIMINARY AND FINAL BIWEIGHT DISTRIBUTIONS FOR SOIL
AND DUST AUDIT PROGRAM
Sample
Type
Dust
Dust
Dust
Dust
Dust
Soil
Soil
SoU
Soil
Soil
Soil
Soil
Soil
Soil
Soil
Soil
Soil
Soil
Audit
Sample
BAL01
BAL02
BAL03
CIN01
CIN02
BOSL
BOSM
BOSH
BALL
BALH
CINL
CINH
REF5
REF6
REF7
REF8
REF9
REF10
Preliminary Values (/tg/g)
Mean
78
331
1,480
2,851
252
3,131
6,090
14,483
639
923
303
13,585
Low
58
• 288
1,346
2,660
216
2,858
5,748
13,071
555
850
284
12,872
High
99
374
1,613
3,042
288
3,405
6,431
15,895
724
997
322
14,297
Final Values
Mean
84
309
1,438
2,617
233
3,101
6,219
13,369
626
1,017
315
12,729
413
936
1,042
2,354
3,913
735
Low
4
138
1,091
1,422
93
2,283
4,742
11,980
468
847
204
11,361
258
738
758
1,950
2,943
615
0*g/g)
High
163
480
1,786
3,812
372
3,919
7,696
14,754
783
1,187
426
14,096
568
1,134
1,326
2,759
4,888
854
3.1.4 Biweight Distribution and Final Interlaboratory Calibration
The nine soil and five dust samples that were used for external standards and audit
samples were reanalyzed in a more detailed round robin exercise near the end of the project.
The purpose of this exercise was to determine the correction factor for statistically converting
the soil and dust data from each study to a common basis and to revise the biweight
distribution values for the audit samples to reflect the multilaboratory variance and systematic
differences between laboratories. Additional analyses by AAS were performed by Baltimore
3-14
-------
and Cincinnati for soil and dust, even though only dust was analyzed by AAS during the
study. Boston and Las Vegas analyzed the samples by ICP for the purposes of obtaining a
broader perspective on the application of this method. The data from this exercise are shown
in Table 3-5 and are the basis for determining the consensus values and correction factors
that appear in Table 3-6.
A data evaluation subcommittee of the Scientific Coordinating Panel was appointed to
determine the consensus values and methods of statistical interpretation of the intercalibration
results. Several methods were discussed in great detail. Tests were made for outliers using
the method of Barnett and Lewis (1984), and none were found. The data were of good
quality and were highly linear. The r2 values ranged from 0.997 to 0.999 using a consensus
based on the simple arithmetic means of the reported values. The subcommittee chose to
explore alternatives to the arithmetic mean and eventually settled on a multiplicative model
weighted for within-laboratory variance. The model was run with GLIM statistical software,
Version 3.77, Update 2, and gave consensus values and correction factors as shown in
Table 3-6. Although several alternatives to simple regression were evaluated, the consensus
values produced by the GLIM procedure differed only slightly from those of a simple linear
regression. The correction factors in Table 3-6 were used by the three studies to convert
their soil and dust data to a common project basis. A plot of the dust (Figure 3-8) and soil
(Figure 3-9) reported values versus the consensus means derived from the GLIM analysis
illustrates the reliability of this method.
3.1.5 Disposition of Audit Data
Based on the results of the second intercalibration exercise, a consensus value was
determined for each dust and soil sample, and biweight distributions were determined for
those that had been used in the audit program. This new distribution incorporated
interlaboratory variation. When the correction factor is applied to the reported results, the
revised number should lie between the upper and lower boundaries of the biweight
distribution. Table 3-3 lists the percentage of these audit sample values that fell within these
new boundaries. Most of the discrepancies were resolved by the corrective measures taken
by the laboratories.
3-15
-------
TABLE 3-5. RESULTS OF THE FINAL INTERCALIBRATION STUDY G*g/g)
Sample
DUST1
DUST2
DUST3
DUST4
DUSTS
SOIL1
SOIL2
SOIL3
SOIL*
SOILS
SOIL6
SOIL?
SOILS
SOIL9
SOIL10
SOIL11
SOIL12
SOIL13
SOIL14
SOIL15
BOSK
120
320
1,430
2,000
280
450
900
1,050
2,200
3,800
710
650
950
2,800
5,600
12,500
310
12,000
810
1,450
BOSX
510
910
1,100
2,300
4,000
770
930
930
2,900
5,300
13,000
290
12,000
850
1,600
XRF
BAL
121
482
1,686
3,771
267
388
808
961
2,100
3,486
640
559
896
2,514
5,200
11,000
283
10,500
793
1,400
AAS
CIN
92
329
1,307
2,924
233
441
1,033
1,080
2,555
4,227
789
675
1,036
3,126
6,493
15,963
305
14,156
929
1,705
LV
78
288
1,288
2,456
212
310
833
923
2,264
3,974
611
532
798
2,972
5,956
15,984
286
13,530
763
1,509
BAL
15
201
1,363
2,335
150
383
1,001
1,100
2,468
4,044
741
567
1,032
3,401
6,861
13,175
321
13,000
875
1,731
CIN
66
236
1,581
2,451
273
452
1,013
1,120
2,502
4,251
798
650
1,067
3,263
6,937
13,955
379
13,195
986
1,766
ICP
BOS
94
284
1,428
2,109
244
401
850
972
2,230
3,748
699
597
944
3,148
5,932
12,652
300
13,167
907
1,631
LV
72
307
1,346
2,296
191
379
912
1,006
2,286
3,843
660
626
998
3,158
6,360
12,608
294
11,440
900
1,650
When the audit sample values fell outside the boundaries of the final biweight
distribution, the batches were flagged. The options could then be to exclude these data from
the statistical analysis, reanalyze the samples, or use the original data based on other
evidence that the data are correct. The quality of soil and dust analysis in this project was
equal to or greater than the generally acceptable standards for reporting soil and dust data in
the scientific literature.
3-16
-------
TABLE 3-6. CONSENSUS VALUES AND CORRECTION FACTORS FROM
THE FINAL INTERCALIBRATION PROGRAM
XRF
AAS
ICP
Interlaboratory Consensus Values for Dust G*g/g)
Sample
DUST1
DUST2
DUSTS
DUST4
DUSTS
Study
BOS
BAL
CIN
92.8
342.7
1,319.0
2,943.4
228.3
Interlaboratory Correction
1.1527
0.7803
1.0074
54.2
221.9
1,492.2
2,378.1
232.4
Factors for Dusta
1.0416
0.9616
81.7
283.4
1,362.3
2,133.4
206.2
1.0707
Interlaboratory Consensus Values for Soil (j«g/g)
Sample
SOIL1
SOIL2
SOILS
SOIL4
SOILS
SOIL6
SOIL?
SOILS
SOIL9
SOIL10
SOIL11
SOIL12
SOIL13
SOIL14
SOIL15
Study
BOS
BAL
CIN
460.2
960.7
1,140.5
2,493.5
4,139.3
761.0
664.1
1,062.3
2,987.8
6,175.2
13,120.7
335.3
12,498.5
941.3
1,663.2
Interlaboratory Correction
1.0370
1.1909
0.8698
430.5
1,002.1
1,106.2
2,474.2
4,164.1
776.9
623.3
1,049.4
3,272.6
6,863.2
13,645.4
361.5
13,041.6
949.5
1,744.1
Factors for Soil3
1.0166
0.9839
426.6
909.6
1,018.8
2,342.1
3,706.1
736.1
656.0
1,005.4
3,274.9
6,411.5
13,224.7
323.6
13,080.0
923.3
1,716.8
1.0166
The correction factor is the value that the reported soil or dust measurement should be multiplied by in order
to adjust each value to a common basis among all three studies.
3-17
-------
3000,
1
A 2500
500
1000 1500 2000
Consensus XRF (|ig/g)
2500
3000
° Boston o Baltimore > Cincinnati xEMSL-LV
Figure 3-8. Departures from consensus dust values for each of the three studies.
Consensus XRF
Thousands
n Boston o Baltimore > Cincinnati x EMSL-LY
Figure 3-9. Departures from consensus soil values for each of the three studies.
3-18
-------
3.2 QUALITY ASSURANCE AND QUALITY CONTROL FOR HAND
DUST
The collection and analysis of hand wipes is an innovative procedure developed just
prior to the beginning of the project. There were few published reports of the measurement
techniques, no certified standards, no internal standards, and little information on which to
base decisions for acceptable analytical precision. Double blind audit samples were provided
to the study QA/QC officer as an external control for hand wipe analysis. These were
prepared as simulated samples by placing a known amount of an appropriate solution of lead
nitrate onto the blank hand wipe at the EMSL/LV laboratory, wrapping and labeling
according to the field protocol and returning to the participating laboratory for insertion into
the sample scheme. There was no attempt to determine interlaboratory Variance or to
calculate correction factors. The study QA/QC officer was responsible for reporting
problems to the laboratory director.
3.3 QUALITY ASSURANCE AND QUALITY CONTROL FOR BLOOD
LEAD
The QA/QC program for blood analysis was directed by Dr. Dan Paschal of the
Centers for Disease Control and Prevention (CDC), using the protocols developed for the
CDC blood lead certification program. Each laboratory received double blind bovine blood
samples from CDC Blind Pool 1 and Blind Pool 2 and inserted these blind samples into the
blood sample stream for the duration of the study. The data from this QA/QC program are
shown in Table 3-7. These data show the number of exceedances to be zero for all three
studies. An exceedance occurs when the mean of two replicates exceeds the 99th percentile
range established by CDC. The data also allow estimation of the probability of analytical
drift during the period of analysis. There was evidence for drift in the Boston Blind Pool 2
and marginal evidence in Cincinnati Blind Pool 1. While the statistical analysis of the QC
data for Boston blood lead analyses suggest the possibility of analytical drift (of unknown
direction) for part of the period where blood lead data were being sampled and analyzed, the
statistical methods for evaluating abatement effectiveness used by the investigators and by
this assessment would compensate for any possible analytical drift.
3-19
-------
TABLE 3-7. QUALITY CONTROL RESULTS FOR
CENTERS FOR DISEASE CONTROL AND
PREVENTION BLIND POOL BLOOD LEAD ANALYSES
Blind Pool 1
Study
Boston
Baltimore
Cincinnati
Dates
Jul 89 - Aug 91
Aug 88 - Oct 90
Aug 88 - Oct 90
n
123
66
53
Number of
Exceedances1
0
0
0
Target
Range2
1.9-6.1
3.9-6.2
1.4-5.6
Drift3
0.2092
0.6382
0.0672
n
112
59
48
Blind Pool 2
Number of
Exceedances1
0
0
0
Target
Range2
8.0-13.1
9.2-13.1
6.5-11.2
Drift3
0.0389
0.4748
0.4732
JNumber of samples that exceeded the target range established by CDC for each batch of QC blood analyses
within a pool.
2The target range is the upper and lower 99th percentile confidence limit established by CDC and differs for
each Blind Pool and each method of analysis.
3The drift test probability is a P-value for the test of the hypothesis that the slope of the difference between
the reported values and the CDC accepted value is significantly greater than zero. A P-value less than 0.05
indicates this slope may be greater than zero and that some analytical drift may have occurred over time, but
the direction of this possible drift is not indicated by this statistic.
3.4 DATABASE QUALITY
Each study maintained rigorous standards for database quality. These included double
entry, 100% visual confirmation, and standard statistical procedures for detecting outliers.
In reviewing the data for statistical analyses contained in this Integrated Report, some
data ambiguities or errors were found, confirmed, and corrected prior to use in this
>
assessment. None of these, however, would have impacted the conclusions drawn by the
individual study reports.
This evaluation of the QA/QC data shows that the three studies were comparable in
their ability to meet the requirements of their QA/QC program. Furthermore, their
performance on the audit program and intercalibration exercises suggests that the data are
comparable among the three studies, with the appropriate correction factors shown in
Table 3-6.
3-20
-------
4. INDIVIDUAL STUDIES
4.1 INDIVIDUAL STUDY INTERVENTION STRATEGIES AND
SAMPLING PLANS
4.1.1 Boston Study
The pathway intervention scheme for Boston is shown hi Figure 4-1. In Boston, all
properties hi the soil abatement group were abated. To be eligible to participate hi the study,
the average or median soil lead concentration was greater than 1500 jwg/g. The approach to
soil abatement was to remove the top 15 cm of'soil, apply a synthetic fabric, and cover with
a layer of about 20 cm of clean topsoil. The new soil was covered with sod or seeded with
grass and watered through dry months. Areas not seeded or resodded were covered with a
bark mulch. Some driveways and walkways were covered with 5 cm soil and 15 cm gravel
or crushed bank (stone with dust). On four properties, the driveway and yard were capped
with 7.5 cm asphalt without soil removal, at the owner's request. A total of 93 Boston
properties, including those abated at the end of the project, were abated hi this manner. The
information on area treated and volume of soil removed from these properties appears hi
Table 4-1. The method of excavation was by small mechanical loader (Bobcat) and hand
labor, for the most part. Initially, six properties were abated with a large vacuum device
mounted on a truck, but this proved to be unsatisfactory due to the size and lack of
maneuverability. During one extreme cold spell, it was necessary to remove large blocks of
frozen soil, often greater than 15 cm thick, by loosening with a jackhammer.
In Boston, loose paint stabilization consisted of removing chipping and peeling paint
with a HEPA vacuum and washing the surfaces with a trisodium phosphate and water
solution. Window wells were painted with a fresh coat of primer.
Interior dust abatement was performed after loose paint stabilization. Families spent
the day off-site during ulterior dust abatement. Hard surfaces (floors, woodwork, window
wells, and some furniture) were vacuumed with a High-Efficiency Particle Accumulator
(HEPA) vacuum, as were soft surfaces such as rugs and upholstered furniture. Hard
surfaces were also wiped with a wet cloth (an oil treated rag was used on furniture)
following vacuuming. Common entries and stairways outside the apartment were not abated.
4-1
-------
Exterior Paint
Dust
Local
Fugitive
Dust
J
Exterior
Dust
^
Interior Paint
Dust
—»T»~r-^S
Hand
Dust
Food
Child
Full Abatement
Stabilization
Figure 4-1. Pathway intervention scheme for dust exposure (Boston Soil Abatement
Study). Bold-line rectangles indicate pathway components monitored by
sequential sampling.
TABLE 4-1. SOIL ABATEMENT STATISTICS FOR THE THREE STUDIES
Number of properties3
Surface area (m2)
Volume soil removed (m3)
Surface area/property (m2)
Volume soil/property (m3)
Boston
36
7,198
1,212
200
34
Baltimore
63
4,100b
690
73
11
Cincinnati
171
12,089
1,813
71
11
'Includes only properties abated during study. Properties abated at the end of the study, where no further
sampling was reported, are not included in this analysis, but are included in the individual study reports.
In Cincinnati, a property is the location of the soil abatement, not the location of the child's residence.
bSurface area not provided by Baltimore report. This was calculated using Boston volume-to-surface ratio,
which is equivalent to an average removal depth of 17 cm.
4-2
-------
Although subsequent measurements of lead-based paint were made, no measurements
were made of the movement of lead from paint to house dust that would reflect the
effectiveness or persistency of paint stabilization. It was believed that any contamination
from lead-based paint would be readily apparent in the dust samples.
Between Rounds 1 and 3, the Boston study lost only three of the original 152 children
enrolled. Twenty-two of the children moved to a new location but were retained in the study
through followup and analysis of their new residence. Children with blood lead
concentrations below 7 jig/dL or above 24 jug/dL had been excluded from the study and two
of the children were dropped from the data analysis when they developed lead poisoning,
probably due to exposure to lead-based paint away from their home.
Baseline characteristics (age, SES as derived from the Hollingshead Index, soil lead,
dust lead, drinking water lead, and paint lead) were similar for the three Boston study groups
(BOS P-S, BOS PI-S, BOS SPI). The preabatement blood lead concentration was higher for
BOS P-S. The proportion of Hispanics was higher in BOS P-S than hi BOS PI-S or
BOS SPI, and the proportion of Blacks was lower. There was a larger proportion of male
children hi BOS P-S.
Data were analyzed by comparison of group means using analysis of covariance
(ANCOVA); which showed a significant effect of group assignment (intervention) for both
the BOS PI-S and BOS SPI groups. These results did not change with age, sex,
socioeconomic status, or any other variable except race and paint loading (P-XRF
measurement). When blood lead was adjusted for paint lead loading, the effect of the soil
abatement relative to the two control groups was somewhat smaller and had a lower
statistical significance (P = 0.06 versus P = 0.02). Likewise, adjusting blood lead for race
reduced the size and statistical significance of the effect of soil abatement (P = 0.09 versus
P = 0.02).
The Boston study has some limitations. Participants were chosen to be representative
of the population of urban preschool children who were already at risk of lead exposure.
The Boston Childhood Lead Poisoning Prevention Program was used to identify potential
participants from neighborhoods with the highest rates of lead poisoning. Because no study
subjects had blood lead levels below 7 /ig/dL,or in excess of 24 jiig/dL at baseline,
4-3
-------
extrapolation of the effect of lead contaminated soil abatement for children above or below
this range is difficult.
Follow-up blood lead measurements were made in Boston eleven months after
intervention and again at 23 months.
4.1.2 Baltimore Study
In Baltimore, 63 properties hi BAL SP were abated between August and November
1990. An additional seven properties that did not meet the requirements for abatement were
transferred to the control group (BAL P2). The pathway intervention scheme is shown hi
Figure 4-2. Soil covered areas on each property were divided into segments or parcels,
usually front, back, and side. Any parcel with soil lead concentrations above 500 pig/g was
abated entirely. Soil and ground cover were removed down to 15 cm and replaced to the
original level with soil having a lead concentration less than 50 /tg/g. These areas were
sodded or reseeded as appropriate. Bare areas were prepped and reseeded even if soil lead
concentrations did not warrant excavation. Additional abatement statistics appear in
Table 4-1.
The exterior painted surfaces of Baltimore homes were wet scraped over the chipping
and peeling surfaces, followed by HEPA vacuuming. The entire surface was pruned and
painted with two coats of latex paint.
The Baltimore study recruited 472 children, of whom 185 completed the study.
Of those that completed the study, none were excluded from analysis. The recruited children
were from two neighborhoods, originally intended to be a treatment and a control group.
Because soil concentrations were lower than expected, some properties in the treatment group
did not receive soil abatement. In their analysis, the Baltimore group transferred these
properties to the control group.
Because of logistical problems, there was an extended delay between recruitment and
soil abatement that accounted for most of the attrition of the participating families from the
study. In their report, the Baltimore group applied several statistical models to the two
populations to evaluate the potential bias from loss of participating children. These analyses
showed that the two populations remained virtually identical hi demographic, biological and
environmental characteristics.
4-4
-------
Exterior Paint
Dust
Atmospheric
Particles
Interior Paint
Dust
Local
Fugitive
Dust
Exterior
Dust
Interior
Dust
Secondary^
Occupational
^ Dust j
Hand
Dust
Food
Child
0 = Full Abatement || = Stabilization
Figure 4-2. Pathway intervention scheme for dust exposure (Baltimore Soil Abatement
Study). Bold-line rectangles indicate pathway components monitored by
sequential sampling.
The Baltimore study design focused on changes in biological parameters (hand dust and
blood lead) over an extended period of time. The study provided limited information on
changes in the movement of lead in the child's environment in response to intervention.
Repeat measurements of soil were done for abated properties only, to confirm abatement.
There were no abatement measurements of exterior dust, no ulterior paint stabilization, and
no interior dust abatement.
Including the prestudy screening measurements of hand dust and blood lead in the
original cohort of participants, the Baltimore study made six rounds of biological
measurements that spanned twenty months.
4-5
-------
4.1.3 Cincinnati Study
The pathway scheme for the Cincinnati study is shown in Figure 4-3. Within each of
six neighborhoods, the Cincinnati study identified all sites with soil cover as discrete soil
parcels. The decision to abate was based on soil lead concentrations for each parcel of land,
and for the depth to which the lead had penetrated. Lead was measured at two depths, the
top 2 cm and from 13 to 15 cm. If the average concentration of the top and bottom samples
was 500 jig/g or greater, the soil was removed and replaced, regardless of the adequacy of
the top cover. If any of the top two cm composite samples exceeded 500 /tg/g, that parcel
was also abated. Initially, there was an option to cultivate by roto-tilling, but this approach
was abandoned as not feasible hi this study. For areas where the top concentration was
greater than or equal to 300 /*g/g, and the average concentration of the top and bottom
samples was less than 500 /*g/g and the cover was inadequate, the soil was resodded.
Excavation was by front end loader, backhoe, and hand tools down to 15 cm, and the
replacement soil lead concentration was less than 50 j^g/g. Further abatement statistics can
be found in Table 4-1.
The approach to exterior dust abatement was to identify all parcels with one of several
types of exterior hard surfaces hi the neighborhood where dust might collect, to obtain
permission to sample and abate these areas, and to clean them once with vacuum equipment
suitable for the parcel. This vacuum equipment had previously been tested and shown to
remove about 95% of the available dust on the area. The types of surfaces identified were
streets, alleys, sidewalks, parking lots, steps, and porches. For data analysis hi the
Cincinnati report, these were grouped as (1) targeted areas adjacent to the exterior of the
buildings where children lived, such as steps, porches, and sidewalks; (2) streets, sidewalks,
and alleys throughout the study neighborhoods; and (3) parking lots and other paved areas
throughout the study neighborhoods.
The exterior dust measurements hi the Cincinnati study (and the ulterior dust
measurements of all three studies) were made in a manner that determined the lead
concentration (jig Pb/g dust), the dust loading (mg dust/m2), and the lead loading (^g Pb/m2)
for the surface measured. This required that a dry vacuum sample be taken over a
prescribed area, usually 0.25 to 0.5 m2. It is important to note that dust abatement is
4-6
-------
Interior Paint
Dust
1 Interior
1 Dust
^SecondanA
•^^^•1 Occupational 1
V. Dust J
Atmospheric
Particles
0 = Full Abatement »»» - Lead Based Paint Previously Removed
Figure 4-3. Pathway intervention scheme for dust exposure (Cincinnati Soil Abatement
Study). Bold-line rectangles indicate pathway components monitored by
sequential sampling.
expected to cause an immediate change in the dust and lead loading, but not necessarily in
the lead concentration on dust surfaces.
The Cincinnati group performed interior dust abatement after exterior dust abatement,
moving the families off-site during this activity. Vacuuming of noncarpeted areas, which
was done two times at a prescribed rate of 1 m2/min, was followed by wet wiping with a
detergent. They replaced one to three carpets and two items of upholstered furniture per
housing unit. Their previous studies had shown that carpets could not be cleaned effectively
with vacuuming alone. Where carpets could not be replaced, these were vacuum cleaned
three tunes at a rate of 1 m2/min, recognizing the limitations of this method.
The Cincinnati study recruited 307 children, including 16 children born to participating
families during the study and 50 children from families recruited after the beginning of the
study. In their main data analysis, the Cincinnati group excluded those children who were
recruited after the start of the study, plus 31 children who were living hi nonrehabilitated
4-7
-------
housing suspected of having lead-based paint and four children (in two families) who had
become lead-poisoned from other causes. Thus, data for 206 children were analyzed in the
Cincinnati report.
The Cincinnati study abated soil on 140 parcels of land scattered throughout the
neighborhoods. In CIN SEI, where soil abatement was performed in the first year, the
arithmetic mean concentration dropped from 680 /ng/g down to 134 /zg/g. In the two groups
where soil abatement occurred in the second year, CIN I-SE(D) and CIN I-SE(F), the soil
lead concentration dropped from 262 to 125 pig/g and 724 to 233 /*g/g, respectively.
If soil were the only source of lead hi the neighborhoods, exterior and ulterior dust
should have responded to the reduction hi soil lead concentrations. Exterior dust lead
loading decreased only slightly following soil and dust abatement, but returned to
preabatement levels within one year. Exterior dust should provide a measure of exposure
intermediate between soil and house dust. Where soil was abated, then exterior dust
abatement should increase the rate at which the impact of this soil abatement can be observed
on the interior dust of homes. But soil is not the only source of exterior lead, especially if
the distance between the soil and the living unit entry way is more than a few hundred feet.
In this case, the recontamination of exterior dust from sources other than soil complicates the
interpretation of the movement of soil lead into the home or to exterior play areas.
Household dust was abated hi the Boston and Cincinnati studies, but hot hi Baltimore.
The BOS SPI and CIN SEI groups received ulterior dust abatement at the same tune as soil
abatement, the BOS PI-S received interior dust abatement hi the first year, with soil
abatement in the second year, and the CIN I-SE received ulterior dust abatement in the first
year followed by soil and exterior dust abatement hi the second year.
4.2 DESCRIPTION OF THE DATA
This section focuses on the actual data that formed the basis for the conclusions reached
by the individual study reports. These data consist of measurements of soil, exterior dust
(sometimes referred to as street dust), ulterior dust (house dust), hand dust, blood lead,
exterior paint, interior paint, and drinking water. The age of the child and the date of
collection were also included hi some analyses. Tables 4-2, 4-3, and 4-4 summarize key
4-8
-------
TABLE 4-2. SUMMARY OF BOSTON STUDY DATA1
Median Soil Pb Cone. 0*g/g)
BOS SPI
BOS PI-S
BOS P-S
Median Floor Dust Pb Cone. 0*g/g)
BOS SPI
BOS PI-S
BOS P-S
Median Floor Dust Load (ing/m2)
BOS SPI
BOS PI-S
BOS P-S
Median Floor Dust Pb Load Og/m2)
BOS SPI
BOS PI-S
BOS P-S
Median Window Dust Pb Cone. 0*g/g)
BOS SPI
BOS PI-S
BOS P-S
Median Window Dust Load (mg/m2)
BOS SPI
BOS PI-S
BOS P-S
Median Window Dust Pb Load (/*g/m2)
BOS SPI
BOS PI-S
BOS P-S
Median Hand Pb Load 0*g/pair)
BOS SPI
BOS PI-S
BOS P-S
Median Blood Pb Cone. (/tg/dL)
BOS SPI
BOS PI-S
BOS P-S
GM Blood Pb Cone. (/tg/dL)
BOS SPI
BOS PI-S
BOS P-S
Round 1
2,396
2,307
2,275
2,100
2,240
2,200
24
24
40
52
59
75
13,240
19,667
17,400
293
304
239
7,005
7,196
4,179
6.75
6.75
5.75
13
12
12
12.36
11.70
11.49
Round 2
125
-
,
845
1,150
950
23
26
28
23
27
27
11,217
10,000
15,500
474
380
239
4,728
4,624
4,441
4.0
5.5
3.5
10
8
9
9.11
8.01
9.19
Rounds
115
2,084
2,212
760
1,030
1,300
15
17
19
16
18
21
21,125
15,650
12,667
373
570
504
5,735
5,697
5,559
3.5
2.0
4.5
10
11
11.5
9.90
10.74
10.75
Round 4
193
278
220
726
806
862
31
31
37
24
28
37
8,780
6,870
12,350
919
500
797
5,402
2,553
6,018
12.5
7.15
9.2
10
8
10
9.07
7.11
8.85
assignments are as used in the Boston study report, and are the same as used in this report.
4-9
-------
1
TABLE 4-3. SUMMARY OF BALTIMORE STUDY DATA
Round 1 Round 2 Round 3 Round 4 Round 5 Round 6
Median Soil Pb Cone. 0*g/g)
BALSP
BALP
440
409
22
Median Floor Dust Pb Cone (/ig/g)
BALSP
BALP
Median Floor
BALSP
BALP
Median Floor
BALSP
BALP
Median Hand
BALSP
BALP
Median Blood
BALSP
BALP
GM Blood Pb
BALSP
BALP
1,600
1,850
Dust Load (mg/m2)
40
37
Dust Lead Load (jiig/m2)
73
72
Pb Load (jig/pair)
10.7
13.6
Pb Cone. (jig/dL)
12.4
10.6
Cone. 0*g/dL)
11.0
10.9
-
_
_
12.9
14.8
11.0
10.2
9.9
10.5
1,068
1,150
37
38
38
41
7.4 8.5 12.6
9.5 6.0 17.3
9.8 8.8 9.9
9.2 7.4 8.0
9.7 8.6 9.6
9.1 7.8 8.1
-
. -
-
14.9
13.0
10.4
8.0
9.7
8.4
'Group assignments are as used in the Baltimore study report, and differ from group assignments used in
this report.
data for all three studies. For the most part, these data are the bases for the results and
conclusions presented in the individual city reports, and also for the statistical analyses in
Chapter 5 of this integrated assessment.
Each study produced similar information about the occurrence of lead in the
environment. The data sets among the studies are not perfectly comparable, however, in that
they differed in the timing of the collection relative to intervention (see Figure 2-1), the
4-10
-------
TABLE 4-4. SUMMARY OF CINCINNATI STUDY DATA1
Round 1 Round 2 Round 3 Round 4 Round 5 Round 6 Round 7
Median Soil Pb Cone, (jttg/g)
CIN SEI 680
CINI-SE 237
CIN NT 339
Median Street Dust Pb Cone, (jig/g)
CIN SEI 3,937
CIN I-SE 3,665
CIN NT 1,583
Median Street Dust Load (mg/m2)
CIN SEI 454
CIN I-SE 649
CIN NT 624
Median Street Dust Pb Load (/*g/m2)
CIN SEI 1,162
CIN I-SE 2,364
CIN NT 1,005
Median Floor Dust Pb Cone, (/tg/g)
CIN SEI 362
CIN I-SE 395
CIN NT 229
Median Floor Dust Load (mg/m2)
CIN SEI 418
CIN I-SE 167
CIN NT 147
Median Floor Dust Pb Load <0g/m2)
CIN SEI 158
CIN I-SE 69
CIN NT 35
Median Window Dust Pb Cone, (jtg/g)
CIN SEI 1,509
CINI-SE 2,000
CIN NT *" 983
Median Window Dust Load (mg/m2)
CIN SEI 710
CIN I-SE 1,258
134
247
346
3,398
3,416
1,156
242
561
755
789
1,618
957
346
388
224
134
38
126
76
18
32
1,287
1,572
816
433
380
CIN NT 2,170 2,534
Median Window Dust Pb Load Otg/m2)
CIN SEI 983
CIN I-SE 2,548
CIN NT 1,782
426
360
1,111
142
240
330
2,118
3,411
891
363
326
481
641
1127
498
325
408
209
135
117
161
54
58
32
922
1,306
548
254
269
324
242
286
172
103
262
256
2,559
2,275
968
452
420
477
968
943
587
474
431
213
197
392
200
130
243
34
1,920
2,017
1,399
4,524
9,860
8,573
15,385
26,364
12,849
122 166 132
125 182 138
331 267 266
3,231
3,040
1,086
310
126
654
808
371
442
158
163
162
76
108
92
502
592
302
966
615
648
397
358
227
4-11
-------
TABLE 4-4 (cont'd). SUMMARY OF CINCINNATI STUDY DATA1
Round 1 Round 2 Round 3
Round 4
Round 5 Round 6
Round 7
Median Mat Dust Pb Cone. 0*g/g)
CIN SEI
GIN I-SE
CIN NT
Median Mat Dust Load
Per Day (mg/m2/day)
CIN SEI
CIN I-SE
CIN NT
109
132
100
738
939
373
549
702
349
767
722
405
659
889
332
-
-
-
-
-
-
Incremental Increase
.
-
-
Median Mat Dust Pb Load Incremental
Per Day (pg/m2/day)
CIN SEI
CIN I-SE
CIN NT
Median Entry Dust Pb
CIN SEI
CIN I-SE
CIN NT
.
-
-
Cone. Otg/g)
334
425
290
6.5
18.7
1.8
Increase
6.54
7.65
3.30
606
492
367
7.7
4.7
2.0
7.62
5.14
4.67
433
468
317
4.4
4.9
2.7
2.38
3.20
0.99
491
632
286
28.2
16.6
12.2
9.80
8.02
5.29
211
102
84
-_
-
-
-
-
-
382
598
317
-
- •
-
-
-
-
488
615
284
Median Entry Dust Load (mg/m2)
CIN SEI
CIN I-SE
CIN NT
Median Entry Dust Pb
CIN SEI
CIN I-SE
CIN NT
Median Hand Pb Load
CIN SEI
CIN I-SE
CIN NT
386
272
348
Load Gig/m2)
112
95
157
Otg/pair)
6.0
7.0
3.0
113
70
238
104
38
80
5.0
7.0
4.0
230
142
294
167
70
88
5.0
5.0
3.0
590
1,394
373
250
588
106
12.0
10.0
5.5
12,671
17,889
14,509
2,502
2,700
1,714
12.5
8.0
7.0
97
161
148
56
103
58
-
•
-
301
513
1,080
150
302
264
-
-
-
Median Blood Pb Cone. (/tg/dL)
CIN SEI
CIN I-SE
CIN NT
9.2
10.8
9.0
_
-
-
7.0
9.2
5.9
8.0
8.9
6.8
-
-
-
7.9
8.0
6.4
8.3
8.8
7.8
GM Blood Pb Cone. (/ig/dL)
CIN SEI
CIN I-SE
CIN NT
8.8
10.8
8.3
-
-
-
6.9
9.3
5.7
8.8
8.6
6.8
-
-
-
8.2
7.6
7.2
8.7
8.9
7.8
JGroup assignments are as used in the Cincinnati study report and differ from group assignments in this report.
4-12
-------
spatial distribution of the sampling points relative to the expected exposure to the child, and
the manner in which the data were reduced to a central tendency.
Data were collected in rounds. That is, during a specific period of time, samples were
taken of soil, dust, etc., for a specific objective, such as establishing the concentration of
lead prior to intervention. Usually a round lasted for several weeks, perhaps three to
four months. It may be important to know when a sample was taken during a round,
especially folio whig intervention, in order to evaluate the impact on exposure. Consider the
pathway from soil => street dust => house dust => hand lead =» blood lead. One would expect,
if soil alone (not house dust) were abated and the exposure were mainly through house dust,
there would be a lag in tune between abatement and response, and the impact of intervention
might become greater with increasing tune. Conversely, the impact of intervention might be
reduced with time if there were recontammation, as would be expected if house dust were
abated but soil or other sources were not.
It is important to know how well the soil concentration measurements and house dust
concentration measurements actually represent the hypothesized pathway between soil and
house dust. If the pathways are valid, it is possible to construct a simple exposure scenario
for the individual child and to analyze these scenarios by structural equation modeling. For
example, a young child may spend most of the time indoors, whereupon the exposure
scenario becomes the lead that is available to the child through food, drinking water, air, and
dust (see Figure 2-1). Each of these proximal sources of lead is influenced by one or more
other sources of lead more remote from the immediate exposure of the child.
Some data are specific to the individual child, such as blood lead and hand lead. Some
are specific for the living unit or family, and some are specific for the property. This
distinction is important where there are several siblings in a family or several families in a
dwelling. In such cases, a single numerical value for soil such as a mean or median for the
premises could be heavily weighted if there were, for example, five children living on the
same property.
4.2.1 Measures of Central Tendency for Property Level Soil and Dust
For soil and dust, mere is a need to reduce multiple measurements within a round to a
single representative data point, or central tendency, for each property or living unit. In
4-13
-------
order to determine the appropriate central tendency for this measurement, the Scientific
Coordinating Committee discussed several alternatives at great length without reaching a
consensus. Therefore, different measures of central tendency were reported hi each of the
three studies. The following is an extended discussion of each of these measures, followed
by an argument for the use of the arithmetic mean as the best measure in these
circumstances.
The procedures for selecting a representative soil sample were based on the statistical
distribution of data hi each study. The Boston study used the median, giving no weight to
extreme values. The Cincinnati study used the geometric mean, a method that is often used
when the measured values are lognormally distributed, because it gives lesser weight to
extreme values. The geometric mean is always lower than the arithmetic mean for any set of
positive values and therefore may be an underestimate of the exposure to the child.
The distribution problem was approached differently in Baltimore, where the tri-mean
was calculated as the weighted average of the first, second, and third quartiles:
X =
Q1 + 2
-------
It would seem reasonable that the ideal method for selecting a representative value
should focus on the relationship between the soil and the child. The best measurement of
central tendency is one that perfectly represents exposure to the child. This means that
outside play activity patterns and exterior dust traffic patterns into the home must both be
evaluated. In the case of outside play activities, a sample would be taken at each location
where the child played and this sample would be weighted according to factors such as the
time spent playing there and the frequency of hand-to-mouth activity during that time.
Because this information is not available, a simplifying assumption is that weight should be
given to the location of the sample rather than concentration. Location, not lead
concentration, is the child's basis of choice for a play environment. An exposure weighted
mean of the soil samples would seem to be the most direct approach. This would be an
arithmetic mean of soil values corrected for the degree of exposure to the child. For
example, a sample taken from bare soil in an area observed to be a play area would be given
a high weighting factor for exposure. Grass covered areas with limited accessibility would
be weighted on the low end of exposure. Although cumbersome, this method is feasible
because such information was collected at the time of sampling in each study. The drawback
is that the method emphasizes the direct, outdoor playtime contact between the child and the
exterior dust, and does not consider other routes of dust exposure, such as soil =» household
dust.
An alternative solution is to consider that the child has equal exposure to the entire
surface of the soil. In this case, the perfect sample would be to scrape up this upper 2 cm of
soil, homogenize it and take a sample. Theoretically, this is equivalent to sampling in a
random pattern and taking the arithmetic mean of these samples. In this project, random
locations were taken along lines specifically selected to represent the expected high- and low-
concentration areas of the plot of soil. In this sense, the arithmetic mean is the best measure
of the central tendency of soil data for a property, and is the statistic used in this report.
Then, for populations of children at the neighborhood or higher level, the median or
geometric mean of the arithmetic property mean is the preferred measure of central tendency
for groups of children where extreme values should be suppressed.
4-15
-------
4.2.2 Adjustments and Corrections to the Data
4.2.2.1 Subjects Dropped from Study
During the analysis of their data, the Boston group discovered that two children of the
same family had apparently become exposed to lead-based paint abatement debris while
staying at a house outside then: neighborhood during a time when it was being remodeled.
Both siblings had blood lead concentrations that had tripled in less than five months, between
Rounds 1 and 3, from 10 to 35 /tg/dL and 17 to 43 ptg/dL. The Boston group analyzed their
data with and without these children, eventually excluding these data from the analyses used
to test their hypothesis. This Integrated Report accepts the conclusion that the data are
outliers and also dropped them from further analysis.
There were four children identified by the Cincinnati investigators who were either
chelated prior to or during the initial stage of the study, or who were victims of careless
remodeling work. These four children were excluded from the Cincinnati analyses and from
this assessment as well. Baltimore did not exclude any children based on medical
intervention or careless remodeling.
The exclusion of these children hi Boston and Cincinnati from statistical analyses was
not arbitrary but followed extensive discussions among all participants in the project. This
exclusion differs from the internal exclusion that occurs within specific statistical tests where
several individuals may not meet the conditions of the test. For example, in one of their
analyses, Baltimore compared the blood lead concentrations for children in Rounds 1 and 6.
By specifying that a child must have been present for both rounds, this selection excluded
children recruited in Rounds 2 and 3, and any children whose blood was not sampled in
Round 6. For other statistical analyses, some or all of these children may have been
included. For all statistical analyses hi this report that involve blood lead measurements for
specific rounds, a child is included if a blood lead measurement was taken for that round and
if other data required for the analysis are also available.
4.2.2.2 Unit Conversion
All data were converted to common units, usually metric. Corrections were made for
analytical blanks or similar analytical adjustments, as reported by each individual city
4-16
-------
research team. In this assessment, all data for soil and dust were adjusted by the
interlaboratory correction factor specific for each study and shown in Table 3-6.
4.3 STUDY DESIGNS
4.3.1 Design Differences
Table 4-5 describes the design differences among the three studies. While considerable
effort was made to coordinate the study designs so as to assure the highest possible degree of
comparability among study results, the investigators in the three cities faced different design
issues that precluded carry ing out completely identical or equivalent studies. Thus, although
participant recruitment and certain other aspects were similar across the three cities, some
salient differences are also worth noting.
The first difference was that there were different levels of remediation or treatment
among the cities. Boston used two comparison or reference groups hi addition to the soil
abatement group, whereas Baltimore used only one such group. In the Cincinnati study,
there were three levels of intervention. Also, the trigger level for soil lead removal varied
somewhat across the cities. In the Baltimore and Cincinnati, a maximum level of 500 /*g/g
or greater hi the parcel or residential property triggered soil removal. In contrast, all Boston
properties had mean or median soil concentrations exceeding 1500 j*g/g. Properties recruited
in the Boston study were scattered across four large neighborhoods or urban areas, although
households were assigned at random to the treatment group for soil removal and not
specifically limited to any given neighborhood. This randomization approach in Boston
provides a more thorough statistical treatment of multiple sources of lead and analysis of
environmental cofactors. The Baltimore study was carried out in two large neighborhoods,
with soil lead removal restricted to only one of the neighborhoods (Lower Park Heights).
Most houses above the soil lead trigger level hi the Lower Park Heights neigborhood hi the
Baltimore study had yard soil removed, but some did not, and no house in Walbrook junction
had "soil removed. The Cincinnati study was carried out in six smaller neighborhoods, with
soil and exterior dust removal only carried hi the Pendleton neighborhood. In the Cincinnati
study, all parcels hi Pendleton above the soil lead trigger level had soil removed in the first
4-17
-------
TABLE 4-5. DESIGN DIFFERENCES AMONG THE THREE STUDIES
Design Feature
Number of treatment groups
Number of rounds with blood Pb
measurement
Interval between abatement and final
Pb measurement (months)
Soil removal trigger level (/tg/g)
Paint stabilization
Number of neighborhoods
Participant recruitment
Treatment assignment to participants
Control groups with no intervention
Age structure of participants (%)
Ethnicity (%)
Black
Hispanic
White
Other
Male/female ratio
Blood sample collection
blood
0-1
1-2
2-3
3-4
4-5
5-6
6-»
Rl
R2
R3
R4
R5
R6
R7
Boston
3
4
22
1,000
Interior
4
Volunteer
Random
No
2.7
24.0
34.0
34.7
4.7
51
15
7
27
47/53
1-2 mo preabate
3-4 mo after Rl
10 mo after Rl
22 mo after Rl
Baltimore
2
6
10
500
Exterior
2
Volunteer
By Neighborhood
No
8.6
17.6
18.1
18.4
20.3
14.5
2.5
100
0
0
0
48/52
24 mo preabate
12 mo preabate
5-8 mo preabate
8-10 mo after R3
14-16 mo after R3
18-20 mo after R3
Cincinnati
3
5
20
500
None3
6
Volunteer
By Neighborhood
Yes
29.9
17.2
17.6
15.8
14.0
5.4
97
0
2
1
44/56
1-2 mo preabate
3-4 mo after Rl
11 mo after Rl
16-18 mo after Rl
22-24 mo after Rl
•Dwelling units had been thoroughly rehabilitated 20 years pripr to study, leaving little exposed lead-based
paint.
year. Soil abatement occurred during the second year in the Back, Dandridge and Findlay
neighborhoods, and in the control groups at the end of the study.
4-18
-------
Paint was stabilized inside all Boston houses and outside all Baltimore houses, but not
in Cincinnati where it was believed that only gut-rehab houses had been recruited into the
study. No Baltimore residence received interior abatement, either of dust or lead paint,
whereas the majority of the residences in the Boston and Cincinnati studies received interior
dust abatement whether or not they were hi the soil removal treatment group. Exterior dust
abatement was performed only in Cincinnati.
Demographic differences among study populations should also be noted. The age
distribution of children at the time of abatement differed among the three studies, , The
Baltimore group had more children of age at least four years, since many of the children had
been initially recruited up to 2 years earlier. Almost all of the children initially recruited in
the Baltimore study were of African-American ancestry; by the final phase of the study,
100 percent of the study group was African-American. The Cincinnati study group was
slightly more diverse, with a small percentage of Caucasians of Appalachian origin. The
Boston group was the most diverse, with substantial subgroups of white and Cape Verdean
children, and also with a large percentage of African-American children. Percentages of
male and female children differed somewhat among the cities. While all of these inner city
households tended to be economically disadvantaged, the majority of the households in
Baltimore were occupied, by the property owner, which was uncommon hi the other two
cities.
Lastly, as for biological measurements indexing changes hi lead exposure, each study
involved collection of preabatement and postabatement blood samples and their analyses.
However, the numbers of sampling points varied across the studies. The studies had four to
six rounds of blood lead collection, with one to three pre-abatement rounds, a short-term
post-abatement round (about two or three months), and two to three rounds up to two years
post-abatement.
4.3.2 Strengths and Weaknesses of Study Designs
In an ideal situation, each study would have been designed around a neighborhood
where soil was a significant source of dust hi the child's environment and this soil contained
an amount of lead sufficient to impact the child's exposure. There would also be no other
sources of lead in the child's environment, and the child's history of lead exposure would
4-19
-------
have been stable. The study would incorporate all children in the neighborhood and these
children would be demographically similar to a representative sample of children across the
United States. Their behavior and activity patterns would also be similar and representative.
Children would be randomly assigned to a study group, and the population would be
sufficiently large to test the main hypothesis as well as any other question that might arise
concerning sibling, ethnic, age, and sex effects. The sample design plan should be sufficient
to establish the pattern of lead exposure for the child population prior to intervention,
including seasonal cycles and long term trends in blood lead and dust lead loading.
None of the three studies in this project met these ideal conditions, nor could any other
neighborhood hi any city. The issue then is whether any departure from this ideal design
seriously impacted the conclusions that the study could have made under ideal circumstances
or did make under these more realistic circumstances. In this respect, the strengths and
weaknesses of the three study designs are discussed and the hypotheses of the individual
studies are reevaluated.
The strong points of the Boston study are that it was designed as a group of
demographically similar neighborhoods where soil was a significant source of dust and lead
exposure. It appears that the children were also similar in terms of behavior and activity
patterns and diverse in age, ethnicity, and sex. The main weaknesses hi the Boston study are
that some children were excluded from selection into the study because of high or low blood
lead concentrations. This truncation of blood leads above 24 jug/dL, excluding these
children may have substantially diminished the impact of intervention, on the assumption that
children with higher blood lead concentrations would show a greater response to reduced
exposure. This hypothesis is .tested hi this assessment.
There was also a sufficient amount of lead from ulterior and exterior lead-based paint in
most residences to partially obscure the impact of soil abatement. The Boston properties
were not contiguous, so that no measure of neighborhood level lead exposure of intervention
can be made. Even though the Boston study may not represent typical U.S. urban
neighborhoods, the study is likely applicable to a broad range of circumstances because the
experimental treatment was assigned at random to children living on properties in four
distinct neighborhoods, and this randomization hi study design is likely to have eliminated
*
many neighborhood level confounding factors.
4-20
-------
The Baltimore study design incorporated two demographically similar neighborhoods,
one designated as a trea.tment group and the other a control. Soil may have been a
significant source of dust in the child's environment, but the soil and dust sampling protocols
were inadequate to test this hypothesis. The Baltimore group identified up to four parcels of
soil on each property and abated any parcel where the maximum soil concentration exceeded
500 /*g/g. In those cases where the abated parcel, by circumstance, was not one that the
child would play on or one that would contribute in any other way the child's lead exposure,
*
then there would be no change in the child's lead exposure hi response to abatement and no
expected decrease hi blood lead, even though the property was hi the "abated" group. This
type of error, normally called "misclassification", could not be evaluated in this assessment
because the soil samples were not identified by parcel identifiers.
For blood lead and hand lead, the Baltimore study sampled three preabatement and
three postabatement rounds. While it is possible that these data may be sufficient to identify
a seasonal cycle or a long term trend similar to that discussed hi Section 2.3.1, analyses for
these effects were not made by either the Baltimore report or hi this assessment. In the
Baltimore study there was insufficient lead in the soil to demonstrate an effect of abatement
and there was a substantial amount of lead in exterior and interior paint to obscure the impact
of intervention. Windows were not included hi the sampling pattern for house dust. The
floor dust was sampled only once for most children, which is not frequently enough to detect
changes hi the child's exposure to environmental lead. For some children there was a
postabatement dust sample taken.
The Cincinnati study alone evaluated intervention on a neighborhood wide basis. The
frequency of environmental sampling was sufficient to identify important features of
environmental dust mobility. However, soil appears not to have been a major contributor to
house dust, nor was the soil lead concentration sufficiently high to impact exposure to
environmental lead through exterior dust. Most of the lead in exterior dust appears to have
come from nonsoil sources. Of the six neighborhoods hi the Cincinnati study, one (Back
Street) was too small to continue in the study, and another (Glencoe) may have been
demographically distinct from the rest. Children were assigned to study groups based on
then: resident neighborhood rather that randomly assigned from the group of neighborhoods,
4-21
-------
and these neighborhoods were grouped by the Cincinnati investigators into treatment and
control groups in a non-random fashion.
4.3.3 Modifications of the Hypotheses
Each study developed a specific hypothesis that was intended to be tested by the data
and observations from the original study design. For the purposes of clarification only, this
report restates the original hypotheses hi a form that is more amenable to statistical
interpretation. This clarification will aid the reader in understanding the statistical analyses
that are presented in Chapter 5. In the case of Boston and Cincinnati, the original
hypotheses are broken into three subhypotheses.
The first subhypothesis for Boston identifies their group "S" as the primary treatment
group, and the second hypothesis identifies group "A" as a positive control group.
An analogous distinction is made for the Cincinnati study, with the first subhypothesis for the
soil abatement neighborhood of Pendleton, and the second subhypothesis for dust abatement
neighborhoods Back Street, Dandridge, and Findlay. No subhypotheses were designed for
the Baltimore study, since the neighborhood effects were confounded with treatment effects.
In the EPA reanalyses, additional subhypotheses were developed based on post hoc
differences between certain subgroups. The formal statement of the original Boston
hypothesis is:
A significant reduction (equal to or greater than 1,000 pg/g) of lead in soil
accessible to children will result in a mean decrease of at least 3 ng/dL in the
blood lead levels of children living in areas with multiple possible sources of lead
exposure and a high incidence of lead poisoning.
The actual hypothesis that was tested is similar to this and might be restated as follows, in
the null form:
(i) A one-time reduction of at least 1000 ppm in average soil lead concentration
of residential property without substantial deteriorating exterior lead paint,
accompanied by remediation of interior household dust and control of
recontamination of interior dust by stabilization of interior paint, will not
result in a reduction of blood lead in children living at the residence.
(ii) A one-time dust lead abatement inside a residential property without
substantial deteriorating exterior lead paint, accompanied by control of
4-22
-------
recontamination of interior dust by stabilization of interior paint, will not
result in a reduction of blood lead in children living at the residence.
(Hi) The reduction in (i) will not be greater than the reduction in (ii), if any.
The original primary Cincinnati hypothesis, pertaining to blood lead levels, was stated
as
(1) A reduction of lead in residential soil accessible to children will result in a
decrease in their blood lead levels.
(2) Interior dust abatement, when carried out in conjunction with exterior dust
and soil abatement, would result in a greater reduction in blood lead than
would be obtained with interior dust abatement alone, or exterior dust and
soil abatement alone.
/
To reflect actual experimental conditions, this hypothesis could be modified as follows
and restated in the null form
(i) A one-time reduction of lead in accessible soil and in street dust in a
neighborhood, accompanied by abatement of household dust in the child's
apartment or residence unit, will not result in a reduction of blood lead in
children living in gut-rehab housing in the neighborhood.
(ii) A one-time reduction of lead in household dust in the child's apartment or
residence unit, will not result in a reduction of blood lead in children living
in gut-rehab housing in the neighborhood.
(in) The reduction in (i) will not be greater than the reduction in (ii), if any.
The original Baltimore hypothesis, stated in the null form, is
A significant reduction of lead (> 1,000 pg/g) in residential soil accessible to
children will not result in a significant decrease (3 to 6 ftg/dL) in their blood lead
levels.
A restatement of this hypothesis that takes into consideration the actual preabatement
conditions and stated hi the null form is
A one-time reduction of at least 500 ppm in the maximum lead concentration in yard
soil, even when not accompanied by abatement of household dust or lead paint inside
the child's apartment or residence unit, will not result in a reduction of blood lead in
children living in housing in which exterior lead paint has been stabilized.
4-23
-------
The actual statistical hypotheses and subhypotheses are all expressed as "null
hypotheses", that abatement has no effect, makes no difference, or causes no change
compared to no abatement, after other processes or changes have been taken into account.
This differs some what from the original statement of the hypotheses by the investigators.
To clarify this distinction, a little algebraic notation may be helpful. Let "E" denote the size
of the abatement effect, for example, a reduction hi blood lead concentration in the
abatement group relative to a control group, and let s.e.(E) denote the estimated standard
error of the effect size estimate. Most statistical tests carried out by the investigators and by
EPA involve use of statistics that are essentially equivalent to a "t" statistic, defined by
t =
s.e.(E)
For the original hypothesis proposed by the Boston investigators, where a blood lead
reduction of at least 3 /zg/dL, was stipulated, a somewhat different t statistic would be
appropriate,
f =
s.e.(E)
Similar alternative hypotheses were adopted hi a roughly similar form by the Baltimore and
Cincinnati investigators for the purpose of power calculations to aid hi study design. In this
report as well as hi their reports, inferences were in fact based on statistics like "t" for
testing no effect.
4.4 INDIVIDUAL STUDY CONCLUSIONS
In their report following the first phase of their study, the Boston group stated their
conclusions:
"...this intervention study suggests that an average 1,856ppm reduction in soil
lead levels results in a 0.8-1.6 pg/dL reduction in the blood lead levels of urban
children with multiple potential sources of exposure to lead."
Following the second phase of the study, they concluded (Aschengrau et al., 1994):
4-24
-------
"The combined results from both phases suggest that a soil lead reduction of
2,060ppm1 is associated with a 2.2 to 2.70 pg/dL decline in blood lead levels."
The basis for their initial conclusions consisted of an analysis of variance comparing
mean blood lead changes among the three intervention groups, paired t-tests for within group
effects, and analysis of covariance with one-at-a-time adjustment for age, SES, race, sex,
paint, water, and mouthing behavior. The analysis of covariance was performed using no
transformation of blood lead data, which appeared to be normally distributed.
The conclusions from the second phase of the study are based on additional analyses of
phase one and phase two data using two-way analysis of variance (ANOVA) with repeated
measures. Soil was abated for the two original control groups (BOS PI-S and BOS P-S) at
the beginning of phase 2. The reduction hi blood lead is based on pre- and postabatement
measurements of all three groups.
The Baltimore group stated their conclusions as follows:
"Statistical analysis of the data from the Baltimore Lead in Soil Project provides
no evidence that the soil abatement has a direct impact on the blood lead level of
children in the study."
"In the presence of lead-based paint in the children's homes, abatement of soil
lead alone provides no direct impact on the blood lead levels of children."
The basis for these statements consisted of an adjusted and unadjusted analysis of
selected covariates. The natural log of the blood lead of children in the treatment group
showed no significant difference from the natural log of the blood lead of children hi the
control group, even when adjustments were made for: age, SES, hand lead, season, dust,
soil, sex, weak mouthing behavior, or strong mouthing behavior. These analyses were made
on two sets of data. The first set consisted of all children enrolled hi rounds one and six.
The second group consisted only of children enrolled hi all six rounds.
The Cincinnati conclusions can be paraphrased as follows based on their individual
report:
Following interior and exterior dust and soil lead abatement, blood lead
concentrations decreased (in Area A) from 8.9 to 7.0 (21%) but increased to 8.7,
This value for soil, 2,060 ppm, cited in their published report, was not adjusted by the Boston group with the
interlaboratory correction factor of 1.037 hi Table 3-6.
4-25
-------
10 months postabatement. Following interior dust abatement alone blood lead
concentrations decreased from 10.6 to 9.2 (13%) four months postabatement and
were 18% below preabatement 10 months postabatement. With no abatement,
blood lead levels decreased by 29 and 6% during these same time periods. Other
comparisons also revealed no effects of the soil or dust abatement.
There was no evidence that blood lead levels were reduced by soil lead or dust
abatement in Area A (with soil, exterior dust, interior dust abatement). There was
a slight reduction (net reduction over control area) of 0.6 pg/dL in Area B that
might be attributed to interior dust abatement. This difference is not statistically
significant.
The basis for the Cincinnati conclusions was a comparison of environmental and blood
lead data for the three treatment groups from Rounds 1, 3, 4, 6, and 7 and of additional
environmental data from Rounds 2 and 5.
4-26
-------
5. RESULTS OF INTEGRATED ANALYSES
5.1 CONCEPTUAL APPROACHES TO EVALUATING RESPONSE TO
ABATEMENT
Many statistical procedures rely heavily on an analysis of the correlation structure of
the data (how one variable changes in response to or in conjunction with another, covarying
variable) within a single measurement period or round of measurement, and on the
longitudinal structure of the sampling scheme, where several rounds of measurement are
made. In preparation for a detailed description of statistical methods, the following
discussion briefly reviews the variability of the key variables and the correlation structure
among their covariates.
In designing the three individual studies, the investigators expected that blood lead,
handwipes, and house dust would be predictors of childhood lead exposure. Changes in
blood lead levels, hi hand lead levels, and in household dust lead levels were expected to
occur hi response to effective intervention and hi response to other biological and
environmental changes that occurred independent of intervention. Many of these patterns of
change were discussed hi Chapter 2 and are reviewed briefly here. As shown earlier in
Figure 2-7, blood lead concentrations in young children often increase up to ages 2 or
3 years, which are peak ages for ingestion of soil and dust during play, and then decrease
slowly in older children (U.S. Environmental Protection Agency, 1986; Clark et al. 1988).
Hand lead loadings increase steadily with age (Bornschein et al., 1988). House dust lead
levels may increase or decrease as changes hi sources or exposure pathways cause changes hi
the amounts of dust that move through the environment and in the amounts of lead hi that
dust.
Childhood blood lead concentrations are, to some extent, a measure of the recent
history of lead exposure and may respond to environmental changes in lead within a tune
frame of a few months (see Figure 2-4). Reductions hi blood lead due to reductions hi
exposure possibly attributable to intervention might be somewhat attenuated by the
remobilization of lead in bone tissue. Figure 2-4 shows the complexity of biokinetic
translocations of lead when the total body burden is decreasing. If the total lead exposure of
5-1
-------
the child decreases, there seems to be no doubt that the blood lead concentrations would also
decrease, but measurements of this decrease would be complicated by the remobilization of
bone and other tissue lead, and interpretation of these measurements would be complicated
by the knowledge that the reduction in exposure cannot be fully attributable to reductions in
soil lead exposure.
Changes in blood lead must be interpreted in the context of four time-dependent effects
that are independent of each other, as summarized below.
(1) Seasonal changes hi children's blood lead concentrations have been
reported hi several longitudinal studies. These usually show a peak in
blood concentration during the late summer months.
(2) Changes occur with age during early childhood such that blood lead
concentrations usually peak between 18 and 27 months of age.
(3) Long-term changes hi national baseline levels of exposure, believed to be
mostly from reductions of lead hi gasoline and hi food, are reflected in a
downward trend for U.S. childhood blood lead levels observed since
1978.
(4) Further changes can be attributed to household- and neighborhood-level
interventions of the types reported hi this project.
The first three of these effects were discussed hi Section 2.3 and are summarized here. The
fourth is the main topic of this chapter.
5.1.1 Expected Impact of Intervention
5.1.1.1 Expected Impact of Soil Abatement on Exterior and Interior Dust
The key to understanding the impact of soil abatement on ulterior dust is to observe
changes hi the three components of the interior dust measurement: lead concentration
(micrograms of lead per gram of dust), lead loading (micrograms of lead per square meter),
and dust loading (milligrams of dust per square meter). Where there was no ulterior dust
abatement, the lead concentration hi interior dust should decrease gradually over tune in
response to soil abatement, provided that the influence of other sources such as lead-based
paint have been minimized. Also, the lead loading should decrease if the dust loading
remains constant. If interior dust abatement has occurred, the lead concentration should
gradually reach some new equilibrium concentration determined by the sources of lead and
5-2
-------
dust, and the lead loading and dust loading should increase and decrease in tandem according
to the pattern of (a) dust movement to the house or other play environment and (b) frequency
and efficiency of housecleaning.
Chipping and peeling lead-based paint can contribute vast amounts of lead to household
dust over a relatively short period of time. A single chip 5 cm x 5 cm at 6 mg Pb/cm2
contains enough lead (150,000 /*g) to increase the concentration of lead in a house with
7 g dust (54 mg/m2 x 130 m2) by 21,000 jwg/g. In this project, chipping and peeling paint
was removed by paint stabilization. But there is no measure of the amount of lead that
passes from paint to housedust by other routes, such as abrasion, weathering, or microscopic
flaking. In the interior of Boston homes, paint stabilization also included repainting the
surfaces with a coat of latex primer paint to slow or minimize recontamination of interior
house dust from paint.
The confounding effect of lead-based paint can be minimized in three ways:
(1) exclude homes with lead-based paint; (2) stabilize me paint so that the rate of
incorporation to house dust is minimized; and (3) compare measurements for areas where the
influence of lead-based paint is probably high relative to soil with data for areas where the
relative influence is low.
Exterior dust was measured and abated in the Cincinnati study only, and the results
suggest a highly fluctuating rate of recontamination from non-soil sources. With a source of
lead of this magnitude at the neighborhood level, it is virtually impossible to measure the
impact of soil abatement on house dust directly. However, if abatement is considered on the
broader scale, where neighborhood cleanup would include soil, external dust, and any other
sources of lead external to the home, then the house dust measurements made immediately
inside the homes can be used to assess this "total neighborhood abatement" .
5.1.1.2 Expected Impact of Soil and Dust Abatement on Hand Lead Loading
It was expected that hand dust would serve as a surrogate measure of changes hi
exposure following abatement, to augment information about blood lead changes. Hand dust
reflects the child's recent exposure (since the latest hand washing) but is only a measure of
lead loading, not lead concentration or dust loading, because the total amount of dust is not
measured. Consequently, it is not possible to infer the source of lead (soil or paint) by
5-3
-------
differences in concentration, nor is it possible to assess the housekeeping effectiveness by
observing changes in hand dust loading, as with house dust loading. In fact, even the
changes in hand lead loading proved to be a poor indicator, hi this project, of changes in
blood lead.
5.1.1.3 Expected Impact of Soil and Dust Abatement on Blood Lead Concentrations
Soil lead remediation in residential yards was expected to have both direct and indirect
effects on childhood lead exposure. The direct effect of removing lead contaminated soils is
to prevent access to the lead in the soil. For most children, direct exposure to lead hi soil is
likely to come from fine particles of loose soil or exterior surface dust that adhere to the
child's hands and are transferred to the child's face and mouth during hand-to-mouth contact
that is part of normal behavior for preschool children and infants. Most children do not eat
large quantities of soil. Quantitative estimates of soil ingestion by children are limited and
highly variable, probably due to differences in methodology and choice of tracer element
used in determining the estimate (U.S. Environmental Protection Agency, 1989).
By calculating a single estimate of soil ingestion for each subject, Stanek and Calabrese
(1995) determined the median daily soil ingestion for 64 children living in Amherst, MA.
Mean soil ingestion estimates were 45 mg/day or less for 50 % of the children and 208
mg/day or less for 95% of the children. Some children may regularly ingest a large amount
of soil either in a non-pica situation where the child is on the upper tail of a unimodal
distribution for soil ingestion, or hi a pica situation, where the child habitually eats nonfood
objects such as soil. Some adults (soil or clay eaters) are known to experience geophagia,
but these are atypical conditions and are not appropriate for assessing soil risks for the
majority of children.
Soil is also a source of lead hi dust hi the child's play areas. Soil hi the residential
yard may be tracked into the house by its occupants (including pets), and fine exterior dust
particles may become re-entrained and carried into the house as micro-scale air contaminants.
Fine dust particles may adhere to the child's hands and may contaminate food during its
preparation. This dust is usually a more important medium of lead intake than is the direct
ingestion of soil. The soil lead pathway is one of several possible sources of lead in dust
(see Figure 2-3).
5-4
-------
Blood lead concentrations should respond to soil and dust abatement through the impact
of intervention on two routes of exposure: (1) hand-to-mouth activity, reflecting the impact
of interior house dust and exterior play area dust on exposure; and (2) food contamination,
reflecting the incorporation of house dust in food during kitchen preparation. There was no
measure of the incorporation of house dust into food during this project. Intuitively, the
impact of interior dust abatement should be the same, or at least comparable, for food and
hand dust. In some homes, however, lead-based paint is found primarily in kitchens and
bathrooms, where the remobilization of lead in dust from lead-based paint following
stabilization would have a greater impact on food than hand dust. There is a limited amount
of data, not yet analyzed, where kitchen floor dust can be compared to bedrooms and other
living areas, and likewise for window wells. Most of these data, however, are from the
Cincinnati study, where there was a minimum influence of lead-based paint.
5.1.2 Evaluation of Specific Statistical Approaches
The studies in each of the three cities are characterized as longitudinal intervention
studies with fixed treatment groups, and there are at least seven statistical methods that could
be used to analyze their data. Of the seven methods, ranging from the simplest analysis of
blood lead data alone to complex analyses of changes in blood lead concentrations that occur
in response to changes in environmental lead over time, this assessment used four methods
for each of the three studies. In this introductory section, each of the seven methods is
discussed, with the rationale given for whether or not the method would be appropriate for
this assessment. A more detailed description of the statistical methodology is given hi
Sections 5.4, 5.5, and 5.6. The seven methods are
1. Cross-sectional analysis of variance
2. Cross-sectional analysis of covariance
3. Cross-sectional structural equation models1
4. Longitudinal analysis of covariance
5. Repeated measures analysis of variance1
6. Repeated measures analysis of covariance1
7. Longitudinal structural equation models1
Methods used in this report.
5-5
-------
By using more than one method, and by using methods that, in general, are more
comprehensive in their treatment of the data, this assessment is able to increase the strength
of the conclusions and, in some cases, provide a more detailed explanation of the nature of
the effect than appeared in the individual city reports.
The statistical methods described above differ primarily in the complexity with which
they treat the data. This complexity arises from the stepwise insertion of mathematical terms
into what is otherwise a very simple equation, usually hi a linear form. These equations are
presented below and, as an aid to the reader, descriptive graphics and simple mathematical
solutions to the equations are provided.
5.1.2.1 Cross-sectional Analysis of Variance
The first method, not used hi this assessment, is a simple cross-sectional analysis of
variance, with each round treated separately, as illustrated hi Figure 5-1. In Figure 5-1, we
show two groups of children, denoted "A" and "B". These groups are observed at several
times (rounds), at least once before and after the soil abatement, shown as separated by a
vertical dashed line. The amount of lead present hi the residence and hi the child at each
round is suggested by the number of dots hi each figure. The changes from soil abatement
hi group B are shown by the smaller number of dots hi the yard and residence, and the
expected effect of abatement is shown by the smaller number of dots in the child, implying a
reduction hi blood lead and therefore a reduction hi the total body burden of lead. A simple
analysis of variance would compare the group mean blood lead between Group A and Group
B. Before abatement, the two groups appear roughly equal. After abatement, Group B has a
lower blood lead than Group A. The effect size would be calculated as the difference
between the groups after abatement. This approach is a simplified description of one of the
analyses used hi the Baltimore report. The children hi each group may be somewhat
different at each round, depending on attrition and recruitment. An alternative version of
this approach uses the change hi blood lead in each child from a preabatement round to a
postabatement round, which is a simplified description of one of the approaches used hi the
Boston report.
The equations of this method are fairly simple and form the basis for the more complex
analyses that use computing capabilities that are much stronger than were available earlier for
5-6
-------
Group A
Group B
Before
After
Figure 5-1. Analysis of variance. Blood lead is compared across treatment groups at
each round. Environmental measurements of dust and soil are not part of
the statistical analyses. The density of dots hi the house and child figures
shows a scenario in which soil abatement in group B reduces soil lead,
house dust lead, and blood lead. Comparison of Group A and Group B is
made only at the postabatement round, where the vertical dashed line
separates pre- and postabatement scenarios.
routine analyses. For an analysis of variance of factors that influence blood lead
concentrations, only the group effect is evaluated, which means that the statistical test simply
looks for a systematic difference between groups, while excluding the effects of covariant
data such as age and sex. If the children in each group are otherwise equivalent, then the
group effect is, by inference, the effect of treatment on the child. In a representative
equation for this method, the response variable, Yir (e.g. blood lead for child i at round r) is
estimated by the' equation
5-7
-------
(5-1)
Yfr is the response variable for child i at round r
Ggr is the fixed mean response for treatment group g at round r
Qir is an error term, referred to as measurement error, for child i at round r
The term Ggr can be used to calculate the effect size. Each of these statistical methods
allows estimation of an effect size, which cannot be defined as an intrinsic property of a
group or treatment. Rather, it is defined in terms of a relationship to a control group or
combination of control and treatment groups. It is important to compare treatment groups
with control groups because even "control" groups may change over time or may respond to
nontreatment environmental influences. The effect size, AG_, for the analysis of variance
o
method would be calculated as the difference in fixed effect between two treatment groups at
each round separately:
- G
2r
(5-2)
In the case where some households in a group have multiple siblings, analysis of variance
could also be used to separate the treatment group effect from a random household effect
for each round r separately, using the equation:
eir
(5-3)
then estimating the treatment group effect at each round as above,
- G
2r
(5-4)
5-8
-------
The advantage of analysis of variance is that this method can deal with multiple groups
or treatments. The limitations are that it can be applied to only one response variable and to
the before/after treatment differences for that variable. The method ignores the correlation
structure among the measurements and, thus, does not have the capability to adjust for
covariates.
5.1.2.2 Cross-sectional Analysis of Covariance
The six remaining methods achieve a progressively more refined estimate of the group
effects, mainly by analysis of the correlation structure within the data set. Cross-sectional
analysis of covariance, illustrated in Figure 5-2, is similar in form to cross-sectional analysis
of variance, adding one or more terms that adjust the estimate for interaction of covariates.
Figure 5-2 is the same as Figure 5-1, but the arrows suggest that child's blood lead used in
the analyses is adjusted for soil lead and dust lead concentrations. The hypothesis used here
is that there is a more or less strong relationship between dust lead and blood lead, and
between soil lead and blood, as two separate media to which the child may be exposed.
Additional adjustments may be made for other household-specific covariates such as
socioeconomic status (SES) or race, or for child-specific covariates such as age, gender, or
mouthing behavior. Adjustments for these covariates allows a better understanding of the
possible processes or mechanisms of abatement, including separation of group mean
differences into those components that related to changes or differences hi soil lead and dust
lead, and other effects that are not attributable to abatement changes hi soil lead or dust lead.
However, the analysis of covariance (called ANCOVA) hi this formulation does not use the
changes in covariate values before and after abatement. Some models in the Baltimore report
used this approach. Similar models in the Cincinnati report used differences in blood lead or
hand lead vs changes hi environmental lead. This method makes adjustments for
between-group effects and environmental covariance, but still doesn't estimate the
before/after effect of intervention.
In this case, a term XfrBgr is added to Equation 5-1 to account for the covariates, and
each round is again analyzed separately:
5-9
-------
Group A
Group B
Before
After
Figure 5-2. Analysis of covariance. Arrows show that the individual child blood lead
concentrations are adjusted for soil lead and dust lead concentrations or
loadings in the child's residence, and may also be adjusted for other
household-specific or child-specific covariates such as age or gender.
However, blood lead is compared across treatment groups at each round
with no analyses of effects between rounds.
(5-5)
Xir is the covariate value for child i at round r
Bgr is the covariate effect for group g at round r
Note that Equation 5-5 has taken on a linear form and, with multiple covariates, could be the
basis for multiple linear regression analysis. The effect of treatment group after covariate
adjustment would still be calculated as equation 5-4, but also has a component that is related
to the covariate difference, which may be different in each group,
A(XB) - X^. -
(5-6)
5-10
-------
where Xj is a typical covariate value for Group 1 and X2 a typical value for Group 2.
Like analysis of variance, this method can deal with multiple groups or treatments. Although
it also ignores the correlation structure among the variables, the method can be used to adjust
for covariates, and these covariates may be either numerical or categorical. A limitation of
analysis of covariance is that covariates are assumed to be known without measurement
error, and this becomes a problem when the analysis is of two environmental variables, each
with its own undetermined measurement error. Neither cross-sectional analysis of variance
nor cross-sectional analysis of covariance was used in this report or in the individual city
reports.
5.1.2.3 Cross-sectional Structural Equation Models
Cross-sectional structural equation modeling (Figure 5-3) has the advantage of modeling
the relationships between several environmental pathways simultaneously, such as soil to
dust, dust to blood, soil to blood. In this simple form, the method provides no longitudinal
information, and therefore cannot fully treat such time dependent variables as blood lead
when the main contributor to blood lead at time t is the blood lead at time t-1 (see
Figure 2-4). Figure 5-3 is essentially the same as Figure 5-2, with arrows showing the
relationship of soil lead and house dust lead to blood lead. There is an additional arrow
showing a relationship between house dust lead and soil lead. This allows assessment of the
hypothesis that house dust lead has a component attributable to soil lead, and possibly other
components that are not attributable to soil lead (such as lead-based paint or secondary
occupational exposure). It then becomes possible to assess the total relationship of soil to
blood lead, including the indirect soil-to-dust-to-blood pathway. While separate regression
analyses of blood lead vs soil lead and dust lead, and dust lead versus soil lead can be
carried out, combining these equations without consideration of the simultaneous estimation
of the equations using the same data involves a potential "simultaneous equation" bias.
Therefore, simultaneous modeling of blood lead and dust lead was carried out hi the
Cincinnati report for each round. Some cross-sectional structural equation model analyses
were performed hi this report using Round 1 or preabatement data to assesss the extent to
which soil might have been a significant source of dust lead and blood lead before abatement.
Structural equation models using differences of blood lead and environmental lead allow
5-11
-------
Group A
Group B
Before
After
Figure 5-3. Cross-sectional structural equations model. Arrows show that the
individual child blood lead concentrations are adjusted for soil lead and
dust lead concentrations or loadings in the child's residence, and may also
be adjusted for other household-specific or child-specific covariates such as
age or gender. There are also arrows showing that house dust lead is
related to soil lead, so that the total soil effect consists of a direct exposure
pathway and an indirect soil-to-dust-to-blood pathway. However, blood
lead is compared across treatment groups at each round with no analyses of
effects between rounds.
some longitudinal structure, but these analyses typically do not include assessment of changes
in blood lead and its environmental covariates.
With cross-sectional structural equation modeling, the response variable Y/r depends on
covariates Zir and Xir, expanding Equation 5-5 to the form
Yir = Ggr H- Xt B + ZF + e (5-7)
gr
+ eir
where Xir is related to Zir, or other covariates represented as Wir, in the following manner:
+d (5-8)
_
5-12
-------
Cgr is the covariate group effect for group g and round r
Z^ is the second covariate, interacting with Xir and also directly affecting Yir
Wir is the third covariate interacting with Xir but not directly affecting Yir
Dgr is the regression coefficient for Zir, and
dir is the measurement error term for covariate X,- at round r
Note that Zir has both a direct and indirect effect on Yir. The direct effect is as a component
of the expression ZfrF«- in Equation 5-7, and the indirect effect is as a component of X/r in
Equation 5-7, as shown in Equation 5-8. With this method, Equations 5-7 and 5-8 would be
solved simultaneously hi an iterative manner to estimate the value of the regression
coefficient that provides the best fit for the combined system of equations. In performing
these calculations, we used several structural equation model calculation algorithms to obtain
the best fit. These algorithms are discussed hi Section 5.6.
The output of the cross-sectional structural equation model provides several types of
treatment group response sizes:
Response to treatment group (base) Glr - G2r
Difference between covariate intercepts Clr - C2r
Difference hi response to covariate X Blr - B2r
Difference hi direct response to covariate Z Flr - F2r
(5-9)
(5-10)
(5-H)
(5-12)
There is also a measure of the total (direct and indirect) relationship of Z to Y, expressed
as
Fgr + BgrDgr (5-13)
and a difference hi total relationship of Z to Y, expressed as
(Fir + BA) - (F2r + B2rD2r)
(5-14)
The output from this model is an estimate of the transfer from one compartment to
another (e.g., for soil to dust, 0.2 /*g Pb in dust per /tg Pb hi soil). Structural equation
models are normally visualized as boxes and arrows, with the boxes representing
environmental components, such as house dust, and the arrows representing transfer between
5-13
-------
boxes. This method establishes that specific pathways exist in the child's exposure
environment.
Cross-sectional structural equation models were used in the Cincinnati report and in this
assessment to examine the pre-abatement environmental lead exposure pathways. While the
population samples of the three studies were chosen for use in a longitudinal intervention
study, and not for the purpose of assessing baseline exposure hi the communities, the
baseline data nonetheless represent a useful snapshot of children in certain neighborhoods in
Baltimore, Boston, and Cincinnati. The limitations of this approach are that the study
populations may not have been a fully representative sample of the true neighborhood
populations, especially in Boston and Baltimore. The impact of these limitations is discussed
in greater detail in Section 5.4.
5.1.2.4 Longitudinal Analysis of Covariance
The next four methods are longitudinal analyses and, as such, are more appropriate for
these studies. These analyses evaluate changes in childhood blood lead and in environmental
lead exposure pathways subsequent to (and by inference, as a result of) soil lead or dust lead
interventions. Among the possibilities considered are: (1) blood lead decreases as a result of
a direct change in environmental exposure to soil; (2) blood lead decreases as a result of both
a decrease in soil lead and a decrease hi dust lead, where the decrease in dust lead load may
be by dust intervention or by the impact of soil intervention on dust lead load; (3) blood lead
decreases as a result of changes hi dust ingestion as a result of other interventions, such as a
decrease in dust loading (cleaner house means less dust ingested) or changes hi the child's
behavior (parental education) that decrease dust ingestion; and (4) blood lead changes as a
result of factors unrelated to intervention, including growth and normal changes hi child
behavior.
Longitudinal analysis of covariance (Figure 5-4) is similar to cross-sectional analysis of
covariance hi that it has fixed treatment groups with continuous or categorical covariates, but
the response variable is also adjusted for the blood lead concentration from the previous
round (Y ir_j). Figure 5-4 is similar to Figure 5-2, but with an additional arrow to show that
blood lead is now controlled or adjusted for the preabatement blood lead concentration in the
5-14
-------
Group A
Group B
Before
After
Figure 5-4. Longitudinal analysis of covariance. The arrow from one child figure to
the next shows that blood lead at the postabatement round is adjusted for
preabatement blood lead for each individual child. Blood lead is adjusted
for covariates at each round, and for blood lead at a previous round, and
then compared across treatment groups at that postabatement round.
same child. The implicit hypothesis is that for each child, the postabatement blood lead
depends on the child's preabatement blood lead on on the current (postabatement) lead
exposures from soil and house dust. This is somewhat analogous to some models used in the
Boston report.
gr
(5-15)
This is the first of this sequence of methods that takes into consideration the before/after
effect of a longitudinal study with intervention, and is especially important to this assessment
because it takes into consideration the store of blood in body tissues, such as the bone
storage discussed hi Section 2.3.4 and illustrated in Figure 2-4. The advantages of
longitudinal analysis of covariance is that this method does not require the use of a simple
5-15
-------
difference between treatment groups in the response variable, but explicitly includes
pretreatment levels of response. Similar to cross-sectional analysis of covariance, however,
this method also assumes the pretreatment level of response is perfectly known and that the
covariates are known without measurement error (see Section 5.1).
The units of the response term are expressed as a ratio of the response to the covariate.
For example, the Boston group used this method to estimate a range of response to
intervention as being a decrease of 0.7 to 1.2 /xg/dL for each incremental decrease of
1000 pg/g Pb hi soil during the first phase of their study, and a somewhat larger decrease
during the second phase.
5.1.2.5 Repeated Measures Analysis of Variance
Two of the remaining methods are repeated measures analyses, and one is a structural
equation modeling method. Repeated measures analyses evaluate whether intervention
affected the child's blood lead, whereas structural equation models assess whether this
change can be attributed to the soil-dust-blood pathway. The two approaches are
complementary, especially for a longitudinal intervention study.
Repeated measures analysis of variance contains a group effect term (Gir), a household
effect term (H,,fep, and an individual child effect term (t,-&/y), in addition to the error term.
= G
gr
ir
(5-16)
5.1.2.6 Repeated Measures Analysis of Covariance
Similarly, repeated measures analysis of covariance adds the covariate term X,y Egi that
first appeared in Equation 5-5.
Yir = Ggr
H
h(g)
ir
(5-17)
5-16
-------
In both of these repeated measures examples, a distinguishing feature is the individual
intercept term, li(gh), that gives greater control over sources of variability, including the
household effect.
The advantage of repeated measures analysis of variance or of covariance, illustrated hi
Figures 5-5 and 5-6, is mat these methods both allow the complete use of the embedded
correlation structure within the data set for multiple treatments and multiple rounds.
Figure 5-5 is similar to Figure 5-1, and Figure 5-6 is similar to Figure 5-2, but with an
additional arrow from the child figure preabatement to the same child figure postabatement to
show that the statistical relationship of postabatement blood lead to preabatement blood lead
for each individual child has been included in the model. There is thus a sense hi which
each child serves as his or her own control, an analytical strategy which frequently increases
the ability to detect treatment group differences in changes in blood lead over time. This
approach also makes no requirement that the change hi individual blood leads between rounds
is the most appropriate index for characterizing change hi group mean effects. These
methods are used hi this report.
In addition, repeated measures analysis, for either method, allows inter-individual
differences to be modeled as a random source of variation, which reduces the problem of
whether the pretreatment response is measured perfectly, as hi longitudinal analysis of
covariance. Repeated measures analysis of variance makes no adjustment for tune varying
covariates. This limitation does not apply to repeated measures analysis of covariance,
although, hi using tune varying information, the covariate is once again assumed to be
known without measurement error.
The longitudinal structural equations model (LSEM) is illustrated hi Figure 5-7. This
combines the features of all of the earlier models. The individual child blood lead at each
round is adjusted for soil lead, dust lead, and other covariates that describe exposure at that
round, as well as for the preceding blood lead concentration. This allows inferences about
effects that are attributable to changes hi soil lead and dust lead concentration, effects that
are attributable to changes in soil or dust exposure after abatement, effects that are
attributable to changes hi the soil-to-dust-to-blood indirect exposure pathway, and other group
effects that are not specifically attributable to changes hi concentration, exposure, or
5-17
-------
Group A
Group B
Before
After
Blgure 5-5. Repeated measures analysis of variance. The arrow from one child figure
to the next shows that blood lead at the postabatement round is adjusted
for preabatement blood lead for each individual child. Blood lead is
compared across treatment groups across different rounds for children in
the study at each round.
transport of lead from soil. The LSEM methods in this report are used in order to evaluate
some of the possible mechanisms by which soil abatement may operate, which was not
required in the individual study reports. The LSEM analyses carried out here also allowed
evaluation of the sensitivity of the repeated measures ANOVA and ANCOVA analyses to an
alternative modelling approach.
5.1.2.7 Longitudinal Structural Equation Models
Longitudinal structural equation model equations (Figure 5-7) are similar to the cross-
sectional form for structural equation models, with the addition of terms for the influence of
previous rounds, expressed with subscript r-1:
Y - G + X + ZF + Y_A + e (5-18)
gr
eir
5-18
-------
Group A
Group B
Before
After
Figure 5-6. Repeated measures analysis of covariance. The arrow from one child figure
to the next shows that blood lead at the postabatement round is adjusted
for preabatement blood lead for each individual child. Blood lead is
adjusted for covariates at each round, and then compared across treatment
groups across different rounds of the study for children in the study at each
round.
_ (**
(5-19)
The advantage of this method is that the full specification of the longitudinal and cross-
sectional pathway relationships are permitted. The model adjusts for the simultaneous fitting
of multiple relationships, but may not fully correct for covariate measurement error.
This summary of seven statistical methods is meant to place into perspective the
detailed analyses that were performed independently on the data of the three studies. The
basis for analyses of abatement effectiveness is comparison of differences in mean blood lead
between groups of children who received different interventions. Figure 5-8 sketches the
probable outcomes of a generalized intervention study.
5-19
-------
Group A
Group B
Before
After
Figure 5-7. Longitudinal structural equation model. The arrow from one child figure
to the next shows that blood lead at the postabatement round is adjusted
for preabatement blood lead for each individual child. Blood lead and dust
lead are adjusted for covariates simultaneously at each round to eliminate
simultaneous equation bias. Postabatement blood lead is adjusted for
preabatement blood lead.
5.1.3 Specific Problems with Statistical Methods
A potential problem arises in simple comparisons of group mean values during a
longitudinal study when different individuals are present at different phases of the study. For
example, some individuals in the preabatement phase of the study may have dropped out by
the time of the postabatement phase, whereas other individuals who were not in the
preabatement phase may have been recruited into the postabatement phase (e.g., infant
siblings who reached enrollment age status during the study). Although it would be
reassuring to think that attrition and recruitment do not depend on the treatment group, and
that children lost or gamed during the progress of the study are no different from those
enrolled throughout the study, this cannot be guaranteed. One of the simplest solutions is to
limit the analyses to children who were present during all key phases of the study.
5-20
-------
1r
Round 1
Round 2
Figure 5-8. Schematic representation of expected outcomes for treatment and control
groups. The intercept, Ggr, from any of the longitudinal methods can be
used to calculate the between round effect size, Er, as the difference
between the AG's for each group.
When the analyses are restricted to subjects with both pre- and postabatement data, then
response to abatement may be assessed by simply taking differences of blood lead
concentrations or differences of their logarithms. Unfortunately, blood lead differences
ignore the intrinsic persistence of blood lead concentrations over tune (see Figure 2-4). The
only part of the preabatement blood lead concentration that can be reduced by intervention is
the .nonpersistent part,
removable blood lead = preabatement blood lead — persistent blood level
where the persistent fraction at one year postabatement may be about 40-60%. The
difference between preabatement and postabatement blood lead cannot be larger than the
amount of removable blood lead. In theory,
5-21
-------
preabatement — postabatement blood lead < fraction of preabatement blood lead.
This suggests that a better index for abatement effectiveness might be a partial difference:
postabatement — (1 — fraction) preabatement blood lead > 0.
Unfortunately, the value of this fraction is not known well enough to define a priori the
partial difference for use as an index of lead effectiveness, because the value of the retained
fraction of lead depends on the tune since abatement, the child's age, and probably other
factors including the initial blood lead level.
Even though statistical models could be based on the partial differences of blood lead
levels between pre- and postabatement phases, the environmental exposure variables are
themselves more or less correlated with earlier measurements of the exposure variables.
This violates one of the most important assumptions about linear regression models, and
generally about linear models such as the analysis of variance and the analysis of covariance.
That assumption is that the predictor variables or regressors are known without statistical
error. Although the statistical error is usually called "measurement error" (Fuller, 1987), the
errors include many other kinds of variability. In environmental epidemiology, the most
common measurement errors in exposure include behavior or activity pattern variability,
repeat sampling variability, sampling location variability, as well as analytical error. That is,
the observed value of the predictor, such as floor dust lead loading, may not perfectly reflect
the activity of the child and the child's actual exposure to dust lead over time.
One way to deal with this is to predict the precursor exposure variables in an
environmental model. For example, suppose that blood lead is predicted by hand lead, soil
lead, dust lead, and by a preceding value of the blood lead. Hand lead may then be
predicted by current dust and soil lead levels, and dust lead by current soil lead, so that in
addition to the direct effect of soil lead on blood lead, there are indirect effects from soil to
dust to hand to blood, and from soil to hand to blood. This approach allows estimation of
the measurement error variance hi the precursor lead exposure variables in terms of residual
deviations between the observed exposure variable and its best estimate from its own
precursors. If the model is correct, this approach will essentially eliminate the bias
5-22
-------
introduced by measurement errors. The usual bias in estimating a regression coefficient or
effect size of intervention will be to attenuate the estimate (i.e., to shrink the estimate
towards zero, which reduces both its magnitude and its statistical significance). However,
with multiple correlated predictors such as lead soil and dust variables for a single residential
premises used in these analyses, this attenuation may not occur (Klepper et al., 1993).
The application of many hypothesis tests to the same set or subset of data may greatly
distort the overall significance level of the entire decision-making process. This problem of
multiple comparisons can be controlled by testing only hypotheses that are specified hi
advance. Because tests of some hypotheses may depend on the results of preceding tests on
the pooling of certain groups within studies, the exact number of tunes that each data set is
used in a test cannot be stated, but is usually not more than six tests. An extremely
conservative approach is to assign experiment-wise significance at level alpha (for example,
alpha = 0.05) only to those tests whose individual test-wise significance is at level alpha /
(number of tests). That is, to assert that all of the results of six tests involving the same data
set are significant at level 0.05, each test should be carried out at level 0.05 16 = 0.0083.
Some authors argue that mis adjustment, which is called the Bonferroni correction, is
exceptionally conservative and that no adjustments are needed for multiple comparisons
(Rothman, 1990). P levels are provided for each test to assist the reader who wishes to form
his or her own judgements about the meaning of the results of the analyses. The decision
level alpha of any statistical test is a subjectively chosen number. For most users of these
tests, the conventional choice of alpha = 0.05 with the conservative decision to use an
experiment-wise Bonferroni adjustment based on five tests per group per variable would
suggest a test-wise level of 0.01 in order to decisively reject the hypothesis of no change,
difference, or effect.
5.2 DIFFERENCES IN GROUP MEANS
Before presenting the detailed statistical analyses, the raw data are presented in two
nonstatistical forms that give the reader a visual perception of the data sets. The first form is
a series of group mean plots that show box and whisker diagrams of several distribution
parameters (arithmetic mean, median, quartiles, etc.), by round, for soil, dust, hand wipes,
5-23
-------
and blood lead independently. In the second form, blood lead data alone are plotted by
round and age for each individual child. This method gives some perception of individual
differences in blood lead concentrations and individual differences in response to
intervention. Each of these two visual presentations has limitations that should discourage
a
the reader from drawing conclusions about the impact of intervention. These limitations are
the same whether blood leads are characterized by the group mean, geometric mean, median
or other percentile values. The first is that some of the children hi any treatment group are
probably not exactly the same children at one phase of the study as at a subsequent phase.
Some children will almost certainly be lost to follow-up by moving or by refusal to
participate (normal processes of attrition in longitudinal studies), whereas other children may
be added by recruitment (such as at Round 3 in the Baltimore study) or as additional
members of households where other children are already enrolled hi the study. Since
children who are lost to follow-up or who are added to the study may differ hi some
systematic ways from children who were retained throughout the study, it may be prudent to
analyze data from these children who were not present separately from those who were
present at all relevant phases. On the other hand, if study results are restricted only to
children who were present at certain specific pre- or postabatement phases of the study, then
repeated measurements on the same child at different phases of the study are not statistically
independent of each other. Although data from one treatment group at a given phase are
independent of data from a different group, data on the same group at a different phase are
not independent of data from an earlier phase.
In the Boston study analyses, the same subset of children was used here as in the
Boston report, excluding the same two children who had become lead-poisoned. For the
Baltimore data, the small group of participants from the treatment group whose properties
were not abated were assigned to a separate control group, rather than merging them with the
main control group. The Cincinnati neighborhoods are treated here as individual study
groups and include all children recruited (both rehab and nonrehab), except for the four
children who were undergoing treatment for lead poisoning.
The presentation of these group mean data, illustrated hi Figure 5-9, uses a similar
format for all of the figures hi this series. Each treatment group is represented in each round
by a box and whisker plot. Each box has a mark approximately midway that shows the
5-24
-------
84th Percentile
-_ 75th Percentile
Arithmetic Mean
Median
25th Percentile
16th Percentile
Figure 5-9. Hypothetical representation of common statistical parameters for a single
group and a single round.
median value for the group, and these medians are connected by a line between boxes.. The
upper and lower ends of the box mark the 3rd and 1st quartiles (75th and 25th percentiles)
respectively. The tick marks on the upper and lower whiskers show the location of the 84th
and 16th percentiles, respectively, these being two statistics useful for estimating geometric
distributions. The diamond on the line or hi the box shows the location of the arithmetic
mean.
5.2.1 Changes in Mean Soil Lead Concentrations
In order to form an effective, permanent barrier between the lead in soil and the human
environment, soil abatement must reduce the concentration of lead hi the soil hi a manner
that is persistent for a period of years. In each of the three studies, measurements of soil
were made prior to abatement and immediately after abatement (within three months).
Followup measurements were made periodically until the end of the study hi Cincinnati and
Boston. The results of these soil analyses are graphically illustrated hi Figures 5-10, 5-11,
5-25
-------
and 5-12. These data represent the percentile calculated from the individual parcel means of
several individual soil samples. They show, for all three studies, a substantial reduction in
the amount of lead hi abated soil areas. In Boston and Cincinnati, where follow-up soil
measurements were taken, this reduction persisted for the duration of the study. In
Baltimore, the postabatement measurements were made only in the locations where soil had
• been excavated and removed.
Each study was able to achieve the targeted concentration for abated soil. The median
soil concentrations following abatement are not substantially higher than the specifications for
clean soil. The amount of soil lead reduction actually achieved directly influences the
expected changes in dust lead and blood lead. Soil lead concentrations vary widely over
relatively small distances. Because it was not feasible to return to the exact spot each time
for sequential soil samples, two sequential samples from the same plot may vary widely.
The median of the soil parcel means for the Boston and Cincinnati studies show that
abated soil concentrations [BOS SPI, CIN SEI(P), CIN I-SE(D), and CIN I-SE(F)] dropped
substantially after abatement (Figures 5-10 and 5-11) whereas unabated soil (BOS PI, BOS P,
and CIN NT) showed virtually no change.
5.2.2 Changes in Exterior Dust Concentrations and Loadings
In Cincinnati, exterior street and sidewalk dust concentrations remained relatively
constant throughout the study (Figures 5-13 and 5-14) and are much higher than the soil
concentrations, suggesting a source or sources with higher lead concentrations than soil that
mix with leaded dust from soil to form exterior dust. A possible conclusion is that sources
of lead in exterior dust, other than soil, impacted each neighborhood or groups of
neighborhoods, differently. This is reasonable because the neighborhoods are geographically
separated. Five of the neighborhoods (Back, Dandridge, Findlay, Mohawk, and Pendleton)
are nearly contiguous and lie hi a larger neighborhood known locally as "Over-the-Rhine".
The sixth neighborhood, Glencoe, lies approximately Vfc mile away. Interpretation of the
spatial distribution of the Cincinnati data within each neighborhood is not possible without
more information on the location of the dust samples.
For Boston and Baltimore, the question arises that there may also be external sources of
lead other than soil that contribute to household dust and to the exposure of children during
5-26
-------
10,000
•3 1,000
100J
10
BOS SPI
10,000
Oct89
Apr 90 JUI91
an 90 I
1,000
100
Soil & Dust
Abate
10
Get 89
B
BOS PI
Apr 90
Jul91
Dust
Abate
Soil
Abate
RD1
RD 2 RD 3
Sample Round
10,000-1
RD4
RD1
RD 2 RD 3
Sample Round
RD4
•3 1,000
1
8
A
= 100
o
V)
10
BOSP
Oct89
Apr 90
Jul91
Soil
Abate
RD 1 RD 2 RD 3
Sample Round
RD4
Figure 5-10. Boston soil lead concentrations (on a log scale) by study group show the
effectiveness and persistency of soil abatement. Note the decrease in soil
lead concentrations (RD 2) immediately post soil abatement and persisting
through RD 2, RD 3, and RD 4 for BOS SPI Group (Panel A); no soil lead
sampling in RD 2 for other two groups (BOS PI-S and BOS P-S); RD 3
values for those two groups similar to their RD 1 soil lead concentrations;
and the later marked decrease in their RD 4 soil lead values following soil
abatement after RD 3.
5-27
-------
i
e
0
V
XI— | |
a'
IHI
5
•HI V
? \
« \
| HI \
? HM
3 \
|Hi i
/
K H
\
^ ^-|
<: si-
ZH
ZH
|^| inn ill j
ZH
«o
g
in
•g
i
• Q
e
CM
•g
I
CQ
§ |
I *
(B/Brt) uon»JJUK)Uoo qd DOS
I D
1 t-n—H
LLI
ID
g
U>
g
gi
8
i !
(D/Bit) uonMiuaauoo qd l\os
II
•§ i
11
It
.S
•g
ii
a |
• «
SI
n ^
t
11?
4> O
•» 8
^
5-28
-------
1,000-1
100-
10-
BALSP
RD1
RD4
1,000-
100-
10-
1-
B BALP-1 1'00°-
' 100-
10-
H
C BAL P-2
• . I 1 I
RD1
RD1
Figure 5-12. Baltimore soil lead concentrations (log scale). Follow-up measurements of
soil were taken for BAL SP only, and show a significant decrease in
response to soil abatement.
outside activities. Because there were no measurements of exterior dust in these studies,
little evidence is available to accept or reject this hypothesis. However, in the context of
exposure pathways, the parcels of soil in Boston and Baltimore were on the individual
properties, whereas in Cincinnati, most soil parcels were in neighborhood areas separated
spatially from the living units, such as parks and vacant lots.
5-29
-------
0
§
8"
g
00
S £
<= z
L
•as
it
,- s
o fc
LU
CM
. Q
(6/Bil) UOHBJJU90UOO Qd isna leans
s
K S
»!
ij
CM
g
.s
S
h -a
55 «
it
-8
t
1
8
(6/Sri) uo|JtJJU»DUOO qj )snQ )8OJ;S
5-30
-------
E
iu
§
!
CO
£
I
. a z
f Inez
H W
Z ! 0
o SH »
JHC
|HH=
w
=H
IH
^H
=H
3— 1
|H~n~H
1 I
1 HltH
8
HI=H
= i-rn—ui
" u
Sj
CM
O
(B/Bt1)uomwiaouooqdlsna>l|BMapis
S
'S
C1
^g
1
1
p
"S
«
Jl
a* ^
fc
"•< a
^
-------
5.2.3 Changes in Interior Dust Concentrations and Loadings
Interior dust is measured in both concentration and surface loading. Concentration is
measured in micrograms of lead per gram of dust, whereas loading is measured in milligrams
of lead per square meter. When dust abatement is performed, the amount of dust changes,
but the concentration of lead hi the dust does not. Therefore, there should be no change in
dust lead concentration unless the source of the dust changes. Where soil abatement has
been performed in connection with dust abatement, the dust lead concentration should also
decrease abruptly if the soil is the major component of the dust. If there is a mixture of dust
sources and only one has been abated, the lead concentration would change less abruptly,
according to the contribution from each source.
The data for the Boston study ulterior dust measurements are shown in Figures 5-15
through 5-20. The high concentrations of lead in individual measurements of window well
dust (5,000 to 22,000 /tg/g) indicate the possible presence of lead-based paint (Figure 5-15).
The Cincinnati study (Figures 5-21 through 5-23) found relatively constant dust lead
concentrations during the first year (Rounds 1-4). Data are not available for Rounds 5-7.
Data for window wells are shown in Figures 5-24 through 5-26 and entry ways, Figures 5-27
through 5-29. The window well concentrations were lower in Cincinnati (1,000 to
2,300 /tg/g) than in Boston, suggesting a minimum influence of lead-based paint.
5.2.4 Changes in Hand Dust Lead Loadings
Because hand-to-mouth activity is one route by which lead may be ingested, the amount
of lead on the child's hand is an indicator of exposure. The hand wipe data for Boston are in
Figure 5-30, for Baltimore, in Figure 5-31, and for Cincinnati, in Figure 5-32. Only lead
loading information is available because there is no measure of the amount of dust removed.
The units of measurement are micrograms per pair of hands rather than micrograms per
square meter. This is an important link in the exposure pathway that measures actual contact
with the child's dust environment. Hand lead loadings were expected to respond more
quickly to environmental changes than blood lead concentrations.
5-32
-------
100,000]
10,000
.0
a
I
u.
1,000
100
BOSSPI
Oct89
10,000
Apr 90
-*~ Sep 90
JuI91
1,000
Soil & Dust
Abate
.100
B
BOS PI
Oct89
Sep 90
Apr 90 _ Ju|91
Dust
Abate
Soil
Abate
RD1 RD2 RD3
Sample Round
100,000
RD4
10,000
c
8
Q 1,000
100
RD1 RD2 RD3 RD4
Sample Round
BOSP
Sep 90
Jul91
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
Figure 5-15. Boston floor dust lead concentration. While dust abatement alone may
temporarily reduce the total dust lead loading (see Figure 5-14), it may not
change the concentration of lead in any remaining dust.
5-33
-------
1,UOU
w
t Load (mg/m
i
§
Q
! 10
E
•i.
A BUS SHI
°^L89 Apr 90 T-
rn T"
• J_ Sep90 -L
*^-^ /^ J
LJ ^v^ td
J_
I
Soil & Dust
Abate
•
1,000
RD 1 RD 2 RD 3
Sample Round
1,000-j Q
RD4
10
BOS PI
°«_89 Apr 90 JU|91
I s°
Dust
Abate
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
BOSP
Floor Dust Load (mg/m ^
0 §
Oc
:t
i
*
I
Ar
B9
>r
T
^
z
90
Se
Pi
»
bU
I
JL
90
119
I
RD1
Soil
Abate
RD 2 RD 3
Sample Round
RD4
figure 5-16. Boston floor dust load Gog scale). The absence of a decrease following
interior dust abatement in the BOS SPI and BOS PI-S groups suggest that
house dust loadings may be replenished back to preabatement levels hi a
tune period shorter than the interval between Round 1 and Round 2.
5-34
-------
10,000.0
1,000.0
I
100.0
10.0
1.0
0.1
BOSSPl 10.000.0i
1,000.0
Oct89
— Apr 90
Jul91
Soil & Dust
Abate
RD1 RD2 RD3 RD4
Sample Round
10,000.0-1
1,000.0-
100.0-
10.0-
1.0-
0.1
Oct89
100.0
10.0
1.0
0.1
B
BOS PI
Oct89
• SepSO
Apr?0 _ Ju|91
Dust
Abate
Soil
Abate
RD1
RD2 RD3 RD4
Sample Round
BOSP
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
Figure 5-17. Boston floor dust lead load (log scale). Even though the dust load in
Figure 5-16 indicates a quick recovery, the lead load did not recover
immediately, indicating that the source of the lead was cut off, at least
temporarily.
5-35
-------
1,000,000
100,000
10,000
1,000
100
10
BOSSPI
Oct 89 Apr90
Sep 90
Jul 91
1,000,000
100,000-
10,000
1,000
100
Soli & Dust
Abate
10
B
BOS PI
Oct 89 Sep 90
Apr90
Jul91
LJ I
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
1,000,000
5! 100,000
D)
10,000
£! 1,000
to
100
10
RD1 RD2 RD3 RD4
Sample Round
BOSP
Oct 89 Sep 90
Apr 90
- I
Jul 91
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
Figure 5-18. Boston window dust lead concentrations (log scale). Paint stabilization and
soil abatement appear to have been effective and persistent for several
hundred days, similar to floor dust. The recovery observed between April
and July 1990 was not observed for the floor dust load data.
5-36
-------
10,0001
1,000 Oct89
100
10
10,0001
1,000
Soil & Dust
Abate
RD1 RD2 RD3 RD4
Sample Round
10,000i
1,000
i
E
100
10
C
Apr 90
Out 89 X
100
10
B
BOS PI
Sep90
AprSO ^ Ju|91
Dust
Abate
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
BOSP
Jul91
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
Figure 5-19. Boston window dust load (log scale). These data show the effectiveness of
window dust abatement, which appears to recover after about 150 days,
similar to floor dust loads observed in Figure 5-16.
5-37
-------
1,000,000.0
100,000.0
j| 10,000.0
a
1
2 1,000.0
TJ
§
2
I 100.0
Q
1
£ 10.0
1.0
O.t
BOS SPI 1,000,000.0
r\
100,000.0
Jul91
Oct 89 Sep 90 p.
pr Apr90 oc
B gf n __ _ 10,000.0
-^ 1,000.0
100.0
10.0
1.0
Soil & Dust
Abate
n BOS PI
Oct89 Sep 90
* Apr90
J _n^ J~ "s
y Q N
Dust soil
Abate Abate
RD1 RD2 RD3 RD4
Sample Round
1,000,000.0
100,000.0
^,
1
d.
¥
CO
5
10,000.0
1,000.0
Oct 89
J Apr90
•^
r
T =
100.0
10.0
1.0
Sep 90
RD1 RD2 RD3 RD4
Sample Round
BOSP
Jul91
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
Figure 5-20. Boston window dust lead load (log scale). As with floor dust lead loads,
the window data indicate that both paint and soil sources of lead were
interrupted, at least temporarily. The data appear to be consistent with
Figure 5-17.
5-38
-------
£
I
3
lldfH
O
tH3Z»
5
Incb
1
g
i
i
1H3DH
IDEM
00
WH
UJ
1
o
IhCCH
IHEP
• HI 3
IHCEH
|HffO
|HIJ
IHEJI
I HEEH
(B/Bit) UOBBJJUBOUOO qd Wtna Joou
(B/6ti) uonaJtuaauoo qd isna Jooid
p -a
li
is
-------
o
1
EE
g
1
!>-i
T
fttzp
»—it r -H
111
:c
O
g
g
O
g
Ul/Biu) p«n W Jooid
5-40
-------
o
I I
ui
2
CD
IHEZ3ZZH
ra a
8 f
t
LLI
• o
E
-Q
E
n
•§
M
g
ii-n—r
m
•o
E
N
•O
^ u^sii) p«n pwi wxi «»u
ui/Bii) PBO-] pran ;«nQjoo|d
5-41
-------
CQ
s
/
ui
H—PU—I
I
a
•g
g
tt B
1 1
(B/Brt) uofwquaauoo Qd Una MOpuiM
ha
CM
g
3 S
1 -a
§ "a
•S10
1 §
gl
"8
la
ri
a
II
§
O
1
I
(B/Bri) uo|vuiu«3uoo qd l«na MOPUIM
••"
(Q Tj
J*
S -g
*PN A
5
5-42
-------
I
5
o
It
•g
§
1
I
QQ
111
. o
cc
8|
w »
§1
N
g
I §
<
\ 1
=« — 1
i 5
i — i
§ s
H
m CC
Si
-------
o
If I—I f I-H
8 I
Inn
iKD^IH
ltC=l
^h-HE=H
CD
I
i"
h-I / H
H—r
& o ^
*"
e e e e e e
I I I S '
*{.*•*"
*• (EUI/Bll)|MOnpMlI«X]MOpU|M
s
i
0)
It
§
i
0> "O
2 at
b* v-
IS
to
.s
ve
«s
V)
5-44
-------
e
!
i
o
§
2
§
00
I-
Q
(D
o
i
• a
DC
I
I
Q
ec
to
a
te
. a
cc
tfi S i
I §
(B/Bil) uojiBqufauoo qd 18"O Aqug
3
(B/Bil) uoivuuiaauoa qd Itiia Aqug
5-45
-------
o
o
\
§
LLI
01
o o
cc •§.
m 1
|§IS
o-
^ UJ/BUJ) p«n wna AWH
r.
o
(C
-------
o
§
2
§
m
I
il
I
g
LU
o?
* S
«r £
I
S.
= «•
£
§
e
g
ii
II!1
8 r
^
{ ui/Bii) peon pee-| wna Aiiua
•g
-------
40
35
30
f 25
1
s. 2°
c
X 15
10
5
A
BOS SPI Jul
Apr 90
Oct89
I Jan 90
I
^»
I
—
be
Soil & Dust
Abate
•
^f^
^^
4
7*
**
T- •
_L
91 40
35
30
. . 25
20
1R
iv
10
5
0
BOS PI
° Jul 91
Apr 90
Jan 90 I -
Oct89 ~T [J ^^^
-r-
JL _L
Dust Soil
Abate Abate
° RD1 RD2 RD3 RD4 RD 1 RD2 RD3 RD4
,
Sample Round Sample Round
40| BOS p
35
30
•§ 25
0.
S
£ 20
•n
^ 15
10
5
c
Apr 90
-i- Jul 91
°?L89 Jan 90 f
I
—
-
p: ^-
/ I
_-^— ~~
I
?
Soil
Abate
RD1 RD2 RD3 RD4
Sample Round
Figure 5-30. Boston hand lead load.
5-48
-------
50
45
40
35
I 3°
J 25
.n
Q.
•g 20
£
15
10
5
0
BAL SP 50
A
45
40
35
30
Jun 91 25
T Sep 91
Apr 89 T 20
Oct88 T _L Hi
T Jan 91 - 15
pJ-i pi Feb 90 T —
-^~-^p 1 / !-,-[
JL T LJSollM ^- 5
ii
R BALP-1
Jun 91
Apr 89
n Sep 91
Oct88 r1] . T
-T- Feb 90
w " f\ I-
"* ^ Jan 91 / ^^_
U ^[KJ7^ Q
RD1 RD2 RD3 RD4 RD5 RD6 RD 1 RD2 RD3 RD4 RD5 RD6
Sample Round Sample Round
50
45
40
35
f 30
i
3 25
£
•g 20
15
to
5
0
C BALP-2
Apr 89
IJUI
Feb 90 f
Oc*88 M j Jan 91 *
-^~^i rL/:
Sep 91
191
] 1
^ s^ * to
L ^
^U VQ ^
RD1 RD2 RD3 RD4 RD5 RD6
Sample Round
Figure 5-31. Baltimore hand lead load. There were no sequential measurements of Baltimore house
dust to compare with the hand lead load.
5-49
-------
E
O
S
S
}>
-HE
8
I C
HZTIH
I
g
»l — 1 t 1— 1
'
S
21 1 l ^ 1-
* \
H
s\
>H~T
0 ' l—J-
S
1 <— 0=
w
g
< l1^
~H
__> 1
Hi
ili-
H
g
E
«
• a
ce
m
o
oe
01
g
g
gj 8 S S
s i? §
Sin o to o
ti <& i- i-
u) o u) e
z
o
S
h-QEH
li-tt:
H3D-
1
i
g
s a a s
(JlBd/Bri) Qd PUSH
•a
O
a
5-50
-------
5.2.5 Changes in Blood Lead Concentrations
5.2.5.1 Baltimore Study Blood Lead Data
The blood lead concentrations for the three Baltimore groups are shown hi Figure 5-33.
The data are for all children participating hi the round. They show that the groups were
similar prior to soil abatement, but no clear difference between groups hi response to
intervention. There is a moderate indication of a seasonal cycle comparable to patterns
reported hi other longitudinal studies, as discussed hi Section 2.3.1. The lack of
postabatement measurements of soil and house dust limits the ability to interpret these data
by more than a simple analysis of variance.
5.2.5.2 Boston Study Blood Lead Data
The blood lead concentrations for the Boston study are shown hi Figure 5-34, where
they graphically illustrate the conclusions of the Boston report, that intervention probably
accounted for a decrease of 0.8 to 1.5 /ig/dL in the blood lead.
5.2.5.3 Cincinnati Study Blood Lead Data
The wealth of information from the more detailed measurements of household dust in
the Cincinnati study presents a proportionally greater challenge to the modeling of dust
exposure pathways. The blood lead concentrations shown in Figure 5-35 show some
evidence for seasonal cycles.
5.3 PRE-AND POSTABATEMENT DIFFERENCES IN INDIVIDUALS
5.3.1 Individual Changes in Blood Lead and Soil Lead
Section 5.2 provides a visual presentation of longitudinal patterns hi population means
for specific parameters over the course of the study. This section presents information on an
individual child basis through the use of a series of simple line plots where the blood lead
concentrations are plotted by round, age, and study group.
Most children hi each neighborhood experienced some change in blood lead, either an
increase or decrease, during the course of the study. This change may be due hi part to
5-51
-------
30,
25
1 20
g
i 1S
i
jo
i 10
5
5
n
A
Oct 88
pi Apr 89
«J
""• •,
i \L
Feb 90 ju
T Jan 91
"*« ••
i y
Soil
Abate
. 1 ,
_--—
E
n!
r
•—
r
IML.S
)1
Se
.
r
t
30
25
20
91 15
10
X
5
— a
g BAL P-1
Oct 88
I Apr 89 Sep 91
-r Feb 90 Jun91 _
1 T Jan91 T I
^^-^ -
y -r x
RD1 RD2 RD3 RD4 RD5 RD6
Sample Round
RD1 RD2 RD3 RD4 RD5 RD6
Sample Round
30i
BAL P-2
25
"Si 20 Oct 88
1 15
i
8
£ 10
S
5
I
^
I
Apr 89
_._ Feb 90 Sep 91
I "T Jun 91
•^ JL, Jan 91 ^ X
^^
-
I
- pi-
1 ^?
!r L
RD1 RD2 RD3 RD4 RD5 RD6
Sample Round
Figure 5-33. Baltimore blood lead concentrations. There appears to be little difference
between study groups. Overall, there seems to be a seasonal cycle of the
type and magnitude discussed in Section 2.3.1.
5-52
-------
25
20
1S
o 10
JQ
Q.
BOS SPI
Apr 90
Oct89 Jul91
I
Jan 90
I I
i
25,
20
15
10
Soil &Dust
Abate
B
BOS PI
Oct89
I
Apr 90
Jan 90
I
I
Jul91
T
Dust
Abate
Soil
Abate
RD1 RD2 RD3
Sample Round
25
RD4
5
JO
a.
m
20
15
10
FID1
BOSP
RD 2 RD 3 RD 4
Sample Round
Oct89
I
Apr 90
Jan 90
i
i
Jul 91
T
Soli
Abate
RD1 RD2 RD3 RD4
Sample Round
Figure 5-34. Boston blood lead concentrations.
5-53
-------
e
o
g
8
*i
-i t
DH
S
H H
DH
/
-------
changes brought about by intervention, or to seasonal effects, age (see Figure 2-6), or
changes in exposure not related to intervention.
This type of plot is especially helpful to the reader in understanding the variability of
the measurements and the possible significance of patterns or clusters. They are designed to
show changes hi only one variable over tune, not the multiple interactions of several
variables. In Sections 5.4, 5.5, and 5.6 statistical techniques such as repeated measures
analysis and structural equation modeling are used to extract information from the systematic
variability using more appropriate methods for comparison than observed on these plots but
in the context of several variables interacting at the same tune.
Figure 5-36 shows the Boston Phase 1 blood lead observations for each individual child
in the three treatment groups broken out by age at the start of the study. Typical patterns
show a decrease in blood lead from Round 1 to Round 2 hi all groups, including controls,
probably attributable in part to seasonal winter decrease. However, hi the control group
(BOS P-S), there was also a substantial rebound or increase hi blood lead at most ages from
Round 2 to Round 3, probably attributable to seasonal summer increases. Decreases
followed by large increases are noted for most children in the control group BOS P-S, and to
a quantitatively greater extent for most children in the Phase 1 dust abatement group
BOS PI-S. The Phase 1 soil abatement group BOS SPI shows large decreases between
Round 1 and Round 2, similar to the dust abatement group BOS PI-S, but unlike the other
groups, many children hi BOS SPI show either continuing blood lead decreases between
Round 2 and Round 3, or at most very slight summer increases. Five exceptionally large
changes are noted. Two children hi BOS SPI suffered clear lead poisoning at Round 3, with
blood lead increasing to 43 and 35 jig/dL respectively, due to accidental exposure during
household renovation. One child in BOS SPI (Round 1 age 18 to 29 months) showed a
rather large blood lead increase between Round 2 and Round 3. One child hi BOS PI-S
(age 18-29 months) shov/ed a large and consistent decrease hi blood lead from Round 1 to
Round 2 to Round 3, possibly reflecting the effectiveness of ulterior dust abatement and paint
stabilization. One child hi P-S (age 42+ months)with blood lead 22 ^tg/dL also showed a
large decrease hi blood lead from Round 1 to Round 2 to Round 3, possibly reflecting the
role of interior lead paint stabilization as an intervention. We omitted the two children who
5-55
-------
Ages
9-17
Months
BOS P-S
BOS PI-S
BOS SPI
18-29
Months
30-41
Months
42+
Months
25
I20
^15
55 5
0
25
I20
f15
I10
m 5
0
25
I20
f15
^10
m 5
0
25
20
v 10
S 5
0
2
Round
2
Round
2
Round
1
2
Round
25
20
15
10
5
0
25
20
15
10
5
0
25
20
15
10
5
0
25
20
15
10
5
0
1
2
Round
2
Round
2
Round
2
Round
25
20
15
10
5
0
25
20
15
10
5
0
25
20
15
10
5
0
25
20
15
10
5
0
1
43
2
Round
2
Round
35
2
Round
2
Round
Figure 5-36. Line plots of blood lead for individual children in Phase 1 (rounds 1-3) of
the Boston study for each treatment group. Where an individual value
exceeds the vertical scale, the actual blood lead concentration is shown
near the point of exit, e.g., for two outliers having 35 and 43 jtg/dL blood
lead concentration.
5-56
-------
were lead-poisoned by an identified source from subsequent analyses, but retained all other
cases.
Figure 5-37 shows the results of Phase 2 of the Boston study, Rounds 3 to 4. The
group BOS PI-S that received dust abatement hi Phase 1 and soil abatement hi Phase 2
showed a substantial decrease in blood lead for most children; only two children at ages
18 to 29 months showed small increases from Round 3 to Round 4, and one child at age
42+ months showed a relatively large increase. By contrast, 10 children in the group
BOS P-S that received only Phase 2 soil abatement showed increases hi blood lead during
Phase 2, with 2 or 3 increases substantial. By contrast, half of the children hi the Phase 2
"control" group BOS SPI, which received Phase 1 soil abatement, showed blood lead
increases, 8 of which were substantial increases.
Figure 5-38 shows the Phase 1 results for the Cincinnati study. The three
neighborhoods on the left side, CIN I-SE(B) (Back St.), CIN I-SE(D) (Dandridge), and
CIN I-SE(F) (Findlay), received ulterior dust abatement. CIN NT(G) (Glencoe) and
CIN NT(M) (Mohawk) received no treatment. CIN SEI(P) (Pendleton), on the right side,
received neighborhood soil and street dust abatement as well as ulterior dust abatement.
Many children showed the typical seasonal pattern of higher blood lead hi summer (Rounds 1
and 4) and lower hi fall or whiter (Round 3). Several children hi Mohawk (a no-treatment
neighborhood) showed large increases, which was much less common in Glencoe, the other
no-treatment neighborhood. The other neighborhoods had some children with increased
blood lead, and some with decreased blood lead. Many of the increases occurred hi children
less than 18 months of age, with decreases more evident in children of ages 42+ months.
There was not any external basis for omitting children from subsequent analyses.
Figure 5-39 shows the Phase 2 results for Cincinnati. There is even less pattern for
Phase 2 than for Phase 1, as was subsequently verified by detailed analyses.
Figure 5-40 shows the results for the Baltimore study. Soil abatement was carried out
in group BAL SP a few months after Round 3. While there were substantial decreases hi
blood lead in some children hi group BAL SP after Round 3, there were many who did not
show any decrease. Two children hi BAL SP (ages 18-29 and 54-65 months) showed very
large increases between Rounds 3 and 4, and one child age 30-41 months showed a large
increase between Round 4 and Round 5. A number of children hi the Area 2 control group
5-57
-------
Ages
9-17
Months
BOS SPI
BOS P-S
BOS PI-S
18-29
Months
30-41
Months
42+
Months
25
I20
J 15
I10
5 5
0
25
I20
H
10
QQ 5
o
25
|20
¥15
^ 10
S 5
0
25
I20
o
m 5
Round
Round
Round
25
20
15
10
5
0
25
20
15
10
5
0
25
20
15
10
5
0
25
20
15
10
5
0
Round
Round
Round
25
20
15
10
5
0
25
20
15
10
5
0
25
20
15
10
5
0
25
20
15
10
5
0
Round
4
Round
Round
Round
Round
Round
Figure 5-37. Line plots of blood lead for individual children in Phase 2 (Rounds 3
and 4) of the Boston study for each treatment group.
5-58
-------
GIN l-SE(B) CIN I-SE(D)
GIN I-SE(F) CIN NT(G)
40r
CIN NT(M)
CIN SEI(P)
1 3
Round
1 3 4
Round
Figure 5-38. Line plots of blood lead for individual children in Phase 1 (Rounds 1-4) of
Cincinnati for each treatment group and four age groups.
(BAL PI) showed large increases, particularly between Round 4 and Round 6, but many also
showed large decreases in blood lead after Round 3, particularly at ages 18-29, 54-65, and
66+ months. The Area 1 control group BAL P2 showed no dramatic changes in blood lead.
Note that the Baltimore cohort was typically much older at Round 3, the last pre-abatement
round. While there were a few children less than 18 months in the study, they were mostly
in group BAL SP and are not shown hi Figure 5-40. No children were omitted from
subsequent analyses based on external evidence.
5-59
-------
Ages CINI-SE(B) CINI-SE(D) GIN I-SE(F)
CIN SEI(P)
3
J30
Izo
a 10
•-^
4U
30
20
10
n
"~~~^\
_
•HI
30
20
10
n
-^^
'•—• — ="^
30
20
10
n
=*~~_ a — "_
<^^>
^^^
30
20
10
0
___
irrr — —
487 487 467 467 467
Round Round Round Round Round
1*
i"
«J
30
^^
===— =
30
20
^^
--5S--J-
30
20
10
30
20
10
\_^
30
20
10
=^^
.^^.^^
487 467 467 467 467 467
Round Round Round Round Round Round
c
1*0
I 10
a 10
0
4U
30
20
n
^S
30
20
n
^^
^==^
30
20
0
30
20
0
____^
^'"^^^^S
30
20
n
^^
487
Round
467
Round
487
Round
467
Round
67
Round
67
Round
Figure 5-39. Line plots of blood lead for individual children in Phase 2 (Rounds 4-7)
Cincinnati for each treatment group and four age groups.
5.4 COMPARISON BY CROSS-SECTIONAL STRUCTURAL
EQUATION MODELS
The cross-sectional structural equation model (XSEM) is a useful tool for answering the
question of whether a source of lead such as soil is a major component of environmental lead
exposure for the children. Lead hi household dust, particularly floor dust, is often the
mostimportant source of lead exposure for children, but house dust is not generally regarded
as a primary source because the lead in house dust is usually derived from other sources.
The sources of lead hi house dust include lead hi soil that is tracked into the house, lead
from deteriorating lead-based paint, lead dust from occupational exposure that is carried into
the house on shoes and clothing, lead particles deposited from the air (including resuspended
5-60
-------
Ages
**•
BAL SP
BALP1
BAL P2
40
18-29 f
Months 320
"8
12345
Round
40
30-41 |3°
Months j| 20
110
o
40
2 30
42-53 f3°
Months 120
110
o
40
123456
Round
12345
Round
54-65 f
Months 320
a 10
o
40
§30
66+ f
Months 120
110
o
123456
Round
1 23456
Round
40
30
20
10
0
40
30
20
10
0
40
30
20
10
0
40
30
20
10
0
40
30
20
10
0
123456
Round
12345
Round
1 23456
Round
123456
Round
123456
Round
40
30
20
10
0
40
30
20
10
0
40
30
20
10
0
40
30
20
10
0
"^ ^~
123456
Round
^^^
1 23456
Round
^— -— -
123456
Round
^-~-_^
Round
Figure 5-40. Line plots of blood lead for individual children in Baltimore for each
treatment group and five age groups.
5-61
-------
surface soil particles), and from household activities such as lead hobbies or home
occupations. Soil lead may therefore contribute to blood lead as both a direct exposure
source during outdoor activity, and indirectly as a source of lead in household dust.
5.4.1 General Issues in Structural Equation Modeling
The general conceptual model shown hi Figure 5-41, based on Figure 2-3, is the basis
for the structural equation models fitted hi this chapter. None of the three studies measured
airborne lead concentrations, which generally provide only small additions to child lead
exposure beyond the soil lead and dust lead particles historically deposited from leaded
gasoline, industrial emissions, and other sources of airborne particles. Fugitive emissions
from deleading or demolition of older structures or from soil excavation may have occurred,
but were not systematically observed.
[Atmospheric
I Particles
Soil
^-
Exterior Paint
Dust
Interior Paint
Dust
: Local ^
Fugitive
Dust j
^^
Exterior
Dust
Interior
Dust
f Secondary)
^ I Occupational!
I Dust )
I
Hand
Dust
Food
Child
Figure 5-41. Typical pathways of childhood exposure to lead in dust showing both the
complexity of the routes of exposure and the mobility of dust lead along
these routes.
5-62
-------
The general graphic representation for both cross-sectional and longitudinal structural
equation models is shown hi Figure 5-42. The variables hi the system can be classified as
independent variables or dependent variables. Independent variables are not predicted by
any other variables hi the system, and may vary by neighborhood, treatment group,
household, residence, or child. They are shown by oval figures with no arrows going into
them, and are also known as predictor variables or exogenous variables. Dependent
variables are shown as rectangular figures with at least one arrow going into them, and are
also known as response variables or endogenous variables. Dependent variables require
input from at least one other component of the system, either from independent variables or
other dependent variables. The arrow shows the direction of the relationship, with the
dependent variable being predicted by a multiple linear or nonlinear regression model using
the variables at the initial part of the arrow. Parameters whose estimates are shown in the
tables of this chapter are either regression coefficients hi the equations, or intercept terms m
the equations from Section 5.1.2.
INDEPENDENT OR
PREDICTOR VARIABLE
Not predicted by other
components of the '
system
INDEPENDENT OR
PREDICTOR VARIABLE
Not predicted by other
component!! of the
system
\
1
Ife
w
1
( \
DEPENDENT OR
RESPONSE VARIABLE
If Input changes,
output changes
V J
DEPENDENT OR
RESPONSE VARIABLE
Predicted by other
components of the
system
Figure 5-42. Explanation of the terms and features of the structural equation model.
5-63
-------
Figure 5-43 shows an expanded version of Figure 5-41, with greater emphasis on the
variables used for structural equation modeling. Exterior and ulterior lead paint dust was not
measured, but data are available on lead paint loadings on walls or trim. However, any
relationship between lead paint and house dust was compromised in all of the studies, either
by interior paint stabilization in Boston, by exterior paint stabilization hi Baltimore, or by
restricting the study to fully rehabilitated houses primarily in Cincinnati. Window dust lead
proved to be a significant predictor of floor dust lead that was distinguishable from lead in
soil in Boston and Cincinnati, but somewhat less predictive of blood lead, and so occupies a
position similar to soil in the pathway scheme. Floor dust lead was strongly correlated with
post-abatement blood lead and is shown as a more proximate predictor of blood lead hi the
pathway models.
;\tmospheric^^
Particles r^
Soil
•^-
Exterior Paint
Dust
^
Interior Paint
Dust
„ {^peuonaary ]
V Dust )
Interior
Window
Dust
"\
Food
Figure 5-43. Dust lead exposure pathway diagram, similar to Figure 5-41, showing the
assumed relationships for interior floor and window dust that were
modeled by cross-sectional SEM. Shaded compartments were not included
in the XSEM analyses.
5-64
-------
The main focus of the structural equation analyses is to evaluate the predictive role of
environmental lead measurements in dust and soil, rather than child-specific measurements hi
which the child is used as a sentinel of changes hi exposure. Therefore, we used a further
simplification of Figure 5-43 to develop the final structural equation model shown hi
Figure 5-44 that was used as basis for the analyses reported for Boston and Cincinnati. Soil
lead and window dust lead were used as independent predictors of both floor dust lead and
blood lead. Window dust lead was used to characterize non-soil exterior exposures and
ulterior exposures not characterized by floor dust lead. Hand dust lead was omitted as a
predictor of blood lead, replaced by the floor dust lead, window dust lead, and soil lead as
direct predictors.
Soil Lead
Concentration
Window
Dust Lead
(Cone, or
Load]
Floor Dust
Lead
(Cone, or Load)
Blood Lead
Concentration
v___ J
Figure 5-44. Adaptation of the soil and dust pathway diagram (Figure 5-43) that
illustrates the general scheme for the cross-sectional structural equation
models, using the notation of Figure 5-42.
5-65
-------
More complex models involving additional dependent or independent variables or
pathways would greatly increase the complexity of the analyses •, especially when the cross-
sectional models are linked longitudinally so as to assess effects over time. The simple
models presented here already include a large number of parameters that need to be
estimated from the data. With increasing complexity, the iterative estimation procedures
used to fit the structual equations fail to converge to a unique optimal solution for the
parameter estimates, and such complexity should be added only if there is a compelling
substantive scientific basis for doing so. In most applications, adding additional parameters
and interaction terms for age effects and gender differences was less informative than simply
stratifying the data by age or gender.
We evaluated a number of two-equation models in which the first equation represented
blood lead concentrations in children derived from environmental pathways and media
including soil and floor dust, and the second equation represented floor dust lead
concentration or lead loading derived from lead in exterior media, including soil lead and
lead on window sills or in window wells.
Blood lead concentrations are related to lead hi soil and to lead loading or concentration
in house dust at or shortly before blood leads are measured, as well as to prior or historic
lead exposures that have accumulated a (primarily skeletal) body burden of lead that
contributes to current blood lead concentrations. The child's age as well as many other
individual behavioral or demographic factors may also affect exposure. Although it is not
necessary to dwell on the concept that there is a "causal" implication for any proposed
predictive relationship, it should be noted that in a longitudinal lead study, some of the lead
hi the child's body (even hi blood and soft tissues) will be circulating hi blood at a later
measurement. Thus, estimates of blood lead concentrations hi earlier samples are expected
to be predictive of measurements from later samples, which are estimates of the same
quantity, in part. The models do not depend on causal interpretations, however, but do
assume a temporal direction in which the dependent variables depend on values of other
variables measured at the same time, or measured previously, but not on values measured in
the future.
Structural Equation Modeling is a computational approach that allows estimation of sets
of niter-related linear or nonlinear models (Buncher et al., 1991). This has been widely used
5-66
-------
for cross-sectional environmental pathway modeling (Bornschein et al., 1985, 1988, 1990;
Marcus, 1991, 1992). Applications to longitudinal lead studies have recently been developed
(Marcus, 1991; Menton et al., 1994; Marcus and Elias, 1994). The PROC MODEL
program in the SAS ETS computer package (SAS, 1992) allows estimation of either linear or
nonlinear models. This procedure is believed to produce less biased estimates of regression
coefficients than other estimation procedures that do not include fitting simultaneous
equations for blood lead to predictor variables such as lead in paint, soil, or dust.
The most complete and technically correct evaluation of the present three studies
requires simultaneous assessment of changes in blood lead levels and changes in
environmental lead pathways following soil lead or dust lead abatement. The underlying
assumptions in the Structural Equation Model approach are that abatement effects can be
inferred from changes in environmental lead exposure variables. Because this is a cause-
effect relationship, it is sequence-dependent or time-dependent. That is, the abatement must
take place before the environmental changes will occur. The cross-sectional SEM models
use correlation structure in the data to infer causal pathway relationships. The longitudinal
SEM models also use correlation structure to infer causality, but the logical basis for
inference is much stronger because the interventions or abatements precede the changes in
blood lead and environmental lead. Changes in control groups during the same period of
time then provide a basis for estimating treatment effects. .Any analysis of time-dependent
relationships should address the following assumptions:
(1) Both preabatement and postabatement blood lead levels reflect, hi part,
contemporary environmental lead exposures that can be characterized by
measurements of lead levels in soil, dust, paint, and other media;
(2) Postabatement blood lead levels may also reflect, hi part, preabatement blood
lead levels due to the contribution of preabatement body burdens of lead
(principally in the skeleton) from earlier exposures;
These models were fitted using indicator or "dummy" variables for different study or
treatment groups. Sometimes these indicator variables were used as "switches", for example
when postabatement soil lead concentration is modeled as a fraction of preabatement soil lead
for soil nonabatement groups, but as a new replacement value for the soil abatement groups.
At other times, indicator variables were used when the data suggested that the effect of
abatement was to modify the regression coefficient for the predicted variable (for example,
5-67
-------
floor dust lead concentration) for a pathway. In that case, separate coefficients were fitted to
the product of the treatment group indicator and the predictor variable (for example, entry
dust lead concentration) as well as separate intercept terms for each treatment group.
The purpose of structural equation modeling is to elucidate pathways for environmental
lead exposure from source to child. From this perspective, the development and testing of
pathway models for urban lead is an exploratory model-building activity that does not readily
lend itself to hypothesis testing. It is well known that "specification searches" such as step-
wise regression have complicated inferential properties (Learner, 1978), and the true P level
for an estimated regression coefficient may be quite different from the nominal P value.
An up-and-down search procedure was employed that started with a plausible pathway
diagram, and dropped nonsignificant blocks of parameters if all estimates of the same or
analogous parameters in different groups were zero or nonsignificant. New parameters were
added for each new pathway in the model, based on prior beliefs and on sample correlation
coefficients.
5.4.2 Boston Preabatement Cross-sectional Structural Equation Models
The model scheme for the Boston cross-sectional structural equation modeling is shown
in Figure 5-45, using the notation of Figure 5-44. The results for the twelve Boston models
with dust lead concentration are shown hi Table 5-1, and an example of Model 1 with the
output parameters is shown in Figure 5-46. The cross-sectional structural equation model
coefficients in these four tables correspond to equations 5-7 and 5-8 of Section 5.1.2, which
are repeated here for convenience.
gr
Z,F
gr
-ir
(5-7)
The blood lead models in Equation 5-7 have a single intercept term (denoted G) for each
round. The dust-to-blood regression coefficient (denoted B) is usually assumed to be the
same for all groups. The soil-to-blood lead regression coefficient (denoted F) is also shown.
The dust lead models follow the form of Equation 5-8, where the floor dust lead
5-68
-------
V
5-69
-------
TABLE 5-1. PREABATEMENT CROSS-SECTIONAL STRUCTURAL
EQUATION MODELS FOR BOSTON STUDY
FLOOR DUST LEAD CONCENTRATION
SEM EQUATION
COEFFICIENTS
INTERCEPT1 Ggr
S
I,
O
P
E
Floor -» Blood2 g
*»
Soil -»• Blood2 p
Window -» Blood2 p
INTERCEPT3 Cgr
S
L
O
P
H
Soil -»• Floor4
Dgr
Whidow -> Floor5
gr
Model 1
11.58s4
0.13
986s2
0.074
0.0657s4
Model 2
11.56s4
0.11
961s2
0.090
0.0652s4
Model 3
10.97s4
0.14
0.16
1008s2
0.075
0.0651s4
FLOOR DUST
INTERCEPT1 Ggr
S
L
O
P
E
Floor -> Blood6 B r
Soil -* Blood2 p r
Whidow -* Blood2
dust Pb cone &
Whidow -* Blood6 p
dust Pb load &
INTERCEPT3 Cgr
S
|,
O
P
E
Soil -* Dust7 D
**
Whidow -»• Floor7 n
dust Pb cone &
Whidow -» Floor8 n
dust Pb load &
11.37s4
0.18
12.0
0.0033
0.0031s4
11.34s3
0.20
28.3S1
0.0024
0.0048s2
11.23s4
0.19
0.0073
28.9S1
0.0022
0.0046s3
Model 4
11.20s4
0.12
0.0182
1014s2
0.076
0.0647s4
Model 5
11.92s4
-0.36s2
0.0694+
993s2
0.081
0.0644s4
Model 6
10.83s4
0.32+
1347s3
0.331s1
LEAD LOADING
11.00s4
0.17
0.0191
11.7
0.0036
0.0031s4
10.96s4
2.74
0.19
28.9S1
0.0023
0.0046s3
11.13s4
0.27+
47.8s2
0.0147s1
Note: In this and all subsequent tables of this chapter, the following notation is used to indicate statistical
significance:
S4 - Significance Level 4, P value < = 0.0001.
S3 - Significance Level 3, P value 0.0002 - 0.0019.
S2 - Significance Level 2, P value 0.002 - 0.0099.
SI - Significance Level 1, P value 0.01 - 0.0499.
IT - Nearly significant or one-tailed significance, P value 0.05 - 0.0999.
M - Marginally significant, P value 0.1 - 0.1999.
are ftg/dL Pb in blood.
2Units are jtg/dL Pb hi blood per 1000 jig/g Pb hi soil.
3Units are pg/g Pb hi dust.
4Units are /zg/g Pb hi dust per /ig/g hi soil.
5Units are /ig/g Pb hi dust per /*g/g hi dust.
6Units are /tg/dL Pb hi blood per 1000 /tg/m2 hi dust Pb load.
7Units are /tg/m2 Pb hi dust per /tg/g Pb hi soil.
8Units are /tg/m2 Pb hi dust per jig/m2 Pb hi dust.
5-70
-------
Window
Dust Lead
Cone.
Concentration/per 1,000 vg/g
0.065 pg/g
perng/g
Blood Lead
Concentration
10.97ng/dl
per^g/g
Floor Dust
Lead Cone.
1,008vg/g
0.
per 1,
Figure 5-46. Pathway diagram for Boston cross-sectional SEM Model 3, with results as
indicated from Table 5-1.
models have a single intercept term (denoted C) and a single soil-to-dust regression
coefficient (denoted D).
In general, any of several models gave an almost equally good fit to the preabatement
blood lead and dust lead data. However, none of the blood models provided a significantly
better prediction than did the group mean. The most significant finding (P=0.12) was for
Model 6 hi Table 5-1, with soil lead concentration as the only preabatement blood lead
predictor. However, this model also used soil lead as the predictor of dust lead
concentration, and provided a significantly worse prediction of dust lead than did any of the
models that used window lead as well as soil lead as predictor of floor dust lead. In fact,
preabatement soil lead was never a significant predictor of dust lead when window lead was
included hi the model. This indicates that, at the Boston preabatement stage, a substantial
amount of the lead hi floor dust may have come from the window dust, perhaps as
lead-based paint, rather than the soil. Very similar results were found when floor dust lead
5-71
-------
loading was used as the dust index, as shown also hi Table 5-1, and the analogous blood
lead-soil lead coefficient was even less significant (P = 0.18).
Because the Boston study excluded children whose blood lead concentration might be
too low (<7 /xg/dL) to be accurately measured following an expected reduction of three or
more jig/dL, or so high (>24 jig/dL) that the child would require immediate medical
intervention, there was a concern that this truncation at both ends might have biased the
statistical analyses. To address this concern, the Boston data set was artificially truncated to
the interval 9-22 /^g/dL, and a number of marginally significant relationships emerged.
Table 5-2 shows that the soil lead concentration by itself, or hi combination with the floor
dust lead concentration or window dust lead lead concentration, provides a marginally
significant preabatement predictor of blood lead. The dust lead variables have positive
coefficients in models with soil lead, and are marginally significant predictors by themselves.
Due to the collinearity between floor dust lead and window dust lead, neither is significant
when used together hi a model with soil lead, and while both were more significant when
used together hi a model without soil lead, the floor dust coefficient was negative. Similar
results were obtained hi the model for floor dust lead loading with blood lead truncation
shown hi Table 5-2.
5.4.3 Cincinnati Cross-Sectional Structural Equations Model
The Cincinnati study collected dust lead measurements at a number of locations. One
of the goals of the Cincinnati cross-sectional structural equations modelling exercise was to
evaluate the ability of different dust lead indices to predict blood lead and, hi turn, the
relationship of that dust lead index to soil lead. Four different dust lead locations were
considered, using both lead concentration and lead loading. The locations were: composite
ulterior floor, interior entry, and window sill. The models evaluated were very simple, with
soil lead-dust lead-blood lead and soil lead-blood lead pathways. Results are shown hi
Table 5-3.
The most useful models for predicting blood lead at Round 1 used floor dust or entry
dust lead concentration and loading, but none of the regression coefficients for blood lead
versus dust lead were statistically significant. In the model hi which both floor dust and soil
lead concentration were used, the blood lead versus floor dust lead coefficient was small
5-72
-------
TABLE 5-2. PREABATEMENT CROSS-SECTIONAL STRUCTURAL EQUATION
MODELS FOR BOSTON STUDY: BLOOD LEAD TRUNCATED (9-22 jig/dL)
FLOOR DUST LEAD
SEM EQUATION
COEFFICIENTS
INTERCEPT1
S
I,
O
P
E
Floor -> Blood2
Soil -» Blood2
Window -» Blood2
INTERCEPT3
S
L
, O
P
E
Soil -» Floor4
Window -* Floor5
Model 1
Ggr 12.43s3
B
gr
Fgr 0.268M
Fgr
Cgr 770S1
Dgr 0.106
Lgr 0.0760s4
Model 2
11.51s4
0.205M
0.305M
750S1
0.113
0.0750s4
Model 3
11.90s4
0.250M
0.0221M
755S1
0.111
0.0752s4
CONCENTRATION
Model 4
11.77s4
0.063
0.267
0.0151
753S1
0.112
0.0751s4
Model 5
12.99s4
-0.377s1
0.07881T
821S1
0.089
0.0765s4
Model 6
12.34s4
0.328M
1544s2
0.335s1
FLOOR DUST LEAD LOADING
INTERCEPT1
S
L
O
P
E
Floor -> Blood6
Soil -» Blood2
Window -* Blood2
dust Pb cone
Window -» Blood6
dust Pb load
INTERCEPT3
S
T,
0
P
E
Soil -» Dust7
Window -> Floor7
dust Pb cone
Window -* Floor8
dust Pb load
Model 7
Ggr 12.40s4
Bf
Fgr 0.264M
Fgr
gr
Cgr 91.7
D_r 0.0043
6
Lgr 0.0031s4
Lgr
Model 8
12.46s4
0.250
30.7S1
0.0019
0.0025s2
Model 9
11.76s4
3.94
0.276M
30.7S1
0.0015
0.0046
Model 10
11.83s4
0.251M
0.025M
9.2
0.0041
0.0031s4
Model 11
12.23s4
0.263M
0 083
30.7S1
0.0014
0.0046s.2
Model 12
12.24s4
0.3581T
51.7s2
0.0139s1
are /tg/dL Pb in blood.
2Units are /*g/dL Pb in blood per 1000 /ig/g Pb in soil.
3Units are /tg/g Pb in dust.
4Units are ^g/g Pb in dust per pg/g in soil.
5Units are /tg/g Pb in dust per /tg/g in dust.
6Units are j«g/dL Pb in blood per 1000 jig/m2 in dust Pb load.
7Units are /tg/m2 Pb in dust per /jg/g Pb in soil.
8Units are /*g/m2 Pb in dust per /zg/m2 Pb in dust.
5-73
-------
TABLE 5-3. PREABATEMENT CROSS-SECTIONAL STRUCTURAL EQUATION
MODELS FOR CINCINNATI STUDY: DUST TYPE MODELS
DUST LEAD CONCENTRATION ALL AGES1-2
SEM EQUATION
COEFFICIENTS
INTERCEPT1 Gg
Slope: Dust -* Blood2
Slope: Soil -* Blood2
INTERCEPT3
Slope: Soil -*
Dust4
B
F
D
With No Soil -* Blood Slope
Floor
7.41s4
4.65
160s4
0.330s3
Entry
7.02s4
5.35
182s4
0.317s3
Window
6.73s2
1,11
908s4
1.24s2
With Soil -
Floor
8.47s3
-0.02
1.72
160s4
0.331s3
Entry
2.52M
26.4s4
-4.081T
185s4
0.310s3
> Blood Slope
Window
9.10
-0.224
2.54
908s4
1.24s2
Soil
8.47s4
1.72S1
160s4
0.33s3
DUST LEAD LOADING ALL AGES
INTERCEPT1
Slope: Dust -*•
Slope: Soil -*
INTERCEPT3
Slope: Soil -*•
INTERCEPT1
Blood6
Blood2
Dust7
Slope: Dust -* Blood2
Slope: Soil -»• Blood2
INTERCEPT3
Slope: Soil -*
INTERCEPT1
Slope: Dust -»
Slope: Soil -*
INTERCEPT3
Slope: Soil -»
Dust4
Blood6
Blood2
Dust7
GR
B
F
D
GR
B
F
D
Gg
B
F
D
8.80s4
0-9,3
15.5
0.146s2
-1.02
38.60s2
211s3
0.1751T
6.53s2
38.47
17.2
0.0871T
8.69s4
0.23
34.6
0.216s1
DUST LEAD
3.16
19.65
1521T
0.267s1
DUST
12.55s4
-0.51s4
129.7
0.068
8.82s4
0.0087
14021T
0.458
8.65s4
-0.25
2.18
15.0
0.149s2
CONCENTRATION for
1.58
6.00
867s3
0.8941T
LEAD LOAD
11.56s4
-0.0016s4
1703
-0.394
8.43s4
-3.34s3
8.06s2
182s2
0.240s1
8.57s4
0.10
0.823
36.0
0.210s1
AGE 42+
8.30s3
7.541T
1801T
0.195
8.45s4
-0.001
1.98
1321s4
0.654
MONTHS1-2
8.13s4
-0.335s4
9.13S1
885s2
0.8331T
8.54s4
1.58
15.0
0.149s2
6.95s4
6.28S1
182s3
0.223S1
for AGE 42+ MONTHS1-2
7.63s4
-0.347s4
8.63M
16.2
0.1281T
10.83M
-0.462M
2.15
111.6
0.099
7.69s3
-0.00167s4
9.04S1
1640
0.0073
6.89s4
6.58S1
18.9
0.1161T
are /tg/dL Pb in blood.
2Units are /tg/dL Pb hi blood per 1000 /*g/g Pb hi soil.
3Units are /ig/g Pb hi dust.
4Units are /tg/g Pb hi dust per jtg/g hi soil.
5Units are pglg Pb hi dust per jig/g hi dust.
6Units are /ig/dL Pb hi blood per 1000 ^g/m2 hi dust Pb load.
7Units are /tg/m2 Pb hi dust per /ig/g Pb hi soil.
8Floor dust lead predicted from soil lead, only soil lead used to predict blood lead.
5-74
-------
(-0.02 /ig/dLper 1000 /*g/g) and the soil lead coefficient was large (1.72 /*g/dL), but not
statistically significant. When both entry dust lead and soil lead concentration were used as
predictors, the blood lead versus entry dust lead coefficient was large (26.4 jug/dL per
1000 /ng/g) and extremely significant (P = 0.0001), but the blood lead versus soil lead
concentration coefficient was also large and negative (-4.08 /xg/dL per 1000 /ig/g) and nearly
significant, suggesting a serious collinearity problem in which neither coefficient was
reliable. The relationship between floor dust lead concentration and soil concentration was
large, 0.33, and statistically highly significant, as was the regression coefficient between
entry dust lead concentration and soil lead, 0.31. The floor and entry dust lead loading
coefficients were only somewhat less significant.
Table 5-3 shows somewhat similar results when the data were restricted to a subset with
ages 42+ months. The best-fitting models for blood lead used entry dust lead concentration
or mat dust lead loading. The blood lead versus entry dust lead concentration regression
coefficient was large (19.65 /ig/dL per 1000 /ig/g lead hi entry dust) but was not statistically
significant, while the entry dust lead versus soil lead concentration coefficient of 0.267 was
large and statistically significant. Floor dust lead concentration produced a somewhat worse
fit to the blood lead data as measured by RMSE, but a statistically very significant regression
coefficient between blood lead and floor dust lead concentration, 38.6 ptg/dL per 1000 pg/g
lead in floor dust.
The Round 1 blood lead and dust lead models evaluated here do not identify an
across-the-board best model using any dust lead index, although floor dust lead appears
adequate in most cases. Models using flood dust lead concentration and floor dust lead
loading are compared in. Table 5-4. The best fitting models are shown schematically in
Figure 5-47.
The best fitting model of all tested, hi the right column of Table 5-3, uses only soil
lead to predict blood lead for children aged 42 months and older. The regression coefficients
for blood lead versus soil lead are large and statistically significant, whether hi simultaneous
fitting of blood lead and floor dust lead concentration (6.28 /*g/dL per 1000 jug/g lead hi soil)
or simultaneously fitting blood lead and floor dust lead loading (6.58 jwg/dL per 1000 ^g/g
lead hi soil). In the longitudinal models described below, we therefore used both floor dust
lead concentration and soil lead concentration as predictors of blood lead lead for Round 1.
5-75
-------
TABLE 5-4. PREABATEMENT CROSS-SECTIONAL STRUCTURAL EQUATION
MODELS FOR CINCINNATI STUDY: FLOOR DUST
FLOOR DUST LEAD CONCENTRATION1-2
SEM EQUATION
COEFFICIENTS
INTERCEPT1 Ggr
S
L
O
p
E
Floor -*• Blood2
Soil -»• Blood2
Window -» Blood2
INTERCEPT3 Ggr
S
L
O
P
H
Soil -* Floor4
Window -* Floor5
Model 1
8.12s4
1.3
0.190s1
99.9s2
0.2264s3
0.0458s3
Model 2 ModelS Model 4 Models
7.64s4 7.55s4 8.02s4 7.58s3
4.17S1 4.10S1 0.70 4.43
0.28 1.12
0.16 -0.01
99.4s2 99.9s2 99.9s2 99.5s3
0.2272s3 0.2247s3 0.2260s3 0.2266s3
0.0457s3 0.0458s3 0.0458s3 0.0457s3
Model 6 Model 7 Model 8
8.47s3 7.41s4 8.47s3
4.65 -0.02
1.72 1.72
160.3s3 160.4s4 160.3s4
0.3308s4 0.3303s4 0.3308s4
FLOOR DUST LEAD LOADING
INTERCEPT1 Ggr
S
L
O
P
E
Floor -» Blood6
Soil -* Blood2
Window -* Blood6
INTERCEPT3 Ggr
S
L
O
P
E
Soil -* Dust7
Window -*• Floor8
Model 9
8.40s4
1.26
0.003
-3.83
0.0898s2
0.00771s4
Model 10
8.29s4
-1.60
2.56
0.0117
-3.98
0.0904s3
0.00772s4
Model 11 Model 12 Model 13
8.54s4 8.80s4 8.65s4
0.93 -0.25
1.58 2.18
15.05 15.54 15.02
0.1486s2 0.1464s2 0.1491s2
^Units are /*g/dL Pb in blood.
2Units are ftg/dL Pb in blood per 1000 pglg Pb in soil.
3Units are jtg/g Pb in dust.
4Units are /tg/g Pb in dust per /tg/g in soil.
5Units are /ig/g Pb in dust per /tg/g in dust.
6Units are jig/dL Pb in blood per 1000 jig/m2 in dust Pb load.
'Units are /tg/m2 Pb in dust per jig/g Pb in soil.
8Units are /ig/m2 Pb in dust per jig/m2 Pb in dust.
5-76
-------
00
Ji
5-77
-------
The Cincinnati cross-sectional structural equation models are shown in Table 5-4. The
floor dust lead model is extended by including window lead as a predictor of both floor dust
lead and blood lead, analogous to the Boston models. While the various models shown in
Table 5-4 provide only a slight additional improvement hi Round 1 blood lead, they provide
a significant improvement in the prediction of Round 1 floor dust lead concentration. In
model 4 hi which floor dust lead concentration, window dust lead concentration, and soil
lead concentration are all used as predictors, none of these variables is statistically
significant. In combinations using one or two of the variables, models 1, 2, 3, and 5 were
best. Model 1 shows a statistically significant relationship between window dust and blood
lead, whereas the soil lead coefficient was not statistically significant. In Models 2 and 3,
there were statistically significant relationships between blood lead and floor dust lead
concentration, 4.17 /tg/dL per 1000 /ng/g lead hi floor dust hi Model 2. The Model 2
regression coefficients between floor dust lead concentration and soil lead concentration,
0.227, and between floor dust lead concentration and window dust lead concentration,
0.0457, were highly significant. This was therefore the starting point for the longitudinal
models described hi Section 5.6. Additional analyses were done using floor dust lead
loading, but the regression coefficients for floor dust lead concentration were generally more
stable and more significant. Note also that Models 7 and 8 are similar to Models 2 and 3
respectively, except that window dust lead is not used as a predictor of floor dust lead. The
floor dust lead regression coefficient hi Model similar hi magnitude to that in Model 2, but is
not statistically significant. The floor dust lead and soil dust lead regression coefficients hi
Model 8 are quite different from those hi Model 3, and are not statistically significant,
whereas the blood lead vs floor dust lead concentration regression coefficient in Model 3 is
statistically significant. This table demonstrates the importance of simultaneous fitting of
pathway model components. Models 9-13 used floor dust lead loading, which hi general was
not as good a predictor of blood lead.
5.5 COMPARISON BY REPEATED MEASURES ANALYSIS
Several approaches are evaluated for analyzing the longitudinal data from the three
cities using "repeated measures" models. In many cases, the ability to identify differences
5-78
-------
among interventions was greatly improved by including covariates in the analyses. For
example, child blood lead is known to change with age. When age is included as a
covariate, some of the variation in blood lead differences before and after abatement can be
attributed to the age of the child when the abatement was carried out. Controlling for the
influence of age increases the ability to more accurately estimate the relationship between
blood lead and other variables, such as soil lead. Similarly, the effect of abatement may
depend on changes in proximate exposure variables such as house dust lead. The effects of
changes in house dust lead may be different at different ages, however, so that other
covariates that may be useful in the analyses include interactions between age, house dust
lead, and treatment group.
The use of baseline preabatement environmental or demographic measurements for
individual subjects as covariates allows one to proceed as if all groups had the same starting
values. The use of differences in environmental measurements before and after abatement
allows one to proceed as if individuals responded similarly to similar changes in lead
exposure, which is a fundamental assumption in a remediation and intervention program.
In general, differences in environmental indices before and after abatement were found to be
no more predictive of blood lead changes than the absolute baseline or final values.
Repeated measures analyses can be carried out using standard statistical programs for
analyses of general linear models. The PROC MIXED program in the SAS statistical
package (SAS, 1992) was used for most of the analyses. Analyses of repeated measures
models with time-varying covariates can be conveniently carried out using these programs.
Repeated measures models with more than two phases or time points may require specific
assumptions about time correlation structure hi some programs, but no such assumptions are
needed here when comparing outcomes at only two time points, pre- and postabatement.
5.5.1 Boston Repeated Measures Analysis of Variance (ANOVA)
The Boston repeated measures ANOVA results for all 150 children (excluding the two
who became lead-poisoned) are shown in Tables 5-5 and 5-6. The repeated measures
ANOVA coefficients of Table 5-5 and following are taken from equation 5-16 of Section
5.1.2, also repeated here for convenience.
5-79
-------
TABLE 5-5. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
BOSTON STUDY: EFFECT OF AGE ON REDUCTION IN BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 3
Study Group
Abate
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
Control
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
Age Group1
All Ages 0-17 Months 18-41 Months 42+ Months
(N=150) (N=19) (N=100) (N=31)
1.87s2
0.33
1.54S1
0.164s2
0.028
0.136S1
Response2 (Er)
0.69 2.51s2
-2.09 1.17M
2.78 1.331T
Log Response3 (Er)
0.056 0.201s2
-0.177 0.077
0.233 0.1241T
1.09
-0.75
1.84
0.146
0.007
0.140
*Age is age in months at time of Round 1 blood sample.
2Units are jtg/dL Pb in blood.
3Units are log /ig/dL Pb in blood.
TABLE 5-6. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
BOSTON STUDY: EFFECT OF AGE ON REDUCTION IN BLOOD LEAD (Er)
BETWEEN ROUNDS 3 AND 4
Age Group1
Study
Abate
BOS P-S
BOS PI-S
BOS PI-S
BOS P-S
BOS PI-S
BOS PI-S
Group
Control
BOS SPI
BOS SPI
BOS P-S
BOS SPI
BOS SPI
BOS P-S
All Ages
(N=147)
1.77M
3,80s3
2.031T
0.125
0.310s1
0.185M
0-17 Months 18-41 Months
(N=18) (N=98)
Response2
(Er)
1.12 0.09
4.33 3.39S1
3.21 2.491T
Log Response3 (Er)
0.228
0.401
0.173
0.042
0.2991T
0.257M
42+ Months
(N=31)
371M
3.76M
0.05
0.359
0.278
0.081
*Age is age in months at time of Round 1 blood sample.
2Units are /tg/dL Pb in blood.
3Units are log /tg/dL Pb in blood.
5-80
-------
H
h(g)
eir
(5-16)
Effect sizes can be calculated very similarly with the methods used in this report. The
simplest effect size calculation, comparing Group g to Group h across rounds 1 and r, is the
difference
- (GM-Gh)
(5-20)
This is used in repeated measures ANOVA, and to calculate the intercept effect in RM
ANCOVA.
All tables show the estimated effect of soil abatement or dust abatement groups versus
reference groups. Table 5-5 shows the mean reduction in blood lead during Phase 1 of the
Boston study, Round 1 to Round 3. The soil abatement group BOS SPI shows a highly
significant (P = 0.0042) decrease hi blood lead of 1.87 j*g/dL relative to the control group
BOS P-S, and a statistically significant decrease (P = 0.016) in blood lead of 1.54 jttg/dL in
the soil abatement group BOS SPI compared to the group BOS PI-S that received only dust
abatement. The groups BOS PI-S and BOS P-S did not show significantly different
(P = 0.62) changes in blood lead between Rounds 1 and 3. Table 5-5 also shows similarly
significant (P = 0.0064) reductions hi the logarithm of blood lead, with the effect size of soil
abatement relative to control of 0.164 and relative to dust abatement of 0.136 (P = 0.021).
This corresponds to estimated effects on a percentage basis of 17.8% for BOS SPI versus
BOS P-S and 14.6% for BOS SPI versus BOS PI-S. This percentage is calculated from the
log transformed data as a function of the geometric mean (GM), where
5-81
-------
GMgl -
GMg2 =
PERCENT CHANGE =
(5-21)
GM,,
--1 100
Table 5-6 shows the mean reduction in blood lead during Phase 2 of the Boston study,
Round 3 to Round 4. Two groups received soil and dust abatement during Phase 2,
BOS PI-S and BOS P-S. Group BOS SPI received no further abatement and was an
appropriate reference group for Phase 2 comparison. The soil abatement group BOS PI-S
shows a highly significant (P = 0.0006) decrease hi blood lead of 3.80 ptg/dL relative to the
reference group BOS SPI, and a marginally significant (P = 0.12 two-tailed,
0.061 one-tailed) decrease hi blood lead of 1.77 /xg/dL in the soil abatement group BOS P-S
compared to the reference group BOS SPI that received no further abatement hi Phase 2.
The groups BOS PI-S and BOS P-S showed near-significantly (P = 0.079) different changes
hi blood lead between Rounds 3 and 4, with group BOS PI-S that had received dust
abatement hi Phase 1 showing an additional 2.03 /ng/dL reduction compared to group
BOS P-S that did not receive dust abatement. Table 5-6 also shows similarly significant
(P = 0.022) reductions hi the logarithm of blood lead, with the effect size of soil abatement
hi BOS PI-S relative to BOS SPI of 0.310 and BOS P-S relative to BOS SPI of 0.125 (NS,
P = 0.37). There was also an indication of a difference between BOS P-S and BOS PI-S of
0.185, but it was very marginally significant (P = 0.18) although hi the expected direction.
This corresponds, on a percentage basis, to an estimated Phase 2 effect of BOS PI-S relative
to BOS SPI of 36.3%.
Age Effects in Boston
Tables 5-5 and 5-6 also show a breakdown of the Boston ANOVA results by age
category. Because of greatly reduced sample sizes, none of the effect size estimates hi the
youngest age group (9 to 17 months at Round 1) are statistically significant, even though they
5-82
-------
are large in magnitude and somewhat similar to those in older children. The largest subset
of Boston children, ages 18-41 months at Round 1, showed an even larger effect of soil
abatement than did the whole group. Table 5-5 shows a mean effect of BOS SPI versus
BOS P-S of 2.51 pig/dL (P = 0.0020). Table 5-5 also shows some differentiation among
treatments, with a nearly significant (P = 0.077 two-tailed, 0.038 one-tailed) effect of
1.33 /Ltg/dL between the Phase 1 soil abatement group BOS SPI and dust abatement group
BOS PI-S. The log-transformed results for Phase 1 hi Table 5-5 are also highly significant
(P = 0.0084) in the 18-41 month age group and somewhat larger than for the whole group,
with an effect size of 0.201 or 22.3% reduction in blood lead in the BOS SPI group
compared to the BOS P-S control group. Effect sizes hi the age group 42+ months are
smaller and not statistically significant, which may also be due to a small sample size.
The Phase 2 results are shown hi Table 5-6. The effects are all hi the expected
direction. The group BOS PI-S that had abatements hi both Phase 1 and Phase 2 showed the
greatest reductions hi blood lead. The group BOS P-S that received Phase 2 soil abatement
showed a greater Phase 2 reduction hi blood lead lead than the group BOS SPI that received
Phase 1 soil abatement, but not further remediation hi Phase 2. All of the effect size
estimates are positive, at all ages. In the youngest age group, there are no statistically
significant effects, but the magnitude of the blood lead reductions is large and similar to
those hi older children. In the 18 to 41 month age group, the significant effects are slightly
smaller than for the group as a whole. The Phase 2 soil abatement group BOS PI-S had a
significantly larger (P = 0.014) decrease in blood lead of 3.39 pig/dL than did BOS SPI, and
a marginally greater reduction of 2.49 /*g/dL (P = 0.076) than did group BOS P-S. The
log-transformed analysis shown hi Table 5-6 found somewhat less significant (P = 0.074)
effects for 18-41 month old children, with an effect of 0.299 for BOS PI-S versus BOS SPI,
and an effect of 0.257 (P = 0.13) for BOS PI-S versus BOS P-S. There was also some
indication in Table 5-6 of an effect of soil lead abatement at Phase 2 hi the older children,
ages 42+ months, amounting to 3.71 pg/dL (P = 0.12) for BOS P-S versus BOS SPI, and
3.76 jtig/dL (P = 0.10) for BOS PI-S versus BOS SPI. The results for older children were
large but not significant (P = 0.21) on a log scale, amounting to 0.359 for BOS P-S versus
BOS SPI, and 0.278 (P == 0.31) for BOS PI-S versus BOS SPI in Table 5-6.
5-83
-------
Race/Ethnicity Effects in Boston
The analysis of subgroups of children clearly identified as Afro-American and as
non-Black did not find large differences in overall response hi Phase 1, although other
analyses using analysis of covariance (ANCOVA) methods suggest some ethnic differences in
response to environmental lead exposure. Table 5-7 shows that there was a significant
(P = 0.050 two-tailed, 0.025 one-tailed) effect of 1.91 /tg/dL comparing BOS SPI to
BOS P-S from Rounds 1 to 3. The logarithmic analysis hi Table 5-7 found a nearly
significant (P = 0.064) effect of 0.163 or 17.7% reduction from soil abatement. The effects
for non-blacks were smaller and only marginally significant, although not significantly
different in magnitude.
Results hi Phase 2 of the study were substantially larger. Both of the Phase 2 soil
abatement groups hi Table 5-7 showed substantial and statistically significant reductions
relative to the Phase 2 non-abatement group BOS SPI. The effect for BOS P-S versus
BOS SPI was 3.65 pig/dL (P = 0.014), much larger than the overall group effect, and the
effect for BOS PI-S versus BOS SPI was 4.46 /ig/dL (P = 0.0044). However, the log-
transformed analyses hi Table 5-7 showed less significant (P = 0.070) changes, with only
the BOS PI-S versus BOS SPI effect of 0.400 nearly significant. None of the effects for
non-Black children were statistically significant, although this may be a consequence of the
relatively small number of children identified hi this subgroup, N = 32, since the estimated
effect of BOS PI-S versus BOS P-S of 4.04 /*g/dL was relatively large.
Gender Effects in Boston
Table 5-7 shows large effects hi male children. The Phase 1 effect of both soil
abatement and dust abatement appear to be very similar and significant, 2.19 /ng/dL for
BOS SPI versus BOS P-S and 2.08 ^g/dL (P = 0.016) for BOS PI-S versus BOS P-S
(P = 0.020). The logarithmic analyses hi Table 5-7 find even more significant
(P = 0.0099) effects for males, 0.197 for BOS SPI versus BOS P-S and 0.194 for BOS PI-S
versus BOS P-S. The Phase 1 effects for female children are smaller and only marginally
significant, 1.45 jig/dL for BOS SPI versus BOS P-S (P = 0.13). The male-female
differences are not significant, possibly due to larger uncertainty hi the estimated female
effect sizes.
5-84
-------
TABLE 5-7. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
BOSTON STUDY: EFFECT OF RACE OR SEX
All
Study Group Rounds 1-3: (N = 150)
Rounds 3-4: (N=147)
Abate
BOS
BOS
BOS
SPI
PI-S
SPI
Control
BOS
BOS
BOS
P-S
P-S
PI-S
Black
(N=75)
(N=74)
Reduction in
1
0
1
.87s2
.33
.54S1
1.
0.
1.
Reduction in
BOS
BOS
BOS
P-S
PI-S
P-S
BOS
BOS
BOS
SPI
SPI
P-S
1
3
.77M
.80s3
2.031T
Reduction
BOS
BOS
BOS
SPI
PI-S
SPI
BOS
BOS
BOS
P-S
P-S
PI-S
0
0
0
.164s2
.028
.136S1
Reduction
BOS
BOS
BOS
P-S
PI-S
P-S
BOS
BOS
BOS
SPI
SPI
P-S
0
0
0
.125
.310S1
.185M
3.
4.
0.
Group
Nonblack
(N=32)
(N=32)
Male
(N=80)
(N=78)
Blood Lead (Er) Between Rounds 1
911T
78
13
1.54
-0.18
1.72M
2.19S1
0.11
2.08S1
Blood Lead (Er) Between Rounds 3
65S1
46s2
81
in Log Blood
0.
T
1631
0.092
0.
070
in Log Blood
0.
0.
T
0.
206
4001
194
-1.86
2.18
4.04M
1.65
3.02S1
1.37
Lead (Er) Between Rounds
0.070
-0.105
0.175M
0.197s2
0.002
0.194s2
Lead (Er) Between Rounds
0.038
0.144
0.106
0.176
0.173
-0.003
Female
(N=70)
(N=69)
andS1
1.
0.
0.
45M
52
93
and4J
2.
5.
2.
1
0.
0.
0.
3
0.
19
00**
801T
and32
130**
050
080
and42
080
0.47017
0.
390**
are jtg/dL Pb in blood.
2Units are log /tg/dL Pb in blood.
The Phase 2 analyses shown hi Table 5-7 find larger and more significant effects in
female children. The Phase 2 effect of soil abatement hi BOS PI-S versus no abatement hi
BOS SPI is 3.02 /ig/dL (P = 0.05) hi males, but 5.00 /tg/dL hi females (P = 0^0039).
There may also be a marginally significant difference (P = 0.093) between female children
in the two Phase 2 soil abatement groups, with a somewhat greater effect of 2.80 ptg/dL hi
BOS PI-S (that had Phase 1 dust abatement) compared to BOS P-S (no Phase 1 abatement).
However, the statistical significance of the log-transformed data hi Table 5-7 was much
lower, with no significant Phase 2 abatement effects for males and less significant effects for
females hi BOS PI-S versus BOS SPI and hi BOS PI-S versus BOS P-S compared with the
5-85
-------
non-transformed data analyses shown in Table 5-7. The male-female differences are not
significant, possibly due to larger uncertainty in the estimated female effect sizes.
Blood Lead Truncation Effects in Boston
The design of the Boston study involved some truncation of the range of starting values
of blood lead concentration. One way of assessing the possible effect of this design choice is
to compare the results of the analyses with the results of similar analyses of the Boston data
set using truncated subsets of the data. After preliminary assessment to determine the
number of observations with different truncation levels, the following truncated data sets
were selected:
Full data set, 7 to 24 jig/dL;
Minimal data set, 10 to 19 /*g/dL;
Upper truncation, 7 to 19 /zg/dL;
Lower truncation, 10-24 #g/dL.
The minimal data set includes only children who would have been considered lead-burdened
at the beginning of the study (blood lead at least 10 /tg/dL) and excludes those children
whose blood lead concentrations were at least 20 /zg/dL and whose residences might have
been considered as appropriate for environmental intervention on the basis of the blood lead
finding. The upper truncation adds children with somewhat lower blood leads to the minimal
data set, and the lower truncation adds children with higher blood leads to the minimal data
set.
The truncated data sets for children ages 9 to 17 months at Round 1 did not show any
significant effects, probably due to the very small sample sizes, and are not discussed
further. The Phase 1 results for children of ages 18 to 41 months are shown in Table 5-8.
The highly significant whole-sample effect of 2.51 /zg/dL for BOS SPI versus BOS P-S in
Phase 1 is largely insensitive to the truncation, increasing to 2.67 /*g/dL (P = 0.013) hi the
minimal data set and 2.73 jig/dL (P = 0.0070) in the lower truncation data set, and
decreasing only to 2.42 /^g/dL (P = 0.0039) in the upper truncation data set. The near
significant effect of BOS SPI relative to BOS PI-S is increased in all of the truncation data
sets, with somewhat lower significance (P = 0.059) due to the smaller sample size except for
the upper truncation, where the effect size is 1.53 pig/dL. The log-transformed analyses
5-86
-------
TABLE 5-8. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
BOSTON STUDY: EFFECT OF TRUNCATION ON REDUCTION IN
BLOOD LEAD (Er) BETWEEN ROUNDS 1 AND 3
Study
Group
Truncation Category
7-24
fjZi 10-19 /*g/dL 7-19/tg/dL 10-24 /*g/dL
Age 18-41: ™-100) (N=67) (N=92) (N=75)
Age 42-52: (N=31) (N=16) (N=29) (N=18)
Abate
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI
BOS SPI
Control
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
Change hi Blood Lead for Age Group 18-41 Months2
2.51s2 2.67S1 2.42s2 2.73s2
1.17M 0.96 0.89 1.29M
1.331T 1.71M 1.531T 1.43M
Change hi Blood Lead for Age Group 42-52 Months2
1.09 0.50 1.45 0.00
-0.75 -1.58 -0.20 -2.25
1.84 2.08 1.65 2.25
Change in Log Blood Lead for Age Group 18-41 Months3
0.201s2 0.212 0.205s1 0.203s1
0.077 0.059 0.068 0.068
0.1241T 0.153M 0.1371T 0.135M
Change in Log Blood Lead for Age Group 42-52 Months3
0.146 0.042 0.176M 0.010
0.007 -0.154 0.035 -0.180
0.140 0.196 0.142 0.191
1Ia the Boston study, children with screening blood lead concentrations below 7
were excluded from the study for reasons discussed hi Chapter 4.
2Units are jtg/dL Pb hi blood.
3Units are log /ig/dLPb in blood.
and above 24 jag/dL
show a similar pattern, with larger but less significant effects of BOS SPI versus BOS P-S
and BOS SPI versus BOS PI-S in the truncated data sets.
The Phase 2 results for children of ages 18-41 months are shown hi Table 5-9. The
significant whole-sample effect of 3.39 j«g/dL for BOS PI-S versus BOS SPI in Phase 2 is
more sensitive to the truncation, increasing to 3.89 pig/dL (P = 0.033) hi the minimal data
set and to 4.85 j^g/dL (P = 0.0040) hi the lower truncation data set, and decreasing to
2.56 fJLg/dL (P = 0.079) hi the upper truncation data set. However, the effect of the other
Phase 2 soil abatement group BOS P-S versus BOS SPI, which was not at all significant hi
5-87
-------
TABLE 5-9. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
BOSTON STUDY: EFFECT OF TRUNCATION ON
REDUCTION IN BLOOD LEAD (Er) BETWEEN ROUNDS 3 AND 4
Truncation Category
Study Group,
Abate
BOS
BOS
BOS
BOS
BOS
BOS
P-S
PI-S
PI-S
P-S
PI-S
P-S
Control
BOS
BOS
BOS
BOS
BOS
BOS
SPI
SPI
P-S
SPI
SPI
P-S
7-24 /ig/dL2
(N=31)
10-19 Atg/dL
(N=16)
Change hi Blood Lead
0.90
3.39S1
2.491T
Change hi
0.042
0.2991T
0.257M
2.10
3.89S1
1.79
Log Blood Lead
0.211
0.372s1
0.161
7-19 /jg/dL
(N=29)
for Age
for Age
Group 18-41
0.53
2.561T
2.03M
Group 18-41
0.013
0.266M
0.252M
10-24 /*g/dL
(N = 18)
months2
2.40M
4.85s2
2.45M
Months3
0
0
0
.226M
.414s2
.188
'In the Boston study, children with screening blood lead concentrations below 7 /*g/dL and above 24 /*g/dL
were excluded from the study for reasons discussed in Chapter 4.
2Units are /tg/dL Pb in blood.
3Units are log /tg/dL Pb hi blood.
the whole sample, is larger and very marginally significant (P = 0.17) hi the truncated
sample with children 7 to 9 /*g/dL omitted. The nearly significant effect of BOS PI-S
relative to BOS P-S hi Table 5-9 is decreased hi all of the truncation data sets, with much
lower significance due to the smaller sample size. The log-transformed analyses show a
similar pattern. The nearly significant effect of BOS PI-S versus BOS SPI hi the whole
sample is larger and much more significant (P = 0.0093) hi the lower truncated sample.
The non-significant effect of BOS P-S versus BOS SPI in the whole sample is similarly
larger and more significant (P = 0.17) hi the lower truncated sample. The Phase 1 results
for children of ages 42+ months are shown hi Table 5-8. None of the effects is even
marginally significant, and this is not substantially changed by truncating the data set. The
log-transformed analyses show a similar pattern, with the only change in significance class
occurring hi BOS SPI versus BOS P-S with P = 0.25 for the whole data set and P = 0.19 in
the upper truncation data set.
The Phase 2 results for children of ages 42+ months are shown hi Table 5-9. The
marginally significant whole-sample effect of 3.76 /xg/dL for BOS PI-S versus BOS SPI in
Phase 2 is somewhat sensitive to the truncation, increasing to 5.76 pcg/dL (P = 0.13) in the
5-S
-------
minimal data set and to 5.90 jtg/dL (P = 0.067) in the lower truncation data set, and
decreasing only to 3.70 jieg/dL (P = 0.13) in the upper truncation data set. The marginally
significant effect of BOS P-S relative to BOS SPI is almost unchanged hi all of the truncation
data sets, with lower significance due to the smaller sample size. The log-transformed
analyses show a similar pattern, with no significant effects of BOS PI-S versus BOS SPI nor
BOS P-S versus BOS SPI in the truncated data sets.
In summary, there appears to be some sensitivity to truncation of the data set. There
were a number of situations hi which larger and more significant results were found hi the
truncated data set, particularly when children with initial blood lead concentrations less than
10 jiig/dL were omitted. The effects of truncation were noted hi both Phase 1 and Phase 2
analyses.
5.5.2 Cincinnati Repeated Measures Analysis of Variance (ANOVA)
The analyses are based on 223 children whose blood lead measurements were taken hi
both Rounds 1 and 4. As noted above, the sample sizes reported are the maximum number
of children who could have been used in the analyses. Due to missing values hi covariates
or classification variables, the actual number used is generally somewhat smaller. The effect
size comparisons are based on neighborhood-by-neighborhood comparisons for the 5
neighborhoods with sufficient numbers of follow-up measurements to allow comparisons.
The Phase 1 dust abatement neighborhoods are CIN I-SE(D) and CIN I-SE(F); CIN I-SE(B)
was omitted. The no-treatment or control neighborhoods for both Phase 1 (rounds 1 to 4)
and Phase 2 (rounds 4 to 7) were CIN NT(G) and CIN NT(M). The Phase 1 soil abatement
neighborhood was CIN SEI(P), which also received no abatement hi Phase 2 of the
Cincinnati study. Comparisons are carried out among neighborhoods hi different treatment
groups, and sometimes between neighborhoods within each group.
Table 5-10 shows the effect size for blood lead between Round 1 and Round 4. One of
the largest differences hi the study is the difference between the two control or no-treatment
neighborhoods, CIN NT(G) and CIN NT(M), with blood lead hi CIN NT(G) decreasing
3.58 /xg/dL more than hi CIN NT(M) between Round 1 and Round 4 (P = 0.073). Blood
lead hi the soil abatement neighborhood, CIN SEI(P), decreased a little more than hi the
no-treatment neighborhood of CIN NT(M), 1.02 /ig/dL (P =. 0.60), but blood lead hi the
5-89
-------
TABLE 5-10. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
CINCINNATI STUDY: EFFECT OF AGE BETWEEN ROUNDS 1 AND 4
Age Group1
Study
Abate
CIN NT (G)
GIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN I-SE (D)
CIN I-SE (D)
CIN I-SE (F)
CIN I-SE (F)
Group
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (G)
CIN NT (M)
CIN NT (G)
CIN NT (M)
All Ages
(N=223)
Reduction
3.581T
-2.56s1
1.02
-2.43S1
-1.20
-0.14
3.441T
-1.37
2.21
9-17 Months
(N=69)
in Blood Lead (Er)
13.10s3
-5.62s1
7.481T
-1.57
-0.91
-4.05M
9.05S1
-4.701T
8.40S1
18-41
Months
(N=80)
Between Rounds
9.211T
-0.97
8.24M
0.10
-0.26
-1.07
8.14M
-0.70
8.511T
42+
Months
(N=70)
1 and42
-2.57M
-1.97M
-4.55S1
-4.34s2
-2.991T
2.36M
-0.21
1.02
-1.55
Reduction in Log Blood Lead (Er)
Between Rounds 1 and 43
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN I-SE (D)
CIN I-SE (D)
CIN I-SE (F)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (G)
CIN NT (M)
CIN NT (G)
CIN NT (M)
0.615s2
-0.403s3
0.212
-0.2411T
-0.170
-0.163
0.453S1
-0.233M
0.3831T
1.903s4
-0.775s2
1.128s2
-0.142
-0.187
-0.634s1
1.269s3
-0.589s1
1.314s3
0.804
-0.199
0.605M
0.007
-0.048
-0.206
0.598M
-0.152
0.652M
-0.129
-0.346s1
-0.475s1
-0.409s2
-0.3361T
0.063
-0.065
-0.010
-0.138
*Age is age in months at time of Round 1 blood sample.
2Units are /ig/dL Pb in blood.
3Units are log /ig/dL Pb in blood.
dust abatement neighborhoods decreased much more, 3.44 /ig/dL for CIN I-SE(D) versus
CIN NT(M) (P = 0.08) and 2.21 ^g/dL for CIN I-SE(F) versus CIN NT(M) (P = 0.30).
On the other hand, blood lead decreased much less in CIN SEI(P) than in the other
5-90
-------
no-treatment neighborhood, CIN NT(G), -2.56 /xg/dL (P= 0.036). In fact, CIN SEI(P)
decreased much less than either of the dust abatement neighborhoods of CIN I-SE(D)
(-2.43 jtg/dL, P = 0.049) or CIN I-SE(F) (-1.20 /xg/dL, P = 0.36). A negative sign on
effect size means that the decrease was smaller in the comparison neighborhood than in the
reference neighborhood., which is generally contrary to what was expected in one-tailed tests.
The log-transformed blood lead analyses for Phase 1 shown hi Table 5-10 exhibit the
same pattern, but with much more significant effects. The largest effect in Table 5-10 is the
difference between the two no-treatment neighborhoods, 0.615 on a log scale (P = 0.0032).
This corresponds to a percentage difference of 85% hi blood lead reduction between Round 1
and Round 4. The soil abatement neighborhood of CIN SEI(P) showed a significantly
smaller reduction compared to CIN NT(G), -0.403 (P = 0.0016). While CIN SEI(P)
showed a reduction of 0.212 or 23.6% compared to CIN NT(M), the effect was not
significant (P = 0.30). Both CIN I-SE(D) and CIN I-SE(F) showed significant or near-
significant improvements over CIN NT(M), respectively 0.453 (P = 0.03) and 0.383
(P = 0.07), but decreased less than the other control neighborhood, CIN NT(G).
The results for Phase 2 are shown hi Table 5-11. The only effect of even marginal
significance is the difference between the two control neighborhoods. However, the effect is
hi the opposite direction to the Phase 1 difference between CIN NT(G) and CIN NT(M),
- 2.55 /tg/dL. The log-transformed analyses for Phase 2 shown hi Table 5-11 show a similar
pattern, but with greater significance. The difference between CIN NT(G) and CIN NT(M)
was larger hi Phase 2, -0.334 on a log scale (P = 0.062), or 40%. While neither CIN
SEI(P) nor CIN NT(G) received abatement during Phase 2, blood lead hi CIN SEI(P)
decreased more than in CIN NT(G) by 0.168 or 18.3% (P = 0.126), and less than hi CIN
NT(M) by -0.166 (P = 0.34).
Age Effects
Tables 5-10 through 5-11 also show age-stratified analyses. Unlike Boston and
Baltimore, the Cincinnati study had almost the same number of children in each of the
Round 1 age categories 9 to 17 months, 18 to 41 months, and 42+ months. In the youngest
age group, the differences identified earlier were much larger and much more significant.
The difference between CIN NT(G) and CIN NT(M) during Phase 1 was 13.1 /*g/dL
5-91
-------
TABLE 5-11. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
CINCINNATI STUDY: EFFECT OF AGE BETWEEN ROUNDS 4 AND 7
Age Group1
Study Group
Abate
GIN NT (G)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN NT (G)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
Control
GIN NT (M)
GIN NT(G)
GIN NT (M)
GIN I-SE (D)
GIN I-SE (F)
GIN NT (M)
GIN NT(G)
GIN NT (M)
GIN I-SE (D)
GIN I-SE (F)
All Ages 9-17 Months 18-41 Months
(N=223) (N=69) (N=80)
Reduction
-2.55M
0.85
-1.70 .
-0.01
0.72
Reduction hi
-0.3341T
0.168M
-0.166
0.033
0.047
in Blood Lead (Er) Between Rounds 4
-4.56
1.06
-3.50
-1.13
1.42
Log Blood
NC
-1.75
0.27
-1.48
0.78
0.23
Lead (Er) Between Rounds
-0.188
0.128
-0.059
0.075
0.062
42+ Months
(N=70)
and72
NC
4and73
-0.304M
0.210M
-0.094
0.123
0.103
'Age is age in months at time of Round 1 blood sample.
2Units are jtg/dL Pb in blood.
3Units are log /ig/dL Pb in blood.
(P = 0.0014) as seen in Table 5-10. GIN SEI(P) was significantly less reduced than CIN
NT(G), -5.6 /ig/dL (P = 0.020) in the youngest age group, and reduced significantly more
than in CIN NT(M), 7.5 jug/dL (P = 0.053). All of the other effect sizes compared to CIN
NT(G) and CIN NT(M) were larger, in the same direction as the whole-group analyses, and
much more significant in this age group. Findings were much more significant for the
log-transformed analyses shown in Table 5-10. The two control groups showed the largest
relative differences, with P < 0.0001. The estimated Phase 1 effect of CIN SEI(P) versus
CIN NT(M) was 1.128 or 200% (P = 0.0098), and CIN SEI(P) versus CIN NT(G) was
-0.775 (P = 0.0046). CIN I-SE(D) and CIN I-SE(F) also had significantly more reduction
than CIN NT(M), and significantly less than CIN NT(G) during Phase 1.
Phase 1 effects for 18-41 month old children were consistently much smaller and much
less significant than for 9-17 month old children. The most nearly significant Phase 1 results
for 18-41 month old children were differences between CIN NT(G) and CIN NT(M) of
9.21 /tg/dL (P = 0.072) and between CIN I-SE(F) and CIN NT(M) of 8.41 /*g/dL
5-92
-------
(P = 0.095). The CIN NT(G)-CIN NT(M) effect for Phase 1 was also nearly significant for
18-41 month children on a log scale, shown in Table 5-10 (P = 0.067).
Phase 1 effects for children of ages at least 42 months showed a very different pattern
than for the younger children. Blood lead decreased less in CIN SEI(P) than in any other
neighborhood. The negative effect of CIN SEI(P) versus CIN NT(G) was -1.97 /ig/dL
(P = 0.20), of CIN SEI(P) versus CIN NT(M) was -4.55 pig/dL (P = 0.017), of
CIN SEI(P) versus CIN I-SE(D) was -4.34 /*g/dL (P = 0.0028), and of CIN SEI(P) versus
CIN I-SE(F) was -2.99 /*g/dL (P = 0.082). Even the sign of the CIN NT(G)-CIN NT(M)
difference was negative hi the oldest age group. Table 5-10 shows the same pattern of
Phase 1 effects using log blood lead of the older children, with similar statistical
significance.
Phase 2 results were consistently non-significant within each age stratum, as shown in
Table 5-11.
Effects of Blood Lead Truncation in Cincinnati
The range of blood lead concentrations in the Cincinnati study was not constrained by
study design and was much larger than in the Boston study. Therefore, truncation of
Cincinnati Round 1 blood lead to the corresponding Boston range (7 to 24 /xg/dL) greatly
reduced the sample size of the truncated data set. For children age 9 to 17 months, the
sample size was reduced from 69 children for the whole data set to 33 in the truncated data
set, with only 15 children remaining in the data set truncated to 10 to 19 jwg/dL. Large and
highly significant differences between CIN NT(G) and CIN NT(M), CIN NT(G), and
CIN SEI, CIN NT(M) and CIN SEI that were found in the whole sample of 9 to 17 month
old children completely lost statistical significance in the truncated samples, and their results
are not reported here.
Fewer children were lost to truncation for ages 18-41 months, and the findings were
statistically significant. Table 5-12 shows the results for Phase 1 effects. The effect size for
CIN NT(G) versus CIN NT(M) Phase 1 blood lead reduction hi the full sample was
9.21 /ig/dL (P = 0.072), which was the same size but less significant (P = 0.10) hi the 7 to
24 pig/dL truncation. However, the effect size for CIN NT(G) versus CIN NT(M) was
larger and more significant hi the smaller data sets: 11.76 pg/dL (P = 0.053) hi the 10 to
5-93
-------
TABLE 5-12. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
CINCINNATI STUDY: EFFECT OF TRUNCATION BETWEEN ROUNDS 1 AND 4
Truncation Category
Study Group
Abate Versus Control
All
7-24 pg/dL 10-19 /ig/dL 10-24 /tg/dL
Abate
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
Reduction in Blood Lead (Er) Age 9-17 Months
(N=69)
13.10s3
-5.62S1
7.481T
-1.57
-0.91
Reduction
(N=80)
9.211T
-0.97
8.24M
0.10
-0.26
Reduction
(N=70)
-2.57M
-1.97M
-4.55S1
-4.34s2
-2.991T
(N=33)
NC
-2.82
NC •
-2.45
3.02
in Blood Lead
(N=67)
9.21M
-0.95
8.26M
1.21
-0.81
in Blood Lead
(N=47)
-1.61
-2.73
-4.341T
-5.17s1
-2.06
(N=15)
NC
0.73
NC
-0.86
-2.83
(Er) Age 18-41
(N=38)
11.761T
-3.52
8.24M
1.86
-0.82
(Er) Age 42+
(N=31)
-1.44
-1.74 -
-3.17
-3.12
-0.02
Months1'2
(N=43)
11.111T
-2.35
8.76M
2.40
-0.33
Months1'2
(N=36)
-1.49
-1.70
-3.19
-4.64M
-0.07
are /tg/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
19 jiig/dL truncation sample, 11.11 pig/dL (P = 0.052) in the 10 to 24 /wg/dL truncation
sample. The CIN SEI versus CIN NT(G) effect size was larger and more negative in the
truncated data sets, but the effects were still not statistically significant. The CIN SEI versus
CIN NT(M) Phase 1 effect was little changed hi either magnitude or statistical significance
by truncation. Similar effects were obtained hi the log-transformed analyses, not shown.
5-94
-------
The Phase 2 results in Table 5-13 showed a few differences, but none of the truncation
effects were statistically significant for 18-41 month old children.
TABLE 5-13. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
CINCINNATI STUDY: EFFECT OF TRUNCATION
BETWEEN ROUNDS 1 AND 4
Truncation Category
Study Group
Abate Versus Control
All
7-24
10-19
Abate
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
t.n.
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
Reduction in Log
(N=69)
1.903s4
-0.775s2
1.128s2
-0.142
-0.187
Reduction in Log
(N=80)
0.8041T
-0.199M
0.605M
0.007
-0.048
Reduction in Log
(N=70)
-0.129
-0.346s1
-0.475s1
-0.409s2
-0.3361T
Blood Lead (Er) Age
(N=33)
NC
-0.153
NC
-0.028
0.421
Blood Lead (Er) Age
(N=67)
0.791M
-0.160
0.631
0.075
-0.091
Blood Lead (E,.) Age
(N=47)
-0.064
-0.332M
-0.3961T
-0.506s1
-0.202
9-17 Months1-2
(N=15)
NC
-0.006
NC
-0.126
-0.194
18-41 Months1-2
(N=38)
NC
NC
NC
NC
NC
42+ Months1'2
(N=31)
NC
NC
NC
NC
NC
are log /*g/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
5-95
-------
The effects of blood lead truncation in children of ages 42+ months, shown in
Tables 5-14 and 5-15 was usually a modest reduction hi the estimated size of the effect, but a
large attenuation of statistical significance. This may be largely attributable to reduction hi
sample size. The Phase 1 reduction hi blood lead for CIN SEI remained smaller than for
any other neighborhood, but the statistical significance of the difference virtually
disappeared. The difference between the two control neighborhoods also lost any statistical
significance. The pattern of findings was also the same for log-transformed blood lead, not
shown here. The Phase 2 results, which showed only a few small and marginally significant
differences among neighborhoods hi the whole sample, exhibited no significant differences hi
the truncated samples.
5.5.3 Baltimore Repeated Measures Analysis of Variance (ANOVA)
The Baltimore study carried out soil lead abatement (without ulterior dust lead abatement) hi
one neighborhood, Lower Park Heights. A small number of houses hi this treatment group
were not abated, almost all of which had no single soil sample above about 500 pg/g. In
other words, the non-abated residences hi Area 1 had a maximum soil lead concentration less
than about 500 /xg/g, whereas almost all of the abated residences had maximum soil lead
above about 500 /tg/g. For this reason, we used the control neighborhood of Walbrook
Junction, where all properties had at least one soil measurement above 500 jtig/g, and the
unabated properties of lower Park Heights as separate controls rather than combining them as
in the Baltimore report. The study group hi Lower Park Heights is denoted BAL SP and the
control groups are denoted BAL PI for the Walbrook Junction group and BAL P2 for the
low soil lead houses hi Lower Park Heights. As hi Cincinnati, because of possible
neighborhood differences, comparisons of BAL SP with each of the reference groups seems
appropriate. We used Round 3, the last preabatement blood lead sample time, as the basis
for comparison even though the measurements were made in February 1990, about 6 months
before the soil abatement, about 11 months before the first postabatement blood lead at
Round 4 hi 1991, and about 19 months before the Sept. 1991 blood lead sample at Round 6.
It was not clear which pre-post comparisons were more appropriate, Round 3 versus Round 4
or Round 6. Both are reported here. Table 5-16 shows the results of the Round 3 versus
Round 4 and Round 6 comparison of group BAL SP versus the two controls. None of the
5-96
-------
TABLE 5-14. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
CINCINNATI STUDY: EFFECT OF TRUNCATION
BETWEEN ROUNDS 4 AND 7
Truncation Category
Study Group
Abate Versus Control
All
7-24 /xg/dL 10-19 /ig/dL 10-24 /ig/L
Abate
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
Reduction
(N=69)
-4.56
1.06
-3.50
-1.13
1.42
Reduction
(N=80)
-1.75
0.27
-1.48
0.78
0.23
Reduction
(N=70)
in Blood Lead
(N=33)
NC
-0.32
NC
3.54
3.18
in Blood Lead
(N=67)
-1.05
-0.48
-1.53
1.18
0.39
in Blood Lead
(N=47)
-2.86
2.83
-0.02
1.18
0.59
(Er) Age 9-17 Months1'2
(N=15)
NC
2.68
NC
6.30
0.68
(Er) Age 18-41
(N=38)
2.42
0.75
-1.67
1.40
-1.96
(Er) Age 42+
(N=31)
-3.97
4.44
0.47
1.75
1.02
Months1'2
(N=43)
0.06
-1.13
-1.07
1.99
-0.49
Months1'2
(N=36)
-4.00
4.45
0.45
1.81
1.00
are ftg/dL Pb in blood.
2 Age is age in months at time of Round 1 blood sample.
effect estimates for the large control group BAL P2 are statistically significant, nor is there
any significant effect against BAL PI using log blood lead, as shown in Table 5-16. The
comparisons of BAL SP versus BAL P2 in Table 5-16 find a marginally significant
difference (P = 0.16) between BAL SP and BAL P2 between Round 3 and Round 4, but the
5-97
-------
TABLE 5-15. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
CINCINNATI STUDY: EFFECT OF TRUNCATION
BETWEEN ROUNDS 4 AND 7
Truncation Category
Study Group
Abate Versus Control
All
7-24 Q*g/dL) 10-19 (/ig/dL) 10-24 /tg/L
Abate
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
Reduction in Log Blood Lead (Er) Age
(N=69)
Reduction
(N=80)
-0.188
0.128
-0.059
0.075
0.062
Reduction
(N=70)
-0.304M
0.210**
-0.094
0.123
0.103
(N=33)
NC
-0.179
NC
0.123
-0.063
in Log Blood Lead
(N=67)
-0.113
0.041
-0.072
0.123
0.082
in Log Blood Lead
(N=47)
-0.292
0.362
0.070
0.272
0.194
9-17 Months1'2
(N=15)
NC
0.251
NC
0.561
-0.075
(Er) Age
(N=38)
NC
NC
NC
NC
NC
(^ Age
(N=31)
-0.402
0.534
0.132
0.382
0.215
18-41 Months1-2
(N=43)
0.052
-0.099
-0.047
0.183
-0.020
42+ Months1'2
(N=36)
-0.403
0.534
0.131
0.363
0.214
are log ^g/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
effect is not statistically significant by Round 6. There is no significant effect after
transforming to log blood lead, as shown in Table 5-16.
5-98
-------
TABLE
5-16. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
BALTIMORE STUDY: EFFECT OF AGE
Study Group
Abate
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BAL SP
Control
BAL PI
BAL P2
BAL, PI
BALP2
BAL, PI
BALP2
BAL PI
BALP2
Age Group1
All Ages < 18 Months 18-41 Months
(N=463) (N=16) (N=88)
Reduction in Blood Lead (Er) Between Rounds 3
0.07 5.70 0.22
l.yyM ... 3.23
Reduction in Blood Lead (Er) Between Rounds 3
-0.54 0.39 -0.44
0.671T — 5.20M
Reduction in Log Blood Lead (E,) Between Rounds
-0.012 NC NC
-0.002 NC
Reduction hi Log Blood Lead (EJ Between Rounds
-0.013 NC NC
0.006 NC
42+ Months
(N=161)
andtf
0.06
1.74M
ande1
-0.18
0.55
3 and 42
0.002
0.124
3 and62
NC
*Age is age hi months at tune of Round 1 blood sample.
2Units are jttg/dL Pb hi blood.
3Units are log /ig/dL Pb hi blood.
Age Effects
The design of the Baltimore study excluded most children younger than 18 months, and
no significant effects were found in the few analyses that could be carried out with only
16 children at most in the three groups. There were 88 children hi the age group
18-41 months, with a marginally significant effect between BAL SP and BAL P2 from
Round 3 to Round 6 (P = 0.17), as found hi Table 5-16. There were 161 children in the
oldest group, and one of the effects was marginally significant for ages 42+ months, BAL
SP versus BAL P2 from Round 3 to Round 4 (P = 0.12).
Effects of Blood Lead Truncation
Results are shown for Round 3 versus Round 6 comparisons only. Because the range
of the blood lead concentrations at Round 3 hi the Baltimore study was not constrained by
study design, truncation of blood lead to the interval 7 to 24 /-ig/dL or smaller resulted in a
substantial reduction hi sample size. At ages 18-41 months, the number of children dropped
from 88 for the full sample to 64 children hi the range 7 to 24 /tg/dL, and to 32 children hi
5-99
-------
the range 10 to 19 /*g/dL. As shown hi Table 5-17, the effects of BAL SP versus BAL PI
were somewhat larger. A similar pattern was seen hi the log blood lead analyses hi
Table 5-17.
TABLE 5-17. REPEATED MEASURES ANALYSIS OF VARIANCE FOR
BALTIMORE STUDY: EFFECT OF TRUNCATION
BETWEEN ROUNDS 3 AND 4
STUDY GROUP
ABATE
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
CONTROL
BAL PI
BALP2
BAL PI
BALP2
BAL PI
BALP2
BAL PI
BALP2
BAL PI
BALP2
BAL PI
BALP2
TRUNCATION CATEGORY1-2-3
ALL 10-19 10-24 7-19 7-24
<18:(N=16) <18:(N=2) <18: (N=4) <18: (N=5) <18: (N=7)
18-42: (N=88) 18-42: (N=32) 18-42: (N=42) 18-42: (N=54 18-42: (N=64)
>42: (N=161) >42: (N=47) >42: (N=53) >42: (N=110) >42: (N=120)
REDUCTION IN BLOOD LEAD (E,.) FOR AGE < 18 MONTHS
5.70 14.7
REDUCTION IN BLOOD LEAD (E,) FOR AGE 18-41 MONTHS
0.22 -1.54 -1.56 -0.52
3.23
REDUCTION IN BLOOD LEAD (Er) FOR AGE 42+ MONTHS
0.06 -0.43 -0.22 -0.17 0.12
1.74M 1.68 1.39
REDUCTION IN LOG BLOOD LEAD (Er) FOR AGE < 18 MONTHS4
REDUCTION IN LOG BLOOD LEAD (E,) FOR AGE 18-41 MONTHS
-0.300
REDUCTION IN LOG BLOOD LEAD (Er) FOR AGE 42+ MONTHS
0.002 0.098 0.087 0.046 -0.020
0.124 ' 0.110 0.123
JUnits are /zg/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
3No optimal solution for missing cells in table.
4Units are log pg/dL Pb in blood.
5-100
-------
The blood lead truncation also reduced the sample size for older children, from 161
children in the full sample to 116 children in the range of 7 to 24 /-ig/dL at Round 3, and
only 47 children in the range 10 to 19 /ig/dL. The effect size for BAL SP versus BAL PI
was larger in the truncated data sets, but not even marginally significant for blood lead in the
range 7 to 24 /xg/dL (smaller P = 0.21). There was, however, a suggestion of a reduction
in BAL SP compared to the Area 1 control group, BAL P2 for children age 18 months and
older, and possibly even a larger benefit relative to BAL PI hi the children < 18 months.
5.5.4 Boston Repeated Measures Analysis of Covariance (ANCOVA)
Many of the important covariate effects of age, race/ethnicity, and gender could be
assessed by stratifying the data set, but the possible changes in response associated with
continuous variables such as lead hi soil and dust required a more general approach. Some
preliminary results suggested that there may be systematic differences hi response to changes
in environmental lead across different treatment groups. These findings were explored more
systematically in the longitudinal structural equation models (SEM) discussed hi Section 5-6.
We will briefly report the more significant findings from the ANCOVA analyses. The
analyses were all carried out using log-transformed soil lead concentrations, dust lead
concentrations, or dust lead loadings, because of the highly skewed distributions of these
environmental lead variables. We also used log-transformed blood leads.
The results of the Phase 1 analyses are shown hi Tables 5-18, 5-19, and 5-20. The
blood lead effect (Er) for repeated measures ANCOVA is calculated hi the same manner as
discussed previously for RM ANOVA. The change hi the dust or soil lead regression
coefficient for RM ANCOVA may be calculated as
Table 5-18 shows the RM ANCOVA for log dust lead concentration, stratified by age.
There were some marginally significant differences hi the relationship between blood lead
and dust lead concentration. The relationship, which was initially quite flat in all groups at
Round 1, became much sharper hi all groups at Round 3. However, the increasing slope of
the log blood lead versus log dust lead concentration grew much more strongly hi the dust
5-101
-------
TABLE 5-18. REPEATED MEASURES ANALYSIS OF COVARIANCE FOR
BOSTON STUDY: EFFECT OF AGE AND LOG DUST LEAD CONCENTRATION
REDUCTION ON LOG BLOOD LEAD (Er) BETWEEN ROUNDS 1 AND 3
Age Group1-2
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
All Ages
(N=142)
0.199
1.330**
-1.131
Covariate: Log Dust
-0.008
-O.ISQM
0.173M
9-17 Months
(N=17)
0.717
3.339
-2.622
18-41 Months
(N=97)
0.951
1.524M
-0.573
42+ Months
(N=28)
-2.421
-3.622
1.201
Pb Concentration3
-0.095
-0.493
0.398
-0.102
-0.199M
0.097
0.342
0.507
-0.165
'Units are log /ig/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
3Units are log jig/dL Pb in blood per log j«g/g Pb in dust.
abatement group BOS PI-S between Round 1 and Round 3 than hi either the soil abatement
group BOS SPI or the control group BOS P-S. The change was most significant in the
largest group, 18-41 months of age, but qualitatively similar in younger children. When log
dust lead loading was used as a covariate, there were virtually no significant differences hi
blood lead response among treatment groups except for the youngest group, ages 9-17
months, where group BOS PI-S showed a relatively much more significant increase hi the
slope of the relationship (P = 0.095) than did BOS SPI or BOS P-S hi Table 5-19. There
were no significant effects on the relationship when log soil lead was used as a covariate.
When both dust lead and soil lead were used as covariates, the effect sizes for slope were
essentially those of the dust lead model, as shown hi Table 5-20, with soil lead effects adding
little to the predictive power of the model.
The results of the Phase 2 analyses are shown hi Tables 5-20 and 5-21. Table 5-20
shows the RM ANCOVA for log dust lead concentration, stratified by age. There were
some highly significant differences among treatment groups hi the relationship between blood
lead and dust lead concentration. The relationship of log blood lead versus log dust lead
5-102
-------
TABLE 5-19. REPEATED MEASURES ANALYSIS OF COVARIANCE FOR
BOSTON STUDY: EFFECT OF AGE AND LOG DUST LEAD AND
SOIL LEAD CONCENTRATION ON REDUCTION IN LOG BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 3
Age Group1'2
Study Group All Ages 9-17 Months 18-41 Months 42+ Months
Abate Versus Control (N=142) (N=17) (N=97) (N=28)
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
SPI
PI-S
SPI
SPI
PI-S
SPI
SPI
PI-S
SPI
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
P-S
P-S
PI-S
Covariate:
P-S
P-S
PI-S
Covariate:
P-S
P-S
PI-S
Log Dust Lead Concentration3
Log Soil Lead Concentration4
1.
0.
0.
-0.
-0.
0.
-0.
0.
-0.
866M
981
884
096
206M
111
103
076
178
-2.
-5.
3.
0.
0.
-0.
-0.
0.
-0.
013
755
742
416
438
022
151
362
513
its are log ftg/dL Pb in blood.
2 Age is age in months at time of Round 1 blood sample.
3Units are log /ig/dL Pb in blood per jtg/g Pb in dust.
4Units are log /*g/dL Pb in blood per log jtg/g Pb in soil.
concentration flattened much more strongly in the Phase 2 soil abatement group BOS PI-S
between Round 3 and Round 4 than in either the soil abatement group BOS P-S or the Phase
2 non-abatement group BOS SPI. The change was most significant (P = 0.0019 for BOS
PI-S versus BOS P-S, P = 0.029 for BOS PI-S versus BOS SPI) hi the largest group,
18-41 months of age, but not separately estimatable hi younger children. When log dust lead
loading was used as a covariate, there were similar significant differences in blood lead
response among treatment groups in the two older age groups, where group BOS PI-S
showed a relatively much more significant decreases hi the slope of the relationship than did
either BOS SPI or P in Table 5-21. For 18-41 month old children, the effects on log blood
lead versus log dust lead loading shown hi Table 5-21 had significance levels P = 0.017 for
BOS PI-S versus BOS SPI and P = 0.033 for BOS PI-S versus BOS P-S respectively. There
were no significant effects on the relationship when log soil lead was used as a covariate.
5-103
-------
TABLE 5-20. REPEATED MEASURES ANALYSIS OF COVARIANCE FOR
BOSTON STUDY: EFFECT OF AGE AND LOG DUST LEAD CONCENTRATION
ON REDUCTION IN LOG BLOOD LEAD (Er) BETWEEN ROUNDS 3 AND 4
Age Group1'2
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
All Ages 9-17 Months 18-41 Months 42+ Months
(N=142) (N=17) (N=97) (N=28)
0.731 -6.685
-3.575s1
-4.306s1
2.828M
-4.123s1
-6.950s2
-3.720
0.077
4.797
Covariate: Log Dust Lead Concentration3
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
-0.098 1.035
0.558S1
0.656S1
-0.432?
0.628?
1.060?
0.582
-0.101
-0.683
are log /ig/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
3Units are /ig/dL Pb in blood per unit logarithm difference in dust lead concentration. Equivalent to decimal
percent change (see Equation 5-20).
TABLE 5-21. REPEATED MEASURES ANALYSIS OF COVARIANCE
FOR BOSTON STUDY: EFFECT OF AGE AND LOG DUST LEAD LOADING
ON REDUCTION IN BLOOD LEAD (Er) BETWEEN ROUNDS 3 AND 4
Age Group1'2
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
All Ages
(N=128)
0.120
-0.607
-0.727M
Covariate: Log
-0.013
0.2671T
0.2801T
9-17 Months 18-41 Months 42+ Months
(N=15) (N=89) (N=24)
NC
NC
NC
Dust Lead Loading
NC
NC
NC
-0.393
-1.2891T
-0.896M
0.099
0.459s1
0.360s1
0.764
2.178
-2.943M
0.106
0.670M
0.776M
are log /ig/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
3Units are /tg/dL Pb in blood per unit logarithm difference hi dust lead concentration. Equivalent to decimal
percent change (see Equation 5-20).
5-104
-------
When both dust lead and soil lead were used as covariates, the effect sizes for slope were
essentially those of the dust lead model, with soil lead effects adding little to the predictive
power of the model.
When stratified analyses by race were carried out, the effects on dust lead slopes were
larger and more significant among Afro-American children than for the sample as a whole.
Table 5-22 is analogous to Table 5-18. The effect on dust lead slope for BOS PI-S versus
BOS P-S between round 1 and round 3 is large, -0.528, and is highly significant
(P = 0.0078). The effect on slope between BOS SPI and BOS PI-S is also large 0.386, and
is nearly significant (P = 0.058). The significant whole-sample dust lead slope effects
reflect the significant effects hi the 18-41 month age group. However, slope effects for
Afro-American children shown in Table 5-23 are nearly all non-significant when log dust
lead loading is used as a covariate. Soil lead concentration is also not predictive for these
children. When both log dust lead concentration and soil lead concentration are used as
covariates, as shown hi Table 5-24, there are large and highly significant differences in
Phase 1 dust lead slope effects hi group BOS PI-S, and large marginally significant soil lead
slope effects as well, although these may reflect the collinearity between soil lead and dust
lead in group BOS PI-S.
Stratified analyses for Phase 2 effects hi Afro-American children are shown hi
Tables 5-25 and 5-26. Table 5-25 shows a large decrease hi slope hi group BOS PI-S, just,
as hi Table 5-20, with P = 0.0349 for BOS PI-S versus BOS SPI hi children of ages
18-41 months. However, hi Table 5-26, slope effects for log dust lead loading are relatively
large and significant, P = 0.015 for BOS PI-S versus BOS SPI and P = 0.071 for BOS PI-S
versus BOS P-S hi 18-41 month old Afro-American children.
In summary, these analyses suggest that there were some fairly substantial differences
hi the relationship between blood lead and dust lead during successive phases of the Boston
study. The relationship between blood lead and dust lead was very flat in Round 1, and
suggests that there may have been an attenuation hi the real relationship due to selection or
recruitment effects. After the Phase 1 abatement, the relationship became much more
evident hi the group BOS PI-S that received dust abatement, but not soil abatement in Phase
1 of the study. When group BOS PI-S received soil abatement during Phase 2 of the Boston
study, much of the apparent relationship seems to have been attenuated or reversed.
5-105
-------
TABLE 5-22. REPEATED MEASURES ANALYSIS OF COVARIANCE
FOR BOSTON STUDY: EFFECT OF AGE AND LOG DUST LEAD
CONCENTRATION ON REDUCTION IN LOG BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 3 FOR AFRO-AMERICAN CHILDREN
Study Group
Abate Versus Control
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
All Ages
(N=71)
0.549
3.008s1
-2.46011
Covariate: Log Dust
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
-0.056
-0.413s1
0.3571T
Age Group1'2
9-17 Months 18-41 Months
(N=ll) (N=44)
NC 1.223
NC 3.865s2
NC -2.6421T
Lead Concentration3
NC -0.142
NC -0.528s2
NC 0.3861T
42+ Months
(N=16)
-3.269
-6.539
3.270
0.448
0.872
-0.423
"Units are log /tg/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
3Units are /tg/dL Pb in blood per unit logarithm difference in dust lead concentration. Equivalent to decimal
percent change (see Equation 5-20).
TABLE 5-23. REPEATED MEASURES ANALYSIS OF
COVARIANCE FOR BOSTON STUDY: EFFECT OF AGE AND LOG
DUST LEAD LOAD ON REDUCTION IN LOG BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 3 FOR AFRO-AMERICAN CHILDREN
Age Group1'2
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
All Ages
(N=71)
0.109
0.439
-0.330
Covariate: Log
0.015
-0.096
0.111
9-17 Months 18-41 Months 42+ Months
(N=ll) (N=44) (N=16)
NC
NC
NC
Dust Lead Load3
NC
NC
NC
0.366
0.820*
-0.453
-0.035
-0.153 +
0.118
5.074
0.597
4.478
-1.261
-0.222
-1.038
'Units are log /tg/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
3Units are /tg/dL Pb in blood per unit logarithm difference in dust lead concentration. Equivalent to decimal
percent change (see Equation 5-20).
5-106
-------
TABLE 5-24. REPEATED MEASURES ANALYSIS OF COVARIANCE FOR
BOSTON STUDY: EFFECT OF AGE AND LOG DUST LEAD CONCENTRATION,
SOIL LEAD CONCENTRATION ON REDUCTION IN LOG BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 3 FOR AFRO-AMERICAN CHILDREN
Study Group
Abate Versus Control
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
Age Group1'2
All Ages 9-17 Months 18-41 Months
(N=71) (N=ll) (N=44)
0.609 NC 1.672
1.916 NC 1.326
-1.307 NC 0.346
42+ Months
(N=16)
NC
NC
NC
Covariate: Log Dust Lead Concentration3
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
-0.066 NC -0.169
-0.436s1 NC -0.574s2
0.3701T NC 0.405s1
NC
NC
NC ,
Covariate: Log Soil Lead Concentration4
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
0.030 NC 0.003
0.163 NC . 0.370M
-0.132 NC -0.367M
NC
NC
NC
1Units are log /tg/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
3Units are jtg/dL Pb in blood per unit logarithm difference in dust lead concentration. Equivalent to decimal
percent change (see Equation 5-20).
4Units are jtg/dL Pb in blood per unit logarithm difference in soil lead concentration. Equivalent to decimal
percent change (see Equation 5-20).
Additional analyses using SEM to further evaluate the changing patterns in the soil lead -
dust lead - blood lead pathways hi different treatment groups seemed to be useful hi
understanding some of these effects.
5.5.5 Cincinnati Repeated Measures Analysis of Covariance (ANCOVA)
The repeated measures analyses for Cincinnati were directed towards assessing the role
of longitudinal group differences associated with different dust indices. The basic Cincinnati
model was run with each of the logarithms of floor dust concentration and loading, entry dust
concentration and loading, window dust concentration and loading. The results are shown hi
Table 5-27.
5-107
-------
TABLE 5-25. REPEATED MEASURES ANALYSIS OF COVARIANCE FOR
BOSTON STUDY: EFFECT OF AGE AND LOG DUST LEAD LOADING ON
REDUCTION IN LOG BLOOD LEAD (Er) BETWEEN
ROUNDS 3 AND 4 FOR AFRO-AMERICAN CHILDREN
Age Group1'2
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
All Ages
(N=64)
0.324
-0.721
-1.045
Covariate: Log
-0.042
0.316M
0.357M
9-17 Months 18-41 Months 42+ Months
(N=8) (N=40) (N=16)
NC
NC
NC
Dust Lead Loading
NC
NC
NC
-0.683
-2.045s1
-1.362M
0.214
0.692s1
0.4781T
0.961
-3.000M
-3.961M
-0.128M
0.941M
1.0691T
'Units are log jtg/dL Pb in blood.
2Age is age in months at time of Round 1 blood sample.
3Units are reduction of natural logarithm /*g/dL Pb hi blood per unit logarithm difference in dust lead
concentration. Equivalent to decimal percent change (see Equation 5-20).
TABLE 5-26. REPEATED MEASURES ANALYSIS OF COVARIANCE FOR
BOSTON STUDY: EFFECT OF AGE AND LOG DUST LEAD CONCENTRATION
ON REDUCTION IN LOG BLOOD LEAD (Er) BETWEEN
ROUNDS 3 AND 4 FOR AFRO-AMERICAN CHILDREN
Age Group1'2
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
All Ages
(N=64)
0.642
-3.629**
-4.271M
Covariate: Log Dust
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
-0.050
0.5881T
0.639M
9-17 Months
(N=8)
NC
NC
NC
18-41 Months
(N=40)
0.340
-5.575s1
-5.915M
42+ Months
(N=16)
0.150
6.796
6.646
Lead Concentration3
NC
NC
NC
-0.034
0.863s1
0.897S1
0.063
-0.886
-0.949
'Units are log /tg/dL Pb hi blood.
2Age is age hi months at tune of Round 1 blood sample.
3Units are reduction of natural logarithm pg/dL Pb in blood per unit logarithm difference in dust lead
concentration. Equivalent to decimal percent change (see Equation 5-20).
5-108
-------
TABLE 5-27. REPEATED MEASURES ANALYSIS OF COVARIANCE
FOR CINCINNATI STUDY: REDUCTION IN BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 4
Study Group
Log
Floor Dust
Concentration
Log
Entry Dust
Concentration
Abate Versus Control
CIN NT(G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT(G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT(M)
CIN NT(G)
CIN NT(M)
CIN I-SE(D)
CIN I-SE(F)
CIN NT(M)
CIN NT(G)
CIN NT(M)
CIN I-SE(D)
CIN I-SE(F)
0.539
0.429
0.969
0.949
0.416
0.015
-0.134
-0.118
-0.189
-0.092
2.208
-0.467
1.741
-0.283
-0.593
-0.244
0.010
-0.234
-0.001
0.067
Log
Window Dust
Concentration
Intercept Effect1
-3.36211
0.078
-3.284M
-1.494
-0.130
Covariate Effect
0.540s1
-0.055
0.4861T
0.187
0.006
Log
Floor
Dust Pb
Loading
1.605M
-0.188
1.417M
0.355
-0.701
-0.277
-0.028
-0.305
-0.100
0.112
Log
Entry
Dust Pb
Loading
1.417"
-0.7551T
0.662
-1.0951T
-0.528
-0.134
0.077
-0.057
0.138M
0.060
Log
Window
Dust Pb
Loading
-0.200
-0.775M
-0.975
-1.846s1
-0.337
0.102
0.055
0.158
0.195s1
0.032
are log ftg/dL Pb in blood.
2Units are reduction of natural logarithm /tg/dL Pb in blood per unit logarithm difference hi dust lead
concentration. Equivalent to decimal percent change (see Equation 5-20).
There was no indication of any strong change hi the response of log blood lead to floor
dust or entry dust during the study. There was, however, an indication that the relationship
of blood lead to window dust changed between some of the neighborhoods during Phase 1 of
the study. Table 5-27 shows a statistically significant difference hi change in log window
dust concentration regression coefficients between CIN NT(G) and CIN NT(M) (P =
0.0496) and a nearly significant difference between CIN SEI(P) and CIN NT(M) (P = 0.08),
while there was almost no difference hi the longitudinal change between CIN SEI(P) and
CIN NT(G) responses to window dust lead (P = 0.95). This suggests that CIN NT(M) was
a neighborhood hi which some substantial external change occurred that was manifested as a
difference hi the response to window dust. There was also a significant effect in covariate
response for window dust lead loading hi CIN SEI(P) vs CIN I-SE(D) (P = 0.025), which
was not manifested as strongly as a change hi response to window dust lead concentration;
perhaps changes hi window dust loadings were an important factor. The window dust
covariate effects that were significant and positive were associated with negative changes hi
5-109
-------
the intercept for the window dust lead concentration model, P = 0.095 for CIN NT(G) vs
CIN NT(M) and P = 0.11 for CIN SEI(P) vs CIN NT(M), and in the intercept for the
window dust lead loading model, P = 0.014 for CIN SEI(P) vs. CIN I-SE(D).
There were some significant differences in the intercept model where the log entry dust
lead loading was used as a covariate, between CIN NT(G) and CIN NT(M) (P = 0.0494),
CIN SEI(P) and CIN NT(G) (P = 0.08), and between CIN SEI(P) and CIN I-SE(D)
(P = 0.06). The only covariate response change of even marginal significance was for CIN
SEI(P) vs CIN I-SE(D) (P = 0.11). The lack of any significant effects of any sort when log
of entry dust concentration was used as a covariate suggests that changes in entry dust
loading may have occured hi several of these neighborhoods.
These analyses suggested that changes hi response to window dust lead may have
played a role in blood leads among the Cincinnati neighborhoods. A more detailed
evaluation of some dust lead pathway models using LSEM showed some modest indications
of changes in the pathway components from window dust to floor dust.
5.5.6 Baltimore Repeated Measures Analysis of Covariance (ANCOVA)
We have not carried out either repeated measures ANCOVA models or structural
equation models for the Baltimore study because of the limited environmental data hi the
Baltimore study. The results of the Boston and Cincinnati studies have shown that there can
be substantial changes hi dust lead concentrations from one round to another, both in abated
and non-abated residences. Since there are strong and statistically significant relationships
between blood lead and current dust lead measurements hi all rounds hi the Cincinnati study
and in all rounds after Round 1 hi Boston, it appears necessary to have dust lead
measurements that reasonably characterize each round of blood lead measurements.
However, the Baltimore study did not collect any post-abatement dust lead measurements for
non-abated residences, nor long-term post-abatement dust lead measurements hi abated
residences. The often large intervals between pre-abatement dust lead measurements and
Round 3 blood lead measurements may not even provide adequate information about baseline
exposures. While soil lead concentrations in non-abated residences appear to change very
slowly over time, some post-abatement soil lead concentrations may increase because of
recontamination, and additional post-abatement soil lead data hi both abated and non-abated
5-110
-------
residences would have been desirable. In earlier drafts of this report, EPA evaluated
several models in which the dust lead and soil lead post-abatement data that were not
available were imputed by assuming that the post-abatement environmental measurements
were equal to the pre-abatement measurements. Based on these earlier assessments and on
reviewer comments, we conclude that this approach does not provide adequate information
about actual post-abatement environmental exposures. Therefore, we were unable to use
these data for tune-varying covariate adjustment models for the Baltimore study. Repeated
measures ANOVA models were stratified for tune-constant covariates such as age and
gender, and our conclusions are based on these analyses.
5.6 COMPARISON USING LONGITUDINAL STRUCTURAL
EQUATION MODELS
5.6.1 Boston Study Longitudinal Structural Equation Models
Recall from Section 5.1.1.2 that the equations for the longitudinal structural equation
model are
ir = Cgr + ZirDgr
(5-18)
(5-19)
and that the model adjusts for the simultaneous fitting of multiple relationships. In a scheme
of thirty-two models run by the SAS PROC MIXED procedure, shown on Table 5-28, each
model was run with a component of the structural equation, either a common intercept term
(Ggi or Cgi) or a pathway regression coefficient (Fgj, Bgj, Lgj, or Dgi), for Round 3 blood
lead or Round 3 dust lead. In some cases, these coefficients were separated into three
intercept terms or three regression coefficients, one for each of the three treatment groups
BOS SPI, BOS PI-S, or BOS P-S. The best-fitting of the 32 models, were models 1, 2, 10,
11, 17, and 30, and these were used for the reported output hi Tables 5-29 through 5-30.
The longitudinal structural equation models allow effects adjusted for changes hi both
concentration (Xgr in Group g at Round r) and regression coefficients,
5-111
-------
TABLE 5-28. MODELS FOR TREATMENT GROUP EFFECTS IN BOSTON
LONGITUDINAL STRUCTURAL EQUATION MODELS
MODEL
NUMBER
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
BLOOD LEAD
INTERCEPT
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
GSPI. GPI_S,
GP-S
GSPI, Gpi_s,
GP-S
GP-S
GSPI. GPJ.S,
GP-S
GSPI. Gpi_s,
GP-S
Gp-s
SOIL-BLOOD DUST-BLOOD
COEFFICIENT COEFFICIENT
F B
FSPI> FPI-S' FP-S B
FSPI> FPI-S> FP-S BSPI> BPI-S' BP-S
FSPI> FPI-S> FP-S BSPI' BPI-S> BP-S
FSPI' FPI-S> FP-S BSPI, BPI-S' BP-S
FSPI» FPI-S» FP-S BSPI' BPI-S> BP-S
FSPI> FPI-S. FP-S B
FSPI> FPI-S> FP-S B
FSPI> FPI-S> FP-S B
F BSPI, BPK, Bp.s
F BSPI. EPI-S' BP-S
F BSPI> BPI-S» BP-S
F BSPI, BM. Bp^
F B
F B
F B
F B
FSPI> FPI-S> FP-S B
FSPI» FPI-S> FP-S BSPI> BPI-S' BP-S
FSPI> FPI-S> FP-S BSPI« BPI-S» BP-S
FSPI> FPI-S» FP-S BSPI' BPI-S' BP-S
FSPI> FPI-S» FP-S BSPI> BPI-S' BP-S
FSPI. FPI-S. Fp-S B
SOIL-DUST
MODEL
C, D
C, D
C, D
Cp p
SPI' PI-S' P-S
DSPJ, Dpi_s, "P-S
C, D
§H>' Sj, Si
CC* C*
SPI' ^-'PI-S' ^P-S
C, D
C.D
p c1 r1
^SPI' ^'PI-S' ^P-S
DSPI, DPI-S' DP-S
DSPI, ^PI-S> ^P-S
C, D
DSPI, ^PI-S> ^P-S
p p p
SPI' PI-S' P-S
C,D
C, D
C, D
C, D
§';; fe; Si
§H,' Dp^; §1
C, D
Cp p
SPI' PI-S» P-S
DSPI, Dpi-s> DP-S
WINDOW-DUST
MODEL
C, L
C, L
C, L
C, L
Cp P
SPI' V^PI-S> ^P-S
Lgpi, '-fl-S' ^P-S
Lsp;; %£: LPP^S
C,L
LSPI, LPI-S' '-P-S
p« pm „ p_ _
V'SPI' V'PI-S' ^P-S
C, L
C, L
Cp P
SPI> *-PI-S> *"P-S
LSPI, LpI-S' MP-S
LSH, 5s>' Si
C, L
Cp P
SPI> *-PI-S> *— P-S
Cp p
SPI' MPI-S' ^P-S
LSPI, Lpi^, Lp_s
C, L
C, L
C, L
C, L
Cp P
SPI> ^PI-S" ^P-S
LSPI. S^; Si
C.L
5-112
-------
TABLE 5-28 (cont'd). MODELS FOR TREATMENT GROUP EFFECTS IN BOSTON
LONGITUDINAL STRUCTURAL EQUATION MODELS
MODEL
NUMBER
24
25
26
27
28
29
30
31
32
BLOOD LEAD
INTERCEPT
GSPI, GPI_S, GP_s
GSPI. Gpi_s, GP_s
*-*spi. GFI.S, vjp_g
GSPI, GPI-S, GP^
GSPI, GPI_S, GF,_s
GSPI, GP1_s, GP_s
^SPI' ^*P1-S> Gp-S
GSPI, Gpi_s, GP_s
GSPI. GP1_s, GP^s
SOIL-BLOOD
COEFFICIENT
FSPI, FPI-S> FP-S
FSPI' FpI-S' FP-S
F
F
F
F
F
F
F
DUST-BLOOD
COEFFICIENT
B
B
BSPI. BPI_S, Bp_s
BSPI, Bpi_s> Bp^
BspLBp^Bp,
BSPI, BM, BP^
B
B
B
SOIL-DUST
MODEL
Cf P
SPI> PI-S* P-S
C, D
C,D
^SPI> -^PI-S' ^P-S
GSPI. CPI_S, CP^J
DSPI. DPi^, Dp^
C, D
^SPI' ^PI-S' ^P-S
DSPi, DPI_S, DP^
*-^SPI' *^PI-S» ^P-S
C, D
WINDOW-DUST
MODEL
GSPI, CPI_S, CP_s
LSPI, ^PI-S' ^P-S
Cp r1
SPI' ^PI-S' V'P-S
LSPI, *-"PI-S> -^P-S
C, L
C, L
LSPI, kpi_s, Lp^
LSPI, ^PI-S' '-'P-S
C.L
C/i /-i
SPI' V'PI-S» ^T-S
LSPI, LPI-S, Lp-s
Cfi /-<
SPI, V^PI-S* P-S
TABLE 5-29. LONGITUDINAL STRUCTURAL EQUATION MODELS
MODEL ASSESSMENT STATISTICS IN BOSTON STUDY
USING ESTIMATED BLOOD LEAD PERSISTENCE FACTOR
MODEL ASSESSMENT STATISTICS
Response Variable
Log Blood Lead RD3
Log Blood Lead RD1
Log Dust Lead Cone. RD3
Log Dust Lead Cone. RD1
All
Statistic
RMSE
RMSE
RMSE
RMSE
N*OBJ
MODEL
1
0.30089
0.37416
0.62147
0.79374
42.75
MODEL
2
0.30049
0.36512
0.62081
0.79092
41.86
MODEL
10
0.29796
0.33282
0.60797
0.78507
12.43
MODEL
11
0.29797
0.33284
0.62998
0.78513
10.95
MODEL
17
0.29696
0.38106
0.61399
0.78683
43.45
1RMSE = Root mean squared error.
5-113
-------
TABLE 5-30. LONGITUDINAL STRUCTURAL EQUATION MODELS FOR
BOSTON STUDY: REGRESSION COEFFICIENTS
USING ESTIMATED BLOOD LEAD PERSISTENCE FACTOR
Predictor
Variable
MODEL MODEL MODEL MODEL
1 2 10 11
MODEL
17
RESPONSE VARIABLE: BLOOD LEAD ROUND 3
ALL GROUPS
, BOSSPI
******* BOSPI-S
BOS P-S
ALL GROUPS
Soil Pb BOS SPI
ROUnd32 BOSPI-S
BOS P-S
ALL GROUPS
Dust Pb BOS SPI
Cone.
Rounds2 BOSPI-S
BOS P-S
Blood Lead Round I3
1.96 2.04 2.42 2.42
0.719 0.032 0.032
0.511
0.680
0.817
0.318 0.302
0.182 0.180
1.275 1.274
1.343 1.342
0.60881T 0.6060M 0.5890 0.5890
1.95
3.28
4.41
0.043
0.429
0.5931
RESPONSE VARIABLE: BLOOD LEAD ROUND 1
Intercept1
Soil Pb Round I2
Dust Pb Round I2
10.75s4 11.11s4 11.21s4 11.22s4
-0.291M -0.293M 0.151 0.149
0.524s1 0.449s1 0.038 0.038
10.63s4
-0.316M
0.574s1
RESPONSE VARIABLE: FLOOR DUST LEAD CONCENTRATION ROUND
1
Intercept4
Soil Pb Round 1s
Window Dust Pb Round I5
1826s4 1736s4 1408s2 1414s2
-0.204M -0.174M -0.073 -0.074
0.0684s4 0.0693s4 0.0714s4 0.0716s4
1256s3
-0.010
0.0608s4
RESPONSE VARIABLE: FLOOR DUST LEAD
CONCENTRATION ROUND 3
ALL GROUPS
Intercept4 BOSSPI
BOS PI-S
BOS P-S
' 791s4 808s4 675s4
849s4
358
130
834s4
5-114
-------
TABLE 5-30 (cont'd). LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS
USING ESTIMATED BLOOD LEAD PERSISTENCE FACTOR
REGRESSION COEFFICIENT
Predictor
Variable
Soil Pb
Cone.
Round 35
Window Dust
Round 35
ALL GROUPS
BOS SPI
BOS PI-S
BOS P-S
Pb Cone.
MODEL
1
0.201s4
0.0111s2
MODEL
2
0.198s4
0.0103s1
MODEL
10
0.198s3
0.0092s2
MODEL
11
-0.062
0.303
0.473
0.0074S1
MODEL
17
0.189s3
0.00671T
are /*g/dL Pb in blood.
2Units are /ig/dL Pb in blood per 1000 pg/g Pb in dust or soil.
3Units are pg/dL Pb in blood at Round 3 per /*g/dL Pb in blood at Round 1.
4Units are pg/g Pb in dust.
5Units are pg/g Pb in floor dust per pg/g Pb in soil or window dust.
TABLE 5-31. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: MODEL ASSESSMENT STATISTICS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR
MODEL ASSESSMENT STATISTICS
Response
Variable
Log Blood Pb
Rounds
Log Blood Pb
Round 1
Log Dust Pb
Cone. Round 3
Log Dust Pb
Cone. Round 1
All
STATISTIC
RMSE
RMSE
RMSE
RMSE
N*OBJ
MODEL
1
0.2915
0.3392
0.6111
0.7918
17.57
MODEL
2
0.2945
0.3393
0.6110
0.7921
17.47
MODEL
10
0.3060
0.3413
0.6094
0.7898
16.47
MODEL
11
0.3102
0.3411
0.6412
0.7916
14.37
MODEL
17
0.3036
0.3400
0.6094
0.7894
16.34
MODEL
30
0.3120
0.3402
0.6283
0.7908
14.54
*RMSE = Root mean squared error.
5-115
-------
TABLE 5-32. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR
REGRESSION COEFFICIENT
Predictor
Variable
MODEL
1
MODEL MODEL MODEL MODEL
2 10 11 17
MODEL
30
RESPONSE VARIABLE: BLOOD LEAD ROUND 3
ALL GROUPS
. BOS SPI
Intercept1
BOS PI-S
BOS P-S
ALL GROUPS
Soil , BOS SPI
Pb Round 32
BOS PI-S
BOS P-S
ALL GROUPS
Floor Dust BOS SPI
Pb Cone.
RoungS2 BOS PI-S
BOS P-S
Blood Pb Round I3
1.37S1
0.4301T
0.795M
0.589
1.41S1 1.05M 1.27M
0.24
1.80S1
2.10S1
-0.202 -0.419 -0.263
0.686
0.424M
0.4731T
0.736M 1.752S1
1.014M 0.820
2.1011T 2.569s1
2.278s1 2.407s1
0.589 0.589 0.589 0.589
-0.03
2.51S1
2.50S1
-0.512
1.931s1
0.589
RESPONSE VARIABLE: BLOOD LEAD ROUND 1
Intercept1
Soil Pb Round I2
Dust Pb Round I2
12.74s4
-0.200
-0.029
12.75s4 12.70s4 12.51s4 12.66s4
-0.204 -0.143 -0.043 -0.150
-0.030 -0.063 -0.066 -0.050
12.55s4
-0.072
-0.055
RESPONSE VARIABLE: FLOOR DUST LEAD
CONCENTRATION ROUND 1
Intercept4
Soil Pb Round I5
Window Dust Pb Round I5
1701s4
-0.175M
0.0678s4
1721s4 1513s4 1647s4 1521s4
-0.180M -0.130 -0.172M -0.125
0.0678s4 0.0683s4 0.0737s4 0.0667s4
1695s4
-0.170M
0.0701s4
5-116
-------
TABLE 5-32 (cont'd). LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR
REGRESSION COEFFICIENT
Predictor
Variable
MODEL MODEL MODEL
1 2 10
MODEL
11
MODEL
17
MODEL
30
RESPONSE VARIABLE: FLOOR DUST LEAD
CONCENTRATION ROUND 3
Intercept4
Soil Pb
Cone.
Round 35
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
Window Dust Pb
Cone. Round 3s
705s4 707s4 713s4
0.222s4 0.221s4 0.209s4
0.0095s2 0.0094s2 0.0091s2
788s4
109
744
0.192
0.392
0.242
0.0082s1
717s4
0.214s4
0.0083S1
787s4
367
485
0.197
0.294
0.347
0.0076S1
are jtg/dL Pb in blood,
2Units are /*g/dL Pb in blood per 1000 /tg/g Pb in dust or soil.
3Units are /*g/dL Pb in blood at Round 3 per fig/dL Pb in blood at Round 1.
4Units are ^g/g Pb in dust.
5Units are /ig/g Pb in floor dust per jug/g Pb in soil or window dust.
TABLE 5-33. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: MODEL ASSESSMENT STATISTICS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR FOR MALES
MODEL ASSESSMENT
Response Variable
Log
Log
Log
Log
All
Blood Pb
Blood Pb
Dust Pb
Round
Round
3
1
Cone. Round 3
Dust Lead Cone.
Round 1
Statistic
RMSE
RMSE
RMSE
RMSE
N*OBJ
MODEL
1
0.26227
0.37303
0.62102
0.77072
19.09
MODEL
2
0.26907
0.37510
0.61741
0.76918
19.20
MODEL
10
0
0
0
0
.28664
.37003
.61503
.77163
19.08
MODEL
1.1
0.26284
0.35280
0.61901
0.78928
13.28
STATISTICS
MODEL
17
0.27831
0.37329
0.61665
0.76631
19.05
MODEL
30
0.26417
0.35432
0.61574
0.78474
12.89
are reduction of /ig/dL Pb in blood per reduction of 1000 /*g/g Pb in soil.
2Age is age in months at time of Round 1 blood sample.
5-117
-------
TABLE 5-34. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR FOR MALES
REGRESSION COEFFICIENT
Predictor
Variable
MODEL
1
MODEL MODEL MODEL MODEL
2 10 11 17
MODEL
30
RESPONSE VARIABLE: BLOOD LEAD ROUND 3
ALL GROUPS
. BOS SPI
***"*** BOSPI-S
BOS P-S
ALL GROUPS
Soil Pb BOS SPI
Round 32
BOS PI-S
BOS P-S
ALL GROUPS
Floor Dust BOS SPI
Pb Cone.
Rounds2 BOSPI-S
BOS P-S
Blood Pb Round I3
2.20s3
1.007s4
0.350
0.589
1.99s2 1.581T 2.37s2
1.37M
2.63S1
2.01M
0.607 0.055 0.505
0.757
1.193s4
0.858s1
0.509 1.185M
1.008M 0.340
1.476 1.401
1.536M 0.945
0.589 0.589 0.589 0.589
1.741T
3.35s2
2.88S1
-0.050
0.871
0.589
RESPONSE VARIABLE: BLOOD LEAD ROUND 1
Intercept1
Soil Pb Round I2
Dust Pb Round I2
14.46s4
-0.786s2
0.016
14.56s3 14.21s4 12.64s4 14.39s4
-0.850s2 -0.735s1 -0.244 -0.812s2
0.054 0.034 0.115 0.054
12.64s4
-0.283
0.137
RESPONSE VARIABLE: FLOOR DUST LEAD CONCENTRATION
ROUND 1
Intercept4
Soil Pb Round I5
Window Dust Pb Round I5
1416s4
-0.141
0.0750s4
1458s4 1315s4 5411T 1410s4
-0.150 -0.115 0.276s1 -0.125
0.0750s4 0.0764s4 0.0510s4 0.0739s4
582S1
0.270S1
0.0514s4
RESPONSE VARIABLE: FLOOR DUST LEAD
CONCENTRATION ROUND 3
ALL GROUPS
. BOS SPI
Intercept4
* BOS PI-S
BOS P-S
895s4
890s4 881s4 892s4
761s4
729si
-292
787s4
6961T
-323
5-118
-------
TABLE 5-34 (cont'd). LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR FOR MALES
REGRESSION COEFFICIENT
Predictor
Variable
Soil Pb
Cone.
Round 3s
ALL GROUPS
BOS SPI
BOS PI-S
BOS P-S
Window Dust Pb Cone.
Round 35
MODEL
1
0.223s4
-0.0021
MODEL
2
0.235s4
-0.0017
MODEL
10
0.232s4
-0.0012
MODEL
11
0.237
0.018
0.918s1
0.0076s1
MODEL
17
0.247s4
-0.0015
MODEL
30
0.217
0.046
0.930s1
0.0068s1
are /ig/dL Pb in blood.
2Units are /tg/dL Pb in blood per 1000 pglg Pb in dust or soil.
3Units are /tg/dL Pb in blood at Round 3 per /tg/dL Pb in blood at Round 1.
4Units are /tg/g Pb in dust.
5Units are /*g/g Pb in floor dust per /*g/g Pb in soil or window dust.
TABLE 5-35. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: MODEL ASSESSMENT STATISTICS USING
FIXED BLOOD LEAD PERSISTENCE FACTOR FOR FEMALES
MODEL ASSESSMENT STATISTICS
ResponSe Variable
MODEL MODEL MODEL MODEL MODEL MODEL
STATISTIC 1 2 10 11 17 30
Log Blood Pb Round 3 RMSE1
Log Blood Pb Round 1 RMSE
Log Dust Pb Cone. Round 3 RMSE
Log Dust Pb Cone. Round 1 RMSE
All N*OBJ
0.31964 0.33652 0.31968 0.31755 0.35483 0.31914
0.34553 0.33859 0.32730 0.38193 0.32931 0.39155
0.70085 0.69726 0.69855 0.69354 0.69200 0.69940
0.89148 0.87654 0.85724 0.86649 0.85785 0.86949
13.15 12.71 11.77 10.87 12.15 11.15
1RMSE = Root mean squared error.
5-119
-------
TABLE 5-36. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR FOR FEMALES
REGRESSION
Predictor
Variable
MODEL
1
MODEL MODEL
2
10
RESPONSE VARIABLE:
Intercept1
SoilPb
Round 32
Floor Dust
Pb Cone.
Round 32
ALL
ROS
BOS
BOS
ALL
BOS
BOS
BOS
ALL
BOS
BOS
BOS
GROUPS
SPI
PI-S
P-S
GROUPS
SPI
PI-S
P-S
GROUPS
SPI
PI-S
P-S
Blood Pb Round I3
1.15M
0.416M
1.063s1
0.589
1.87S1
-2.598
0.288
0.505M
0.677
•
0.589
1.421T
0.090
0.370
1.2821T
1.668s1
0.589
RESPONSE VARIABLE:
Intercept1
Soil Pb Round I2
Dust Pb Round I2
10.08s4
-0.005
0.392M
10.21s4
0.029
0.329M
RESPONSE VARIABLE:
10.99s4
-0.062
0.197
FLOOR
COEFFICIENT
MODEL
11
MODEL
17
MODEL
30
BLOOD LEAD ROUND 3
0.14
0.080
1.477S1
2.182s3
2.333s2
0.589
1.84M
2.75S1
3.97s3
0.276
0.070
0-.589
-0.21
0.64
0.77
0.087
' 1.867**
0.589
BLOOD LEAD ROUND 1
8.44s4
0.394
0.6291T
DUST LEAD
10.77s4
-0.049
0.229
8.26s4
0.400
0.693S1
CONCENTRATION
ROUND 1
Intercept4
Soil Pb Round 1s
Window Dust
Round I5
Pb
2353s4
-0.2771T
0.0511s2
2100s4
-0.238M
1689s3
-0.180M
0.0547s3 0.0647s4
RESPONSE VARIABLE:
FLOOR
1827s3
-0.203M
0.0566s3
DUST LEAD
1694s3
-0.178
0.0629s4
1875s3
-0.206M
0.0541s3
CONCENTRATION
ROUND 3
Intercept4
ALL
BOS
BOS
BOS
GROUPS
SPI
PI-S '
P-S
1103s4
1111s4
1164s4
345
1557
959
1107s4
299
1534
1004
5-120
-------
TABLE 5-36 (cont'd). LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS USING
FIXED BLOOD LEAD PERSISTENCE FACTOR FOR FEMALES
Predictor
Variable
Soil Pb ALL
Cone. GROUPS
Round 3s BOS SPI
BOS PI-S
BOS P-S
Window Dust Pb
Cone. Round 35
REGRESSION
MODEL MODEL MODEL
1 2 10
0.179s2 0.168s1 0.146S1
0.0010 0.0013 0.0020
COEFFICIENT
MODEL MODEL
11 17
0.162s1
3.63$*
0.064
0.075
0.0073 0.0022
MODEL
30 UNITS
4.003M
0.077
0.065
0.0065
its are jug/dL Pb in blood.
2Units are jtg/dL Pb in blood per 1000 /tg/g Pb in dust or soil.
3Units are /ig/dL Pb in blood at Round 3 per /*g/dL Pb in blood at Round 1.
4Units are fig/g Pb in dust.
5Units are pg/g Pb in floor dust per /zg/g Pb in soil or window dust.
TABLE 5-37. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY; MODEL ASSESSMENT STATISTICS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR FOR AGES 18-41 MONTHS
Response
Variable
Log Blood
Pb Round 3
Log Blood
Pb Round 1
Log Dust Pb
Cone.
RoundS
Log Dust Pb
Cone.
Round 1
All
Statistic
RMSE1
RMSE
RMSE
RMSE
N*OBJ
MODEL
1
0.29243
0.33790
0.62465
0.80410
20.86
MODEL
MODEL
2 •
0.30647
0.33689
0.62430
0.80655
21.07
ASSESSMENT STATISTICS
MODEL
10
0.32785
0.33729
0.62544
0.80448
18.73
MODEL
11
0.40901
0.34131
0.63384
0.80124
16.98
MODEL
17
0.31414
0.33631
0.62604
•
0.80346
18.74
MODEL
30 UNITS
0.34253
0.33827
0.63126
0.80124
17.22
JRMSE = Root mean squared error.
5-121
-------
TABLE 5-38. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS USING
FIXED BLOOD LEAD PERSISTENCE FACTOR FOR AGES 18-41 MONTHS
REGRESSION COEFFICIENT
Predictor
Variable
MODEL
1
MODEL
2
MODEL
10
MODEL
11
RESPONSE VARIABLE: BLOOD
Intercept1
Soil Pb
Round 32
Floor Dust
Pb
Cone.
Round 32
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
Blood Pb Round I3
2.01s2
0.261
0.772
0.589
1.76S1
1.597
0.337
0.183
0.850
0.589
1.441T
-0.574
1.020
2.826s1
2.975S1
0.589
0.51
-1.337s1
1.903M
5.378s2
5.435s2
0.589
RESPONSE VARIABLE: BLOOD
Intercept1
Soil Pb Round I2
Dust Pb Round I2
Intercept4
Soil Pb Round 1s
Window Dust Pb
Round I5
Intercept4
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
12.42s4
0.005
-0.068
RESPONSE
1834s4
-0.152
0.0543s4
RESPONSE
593s4
12.18s4
-0.047
0.013
VARIABLE
1910s4
-0.178M
0.0553s4
VARIABLE
581s4
12.21s4
0.056
-0.064
: FLOOR
1520s4
-0.094
0.0556s4
: FLOOR
628s4
12.90s4
-0.105
-0.131
DUST LEAD
1446s4
-0.027
0.0472s4
DUST LEAD
580s4
684
236
MODEL
17
MODEL
30 UNITS
LEAD ROUND 3
0.47
2.68s2
3.02s2
-0.592
2.0271T
0.589
-0.38
2.92S1
3.13S1
-1.0031T
2.800s1
0.589
LEAD ROUND 1
12.00s4
0.061
-0.018
12.55s4
-0.052
-0.068
CONCENTRATION ROUND 1
1536s4
-0.087
0.0531s4
1522s4
-0.042
0.0462s4
CONCENTRATION ROUND 3
615s4
589s4
708
206
5-122
-------
TABLE 5-38 (cont'd). LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR BOSTON STUDY: REGRESSION COEFFICIENTS USING
FIXED BLOOD LEAD PERSISTENCE FACTOR FOR AGES 18-41 MONTHS
REGRESSION
Predictor
Variable
ALL
Soil Pb GROUPS
Cone.
Rounds5 BOSSPI
BOS PI-S
BOS P-S
Window Dust Pb
Cone. Round 35
MODEL
1
0.250s4
0.0080s2
MODEL
2
0.250s4
0.0082s2
MODEL
10
0.244s4
0.0069s1
COEFFICIENT
MODEL
11
0.626s2
0.144
0.354
0.0082s3
MODEL
17
0.251s4
0.0067s1
MODEL
30 UNITS
0.533s1
0.143
0.406
0.0078s2
are /tg/dL Pb in blood.
2Units are /tg/dL Pb in blood per 1000 /tg/g Pb in dust or soil.
3Units are /tg/dL Pb in blood at Round 3 per /tg/dL Pb in blood at Round 1.
4Units are /tg/g Pb in dust.
5Units are jtg/g Pb in floor dust per /tg/g Pb in soil or window dust.
(5-23)
Model 1 has a single common coefficient for all groups for each pathway coefficient or
intercept in the basic model configuration. Model 2 has the same configuration, but instead
of a common coefficient for Round 3 blood lead regression on Round 3 soil lead, there are
different coefficients in each of the three treatment groups. Otherwise, all parameters are the
same in both models. However, when the parameters of the models are estimated from the
data, the estimates of the parameters may be different in different models, since parameters
cannot be estimated independently of the other parameters. Likewise, Model 10 differs from
Model 1 in that the Round 3 blood lead vs dust lead coefficients may be different hi different
groups. In Model 17, all of the pathway regression coefficients are common among
treatment groups, but the effects of the interventions are characterized by different intercepts
for blood lead hi each treatment group Round 3. Common parameter values hi Tables 5-30,
5-32, 5-34, 5-36, and 5-38 are entered as "all" hi the parameter cell, whereas the "all" cell
is left empty and the treatment group cells are filled hi hi the models hi which separate group
parameters are estimated.
5-123
-------
The two-equation blood lead and dust lead model for Round 1 was also used for
Round 3, and there was an additional component for persistence of a fraction of Round 1
blood lead extending to Round 3. Abatement effects at Round 3 were modelled by
separating a single pathway regression coefficient or intercept (group mean) into three
separate coefficients, one for each of the treatment groups BOS SPI, BOS PI-S, and BOS
P-S. The effect of abatement could then be assessed as a difference among these separate
coefficients. All possible combinations of abatement effects could be modelled by one of
32 possible models. Detailed analyses found that only five or at most six of the models gave
good results, as assessed by small root mean squared errors (RMSE) for log blood lead and
log dust lead hi both Round 1 and Round 3. These are shown hi Table 5-29. Models with a
larger number of free (estimatable) parameters gave smaller overall objective functions across
all four state variables. However, assessment of RMSE for the four variables found small
differences, at most.
Model 1 did well hi all of the analyses. This is the simplest model, assuming that there
were no differences hi any pathway coefficients or intercepts hi any treatment group, and
therefore any differences hi blood lead or dust lead by Round 3 can only be attributed to
differences hi soil lead and window dust lead concentrations. In Model 2, this hypothesis is
extended to include different regression coefficients hi BOS SPI, BOS PI-S, and BOS P-S.
However, the Round 3 blood lead versus soil lead regression coefficients shown hi
Table 5-29 for these groups are 0.517, 0.680, and 0.817 jttg/dL per 1000 /ig/g lead in soil
respectively, compared to 0.719 in Model 1 for all groups combined, none of which are
statistically significant or different from each other. Furthermore, as shown in Table 5-29,
the overall assessment statistic N * objective is 41.86 hi Model 2, only slightly smaller than
42.75 hi Model 1, and the RMSE are only slight smaller for Round 3 log blood lead (0.3005
hi Model 2 versus 0.3009 hi Model 1), for Round 1 log blood lead (0.3651 for Model 2
versus 0.3742 for Model 1), for Round 3 log dust lead (0.6208 hi Model 2 versus 0.6215 hi
Model 1), and for Round 1 log dust lead (0.7909 for Model 2 versus 0.7937 for Model 1).
Models 10 and 11, which allow for different regression coefficients for Round 3 blood
lead versus dust lead regression coefficients, showed the best fit for at least three of the four
state variables hi Table 5-29. RMSE for Round 3 log blood lead was 0.2980 hi Models 10
and 11, compared to 0.3009 for Model 1. RMSE for Round 1 log blood lead was 0.3328
5-124
-------
for Models 10 and 11, compared to 0.3742 for Model 1. MSB for Round 3 log dust lead
was 0.6080 for Model 10, somewhat better than 0.6215 for Model 1, and RMSE for Model
11 was 0.6300 which was somewhat worse than Model 1. RMSE for Round 1 log dust lead
was 0.7851 for both Models 10 and 11, lower than 0.7937 for Model 1. In Table 5-30, the
Round 3 blood lead versus dust lead regression coefficients for Models 10 and 11 were
0.18 /ig/dL per 1000 /*g/g dust lead for BOS SPI, 1.27 jtg/dL per 1000 jttg/g for BOS PI-S,
and 1.34 /*g/dL per 1000 pig/g for BOS P-S, which were not statistically significant and were
not significantly different from each other nor from Model 1. In Model 11, the Round 3
dust lead versus soil lead regression coefficients were allowed to vary by treatment group
also. As shown in Table 5-30, the Round 3 dust lead versus soil lead coefficients were
-0.06 /ig/g dust lead per jwg/g soil lead hi BOS SPI, 0.30 in Group BOS PI-S, and 0.47, none
significant and none significantly different, even though the single dust lead versus soil lead
coefficient of 0.20 in Models 1, 2, and 10 was highly significant (P < 0.0001 hi Models 1
and 2, < 0.002 hi Model 10).
Model 17 was the only model providing a good fit in which there were separate
intercepts for each group. As shown hi Table 5-29, Model 5-17 has a slightly lower RMSE
for Round 3 log blood lead than other models, a slightly higher RMSE for Round 1 log
blood lead than Model 1, RMSE for Round 3 log dust lead that is somewhat higher than
Models 10 and 11 but lower than Models 1 and 2, and slightly higher RMSE for Round 1
log dust lead than Models 10 and 11 but lower than Models 1 and 2. The intercept terms in
Table 5-30 are also not significant, but provide easily calculated components of effect size,
4.41 - 1.95 = 2.46 #g/dL for BOS SPI-S versus BOS P-S, 3.28 - 1.95 = 1.33 j^g/dL for
BOS SPI versus BOS PI-S, 4.41 - 3.28 = 1.13 jtg/dL for BOS PI-S versus BOS P-S, which
is very similar to the repeated measures ANCOVA effect size estimates hi Table 5-5.
The single most sensitive parameter estimate hi these models is Round 3 versus
Round 1 blood lead regression coefficient, which is extremely consistent at 0.59 to
0.61 jug/dL in Round 3 per /Ag/dL in Round 1, hi all 5 models in Table 5-30. However, this
coefficient is statistically significant only in Model 1. Since the persistent effect of pre-
abatement blood lead is the largest single component of post-abatement blood lead for most
of the children hi the sample, we decided to also evaluate models hi which this regression
coefficient was held fixed at the Model 10 and 11 optimal value 0.589.
5-125
-------
Longitudinal SEM with Fixed Estimates of Blood Lead Persistence
Results with the Round 3 versus Round 1 blood lead regression coefficient fixed at
0.589 are shown in Tables 5-31 and 5-32. Table 5-31 shows that fixing the value of this
coefficient tends to flatten the effects of separating the other regression model parameters, so
that Model 1 provides a smaller RMSE for Round 3 log blood lead than other good models,
and smaller or nearly smaller values of RMSE for the other three variables. Model 30 also
provides a good fit to all variables, with statistically significant treatment group effects for
the blood lead intercept and for the dust lead versus soil lead intercept.
Table 5-32 shows that there are statistically significant post-abatement relationships
between blood lead concentration and dust lead concentration hi Models 10, 11, 17, and 30.
Furthermore, the differences among treatment groups in blood lead versus dust lead
regression coefficients hi Model 11 is nearly significant, and is marginally significant hi
Model 10. These differences did not exist for Round 1 blood lead. The effects of abatement
are seen in Model 11 in Table 5-32. In the soil abatement group BOS SPI, the typical
Round 3 dust lead concentration has been reduced to about 800 fig/g, and there is only a
small relationship between dust lead and soil lead (0.19 /ig/g dust lead per /jg/g soil lead),
and in any case the Round 3 soil lead is low hi BOS SPI. In the dust abatement group BOS
PI-S, the Round 3 dust lead intercept is lower than for BOS SPI or BOS P-S, about
100 /tg/g, but the relationship to soil lead is much stronger than hi BOS SPI or BOS P-S,
0.39 jtg/g dust lead per yxg/g soil lead. The difference between BOS SPI and BOS PI-S is
that the soil lead is much lower hi BOS SPI, so that dust lead tends to be lower hi BOS PI-S
than hi BOS SPI. On the other hand, while the Round 3 dust lead intercept and the Round 3
dust lead versus soil lead regression coefficient are similar hi BOS SPI and hi BOS P-S,
Round 3 soil lead is much lower in BOS SPI, as is Round 3 dust lead. Therefore, the soil
lead abatement effect appears to be related simply to reduction of lead concentrations in soil
and dust. The partial effect of the dust lead abatement in BOS PI-S may have a component
that could be attributed to a change (possibly temporary) in the soil lead to dust lead
pathway.
A simple calculation of effect size based on separate group lead intercept terms for
Round 3 blood lead is also informative. In Table 5-32, for Model 17, the BOS SPI versus
BOS P-S effect is 2.10 - 0.24 = 1.86 jtg/dL (nearly significant), the BOS SPI versus
5-126
-------
BOS PI-S effect is 1.56 /ig/dL, and the BOS PI-S versus BOS P-S effect is not significant.
In Model 30, the BOS SPI versus BOS P-S effect is significant, 2.53 jug/dL, and the
BOS PI-S versus BOS P-S effect of 2.54 /tg/dL is nearly significant. There is a significant
relationship between blood lead and dust lead in these two models, and significant (possibly
different) relationships between dust lead and soil lead, so that there is again evidence of the
operation of soil lead abatement by a soil lead to dust lead to blood lead pathway hi the
Boston study.
Effects of Gender in Longitudinal SEM
Sensitivity of the LSEM models was also evaluated by stratifying the sample by gender
and fitting different models for males and females. Results for males are shown hi
Tables 5-33 and 5-34 and results for females in Tables 5-34 and 5-35. As shown in
Table 5-33, Models 11 and 30 provided a good fit to blood lead data for males in Rounds 1
and 3. Table 5-35 shows a somewhat different pattern for females, with Models 1 and 10
providing a somewhat better fit to blood lead than models 11 and 30.
Table 5-34 shows that the Round 3 blood lead versus dust lead regression coefficients
for males are less significant, and the Round 3 blood lead versus soil regression coefficients
much more significant in Models 1 and 2 hi the male subgroup than in the whole sample.
Table 5-36 presents a different finding for females, with blood lead versus soil lead
coefficients marginally significant or not significant in most models, and blood lead versus
dust lead coefficients significant or highly significant hi most models. This suggests that
direct soil lead exposure may be more important for boys and dust lead exposure inside the
home somewhat more important for .girls hi the Boston study.
Effect size estimates for males may be taken from Model 30 hi Table 5-34. The effect
of BOS SPI versus BOS P-S is 1.14 /ig/dL (not significant) and the effects of BOS SPI
versus BOS PI-S is 1.61 /tg/dL (marginally significant). The effect of BOS PI-S versus
BOS P-S is not statistically significant. However, this must be combined with an assessment
of differential abatement effects hi Table 5-34 on the dust lead versus soil lead relationship hi
the male residences, where a very large and statistically significant Round 3 relationship
between dust lead and soil lead exists in the control group, 0.93 /^g/g lead hi dust per
lead hi soil, but not hi the two abatement groups BOS PI-S and BOS SPI.
5-127
-------
Effect size estimates for females may be taken from Model 17 in Table 5-36. The
effect of BOS SPI versus BOS P-S is 2.13 /*g/dL (significant) and the effects of BOS SPI
versus BOS PI-S is 0.91 pg/dL (not significant). The effect of BOS PI-S versus BOS P-S is
not statistically significant. In Table 5-36, the dust lead versus soil lead relationship hi the
female residences is statistically significant at Round 3, 0.162 jug/g dust lead per jwg/g soil
lead, whereas hi male residences in Table 5-34, the Model 17 relationship between dust lead
and soil lead is larger and much more significant, 0.25 jig/g lead hi dust per /zg/g lead hi
soil, but not hi the two abatement groups BOS PI-S and BOS SPI. However, the differential
treatment group relationships hi Models 11 or 30 are strikingly different between males and
females. The BOS SPI Round 3 dust lead versus soil lead coefficient for females is very
large, about 4 pg/g dust lead per /ig/g soil lead, and the coefficients are negligible for
BOS PI-S and BOS P-S, whereas the estimated BOS SPI coefficient for males is small hi
BOS SPI, about 0.2, and large for BOS P-S, about 0.9. These differences may be statistical
artifacts, since homes were abated similarly whatever the gender of the resident children.
An alternative hypothesis, that the dust lead versus soil lead relationship depends on the
gender of the child residing hi the house, seems implausible. Additional studies of gender
effects may be of considerable scientific interest. However, it is clear that soil lead
abatement is associated with reduced childhood blood lead hi the Boston study in both male
and female children, even if there is some possibility that the soil and dust exposure
processes may differ by gender.
Longitudinal SEM by Age Group
There were not enough children for separate analyses of Boston children in age groups
< 18 months or > 41 months. The results for 18-41 month old children shown in
Tables 5-37 and 5-38 were very similar to those for the group as a whole, with Models 1, 2,
and 17 providing the best fit to blood lead data. Effect size estimates for Model 17 were
also similar to repeated measures ANOVA results, 2.55 jttg/dL for BOS SPI versus BOS P-S
and 2.21 pg/dL for BOS SPI versus BOS PI-S (both statistically significant). There appears
to be a significant group difference hi the relationship between dust lead and soil lead at
Round 3, with a significantly stronger relationship hi the BOS SPI group and a similar but
weaker relationship hi the BOS P-S group compared to the BOS PI-S group.
5-128
-------
5.6.2 Cincinnati Study Longitudinal Structural Equation Models
The very simple model that assumes the same relationships among blood lead, dust
lead, and soil lead hi all neighborhoods (called Model 1) provided a reasonably adequate
description of the variability hi the data. A large number of alternative models were
investigated, but only Models 2, 5, and 6 significantly unproved the goodness of fit. The
most important of these was Model 5, which tested the hypothesis that there were differences
hi average residential floor dust lead among the five neighborhoods at Round 4. The
question of whether these differences should be attributed to the soil or dust lead abatements
is discussed below. The models with neighborhood group mean differences in blood lead
that were analogous to these modifications were used to calculate effect sizes (Models 2 and
6). Other modifications that somewhat unproved the fit were that there were neighborhood
differences hi neighborhood mean floor dust lead concentrations at Round 1 that could not be
attributed to a common relationship of floor dust lead to soil lead (called Models Jl to J6
respectively).
The most sensitive parameter hi the model was regression coefficient of Round 4 blood
lead on Round 1 blood lead, which we interpreted earlier as the blood lead persistence
parameter. The optimal value of the parameter for fitting all four state variables (log blood
lead at Rounds 1 and 4, log dust lead at rounds 1 and 4) usually provided a somewhat
inferior fit for Round 4 blood lead, so we modified the fitting procedure first to estimate the
value of this coefficient that optimized prediction of Round 4 blood lead, then optimized all
of the other parameters. The results are shown hi Table 5-39 and 5-40. Table 5-39 shows
the model assessment statistics that were used, including the global objective function for the
iterated generalized method of moments procedure and the RMSE of the four state variables
that were fitted. The regression models for blood lead at Rounds 1 and 4, and the dust lead
regression models are shown hi Table 5-40.
The optimized models all suggest values of the blood lead persistence parameter that
are very similar to that hi the Boston longitudinal structural equation model, hi the range
0.54 to 0.63. The models also show that adjustments for neighborhood differences in blood
lead and floor dust lead at each round clarify the pattern of effects. In Table 5-40, it is clear
that blood lead intercept terms differ substantially across the neighborhoods even after
5-129
-------
TABLE 5-39. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR CINCINNATI STUDY: MODEL ASSESSMENT STATISTICS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR
MODEL ASSESSMENT STATISTICS
Response
Variable
Log Blood
Pb Round 3
Log Blood
Pb Round 1
Log Dust
Pb Cone.
Rounds
Log Dust Pb
Cone.
Round 1
All
Statistic
RMSE1
RMSE
RMSE
RMSE
N*OBJ
MODEL
1
0.50410
0.52740
0.88351
0.72687
32.59
MODEL
5
0.51099
0.52134
0.84617
0.73471
22.22
MODEL
6
0.53409
0.53216
0.83267
0.73004
17.82
MODEL
J5
0.51589
0.76285
0.83566
0.82747
20.16
MODEL MODEL
J6 30 UNITS
0.49415
0.78434
0.82889
0.84609
19.33
JRMSE = Root mean squared error.
adjustment for individual household dust lead concentrations. The Round 4 intercepts are
large and statistically significant for CIN NT(M) and CIN SEI(P) in Models 2, J2, and J6,
and large for CIN NT(M) in Model 6. At least some of the children in CIN NT(M) (and
possibly CIN SEI(P)) may have been exposed to a lead source or medium contaminated with
lead, other than soil and floor dust, to which most children hi the other neighborhoods were
not exposed. The adjustment for changes in dust lead are suggested hi Table 5-40, which
suggests large increases hi average dust lead hi CIN I-SE(D), CIN I-SE(F), and CIN SEI(P)
from Round 1 to Round 4, everything else being equal.
5.6.3 Calculating Effect Sizes from Longitudinal Structural! Equation
Models
This section illustrates how effects sizes can be calculated from the results for certain
longitudinal structural equation models for Boston and Cincinnati. The effect size
comparisons for Boston are shown hi Table 5-41. The Boston results are based on
Table 5-32. While several different models were evaluated, Model 17 offered the smallest
global objective function and among the smallest RMSE of all models fitted by the Iterated
5-130
-------
TABLE 5-40. LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR CINCINNATI STUDY: REGRESSION COEFFICIENTS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR
REGRESSION COEFFICIENT
Predictor
Variable
ALL GROUPS
CIN I-SE(D)
, CIN I-SE(F)
Intercept1
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Floor Dust Pb Round 42
Soil Pb Round 42
Blood Pb Rount 1 (Fixed)3
Intercept1 ALL GROUPS
CIN I-SE(D)
CIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Floor Dust Pb Round I2
Soil Pb Round I2
MODEL MODEL
1 5
RESPONSE
1.73s2 1.161T
3.84s2 4.70s2
-0.05 0.67
0.4824 0.5221
RESPONSE
8.84s4 7.35s4
3.00S1 2.491T
-0.39 3.04S1
MODEL
6
VARIABLE:
-0.14
-1.50M
-1.00s1
3.27S1
-0.09
5.70s4
0.6953
VARIABLE:
6.73s4
4.02S1
2.681T
RESPONSE VARIABLE: FLOOR
MODEL
J5
MODEL
J6 UNITS
BLOOD LEAD ROUND 4
0.46
4.74s3
0.90
0.5456
1.19
0.66
0.49
3.27S1
1.841T
3.34s3
0.5801
BLOOD LEAD ROUND 1
10.25s4
8.98s4
7.02s4
6.82s4
6.87s4
4.52s3
DUST LEAD
10.61s4
9.34s4
7.05s4
6.42s4
7.23s4
3.84s2
CONCENTRATION
ROUND 1
Intercept4 ALL GROUPS
CIN I-SE(D)
CIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Window Dust Pb
Round I5
Soil Pb Round I5
150s4 156s4
0.0292s3 0.0255s3
0.188s2 0.157s2
136s4
0.0324s3
0.166s2
166s4
741T
115S1
217s4
132s3
0.1153s4
150s3
51M
108S1
170s4
128s2
0.1293s4
5-131
-------
TABLE 5-40 (cont'd). LONGITUDINAL STRUCTURAL EQUATION MODELS
FOR CINCINNATI STUDY: REGRESSION COEFFICIENTS
USING FIXED BLOOD LEAD PERSISTENCE FACTOR
REGRESSION COEFFICIENT
Predictor
Variable
MODEL
1
MODEL
5
MODEL MODEL
6 J5
MODEL
J6 UNITS
RESPONSE VARIABLE: FLOOR DUST LEAD CONCENTRATION
ROUND 1
Intercept4
ALL GROUPS
CIN I-SE(D)
CIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Window Dust Pb
Round I5
Soil Pb Round I5
150s4
0.0292s3
07188s2
156s4
,
0.0255s3
0.157s2
136s4
166s4
741T
115S1
217s4
132s3
0.0324s3 0.1153s4
0.166s2
150s3
51M
108S1
170s4
128s2
0.1293s4
RESPONSE VARIABLE: FLOOR DUST LEAD CONCENTRATION
ROUND 4
Intercept4
ALL GROUPS
CIN I-SE(D)
CIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Window Dust Pb Round I5
Soil Pb Round I5
324s4
0.0826s4
-0.121s1
235s4
227s4
79s3
3271T
254s4
0.0813s4
235s4 295s4
284s3 212s4
82s4 86s4
53 89
235s4 295s4
0.0781s4 0.0740s4
293s4
210s3
78S4
71
237s4
0.0832s4
'Units are /tg/dL Pb in blood.
2Units are /xg/dL Pb in blood per 1000 /ig/g Pb in dust or soil.
3Units are /tg/dL Pb in blood at Round 3 per /tg/dL Pb in blood at Round 1.
4Units are /ig/g Pb in dust.
5Units are /zg/g Pb in floor dust per jtg/g Pb in soil or window dust.
Generalized Method of Moments (TTGMM) method in SAS PROC MODEL. In Model 17,
the three treatment groups are assumed equal at Round 1. However, at Round 3, the three
groups are assumed to have different blood lead intercept values due to differences in
treatment. If G: denotes the overall blood lead intercept at Round land Gg3 denotes the
5-132
-------
TABLE 5-41. COMPARISON OF STATISTICAL METHODS
FOR BOSTON STUDY: REDUCTION IN BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 3
Study Group
Abate Versus Control
Statistical Method
Repeated Measures Analysis
of Variance1
Longitudinal Structural
Equation Model1
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
1.87s2
0.33
1.54S1
1.86
0.30
1.56
its are jig/dL Pb in blood.
intercept for group g at Round 3, then the general form of the two blood equations for
Model 17 as fitted are:
Bloody =Gr +
+
for all groups
Bloodi3 = Gg3 + Bloody*A13 + B3*Dustj3 + F3*Soili3 In group g,
where
Bloody, Bloody == blood lead concentration in rounds 1 and 3 respectively,
, Dustj3 = floor dust lead concentration hi rounds 1 and 3,
il^ Soili3 = soil lead concentrations hi rounds 1 and 3, .
E^E3 = regression coefficient of blood lead on floor dust lead at rounds 1 and 3,
Fl9 F3 == regressions coefficients of blood lead on soil lead at rounds 1 and 3,
A13 = regression coefficient of blood lead at round 3 on blood lead at round 1.
The "base" effect size estimates can be calculated as the difference of changes hi intercepts,
-5-133
-------
-[SPI, P-S] - V^l (l-Ai3) - G[SPI,3]) - (Gl(l-A13) - G[P-S,3])
= Gp-s,3] - G[SPi,3]) = 2.10 - 0.24 = 1.86 /ig/dL,
where the adjustment for removable blood lead in Phase 1 is (1 - A13). The other LSEM
entries hi Table 5-41 are similarly calculated from Model 17 Round 3 blood lead intercept
terms hi Table 5-31. Additional adjustments of effect size can be made using dust lead and
soil lead regression coefficients to standardize the comparisons.
The Cincinnati analyses are based on Model J6 hi Table 5-39 and are more complicated
because the neighborhoods are initially quite different with respect to both soil lead and dust
lead concentrations. In general, the ITGMM objective function is greatly reduced by
including separate intercept terms at both Round 1 and Round 4, both for blood lead and dust
lead. Thus the model fitted to the blood lead data are hi the form:
Bloodn = Ggl
Bloodi4 = Gg4 + Bloodu*A14 + B4*Dusti4 hi group g,
so that the effect sizes for the "base" model are in the form
E[CIN NT(G), CIN NT(M)] = (G[CIN NT(G),1]
~ G[CIN NT(G),4]} ~ (G[CIN NT(M),1]
G
[CIN NT(M)
,4]}
= { 7.05 (1 - 0.5801) - 0.49} - (6.42 (1 - 0.5801) - 3.27}
= 2.47 - (-0.57) = 3.04 jtg/dL.
The entries in Table 5-43 are the group intercept changes adjusted for removable blood lead.
The final effect sizes hi Table 5-42 were calculated hi the same way as differences of
changes hi Table 5-43.
The right columns of Table 5-43 shows the effect of adding a dust lead adjustment.
The blood lead effect size adjustments were calculated from the blood lead - dust lead
regression coefficients Bl and B4, and the median or average dust lead coricentrations hi
each neighborhood at Rounds 1 and 4 respectively. The defining equation is:
5-134
-------
TABLE 5-42. COMPARISON OF STATISTICAL METHODS FOR
CINCINNATI STUDY: REDUCTION IN BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 4
STATISTICAL METHOD
STUDY GROUP
ABATE VS CONTROL
CIN NT(G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CINNT(M)
CIN NT(G)
CIN NT(M)
CIN I-SE(D)
CIN I-SE(F)
LONGITUDINAL
STRUCTURAL EQUATION
MODEL1
REPEATED MEASURES
ANALYSIS OF
VARIANCE1
3.58
-2.56
1.02
-2.43
-1.20
BASE
MODEL
3.04
-1.27
1.77
-2.07
-2.06
ADJUSTED FOR
MEDIAN DUST
3.67
-1.45
2.22
-1.71
-1.64
are /tg/dL Pb in blood.
ED
[CIN NT(G),ciN.NT(M)]
NT(G)) - B4*(Dusti4-CIN NT(G))}
NT(M)) - B^Dusttf-CIN NT(M)}.
The entries in Table 5-43 are the estimated changes in blood lead attributable to dust lead
between Round 1 and Round 4. There is no "standard" standardization for dust lead or
other environmental covariates. Blood lead changes and effects sizes were also calculated
using dust lead intercepts from Model 17 hi Table 5-39, and were similar to those shown for
mean or median dust lead. The effects in Table 5-42 show a large Phase 1 blood lead
reduction.in CIN NT(G) compared to CIN NT(M), a moderately large blood lead reduction
hi CIN SEI(P) compared to CIN NT(M), and and moderate to large blood lead increases in
CIN SEI(P) compared to CIN I-SE(D), CIN I-SE(F), or CIN NT(G).
5-135
-------
eg
tt# 3
•2, g Q
•3T.2
i2 9
HH (^ +j
ts 3 3
.3,11 O
vi
a s
CO
"
+ 8
O
T—*
8
8
O
CO
00 VO
ON CO
CS r-5
T-H t~- CO i—I 1/1 ON
— I-H eo r* r- >o
•§
O t-
to • c-
O O CT\ VO O\ t- .•*
OO T—I T—I VO Tj- tS OO
J CO O
en
CS
S
T) Q\ f«q
9 5 d
T"? ^'90009
00
SI S5
co to H H
HH HH i^j )^|
03 O
u u
•3 75
B 6
t>OO
PH £3
5-136
-------
5.7 SUMMARY OF RESULTS OF STATISTICAL ANALYSES
5.7.1 Synthesis of Results by Repeated Measures ANOVA and
Longitudinal Structural Equations Modeling
The different statistical methods used in this study produce quantitatively similar
results, with no essential qualitative differences in conclusions from the studies. The
preceeding Tables 5-41 and 5-42 compare effect size estimates from repeated measures
ANOVA analyses and longitudinal structural equations models for Boston and Cincinnati.
Table 5-41 shows the estimated Phase 1 blood lead reductions of the soil abatement group
(BOS SPI) and dust abatement group BOS PI-S vs the Phase 1 control group BOS P-S. The
effect sizes are almost identical, even though the estimates from Model 17 are based on
differences hi intercepts after adjustments implicit in a four-equation model involving floor
dust lead and blood lead from Rounds 1 and 3.
Table 5-42 shows the estimated Phase 1 reduction hi blood lead of one of the Cincinnati
no-treatment groups (CIN NT(G)) vs the other no-treatment neighborhood (CIN NT(M)), and
the Phase 1 soil abatement neighborhood (CIN SEI(P)) vs these two neighborhoods and the
two Phase 1 neighborhoods that received only interior dust abatement (CIN I-SE(D) and
CIN I-SE(F)). The longitudinal SEM effects were calculated from Model J-6 in Table 5-39.
The effect size estimates from Model J6 were calculated in two different ways. The "base"
model calculated treatment group effects as the difference in the change from Round 1 to
Round 4 hi blood lead model intercept terms between the two neighborhoods. The model
"adjusted for mean dust lead" adds analogous terms using the mean dust lead concentrations
hi the neighborhoods at each round to adjust the overall difference. The worksheet for this
calculation is shown in Table 5-43. The differences between methods are larger than for
Boston, but less than 1.3 /*g/dL in each group. The largest differences reduce the advantage
of CIN NT(G) over CIN SEI(P) from 2.6 /*g/dL hi repeated measures ANOVA to 1.4 jig/dL
hi longitudinal SEM, and increased the advantage of CIN SEI(P) over CIN NT(M) from
1.0 fig/dL to 2.2 jtig/dL. The difference between control neighborhoods CIN NT(G) and
CIN NT(M), the largest statistically significant difference between any two neighborhoods,
remained at about 3.6 to 3.7 /tg/dL by either method.
The longitudinal structural equation Model 17 for Boston, the parameters for which are
shown hi Table 5-33, is sketched in Figure 5-48. The model shows intercepts and regression
5-137
-------
Round 1
Figure 5-48. Pathway diagram showing the results of the longitudinal structural
equation Model 17 for the Boston study, from Table 5-32, using the
standard terminology of Figure 5-44. Width of the arrow is proportional
to the regression coefficient for Model 17 in Table 5-32. Regression
coefficients shown by thin lines are not statistically significant (P > 0.2),
and those shown by shaded thick lines are statistically significant
(P < 0.05). Height of the bars in the boxes is proportional to the
intercept of the regression model, with separate intercepts for each
treatment group shown in the Round 3 blood lead box.
5-138
-------
coefficients that have been scaled so as to exhibit the relationships among lead in soil,
window dust, floor dust, and blood, before and after Phase 1 abatement. In Figure 5-48, the
width of the arrows connecting each of the independent and dependent variables, and the
height of the bars in each box is proportional to parameters in Table 5-32. In the pre-
abatement model (Round 1), soil lead concentration has little relationship to blood lead or
dust lead. This is contrary to experience hi other urban areas and may reflect a selection
effect. Window dust lead is correlated with ulterior floor dust lead, however. The post-
abatement model (Round 3) shows several strong statistical relationships. The most
important predictor of post-abatement blood lead is the pre-abatement blood lead, which
includes prior exposures from elevated soil lead and dust lead concentrations, but post-
abatement dust lead is also a significant predictor of blood lead. Window lead is also a
statistically significant predictor of dust lead, although much smaller in magnitude than soil
lead hi the post-abatement data.
While post-abatement soil lead has a small and non-significant direct relationship to
blood lead, the indirect relationship of soil lead to blood lead through house dust is
statistically significant. The combined effect is small, but significant: 0.214 /ig dust
Pb/g soil and 1.752 pg blood Pb/ 1000 g dust corresponds to a composite effect
(0.214 * 1,752) = 0.375 /*g/dL per 1000 /*g/g hi soil. However, this corresponds only to
differentials hi blood lead associated with differences hi soil lead at Round 3. There is also
an overall effect characterized by the abatement group intercepts, which were not
significantly different at Round 1, but are significantly different at Round 3, as shown by the
height of the bars hi the intercept box. The control group BOS P-S had the highest
post-abatement intercept, 2.10 /ng/dL, and the soil abatement group BOS SPI had the lowest
intercept, 0.24 jig/dL. The difference of 2.10 - 0.24 = 1.86 yug/dL is the intercept
difference effect size. However, hi addition to this, there is additional benefit to the group
BOS SPI because of the lower dust lead and soil lead concentrations in most of the BOS SPI
residences. The median difference is about 2.43 /*g/dL, taking differences hi median levels
of soil lead and dust lead into account.
Figure 5-49 shows Cincinnati LSEM Model 5, the parameters for which are given hi
Table 5-40. The scaling is the same as in the Boston model, Figure 5-48, except for the
floor dust concentrations which are shown 1/10 as high as hi the Boston figure. The floor
5-139
-------
Round 1
Window-C
Window-C
Figure 5-49. Pathway diagram showing the output from the longitudinal structural
equation Model 5 for the Cincinnati study, from Table 5-40. Format is
the same as Figure 5-48. Width of the arrow is proportional to the
regression coefficient for Model 5 in Table 5-39.
5-140
-------
dust concentration intercept in the Cincinnati Round 1 is 156 jig/g, compared to 1520 pig/g in
the Round 1 Boston model 17. The Cincinnati model 5 differs from the Boston model 17 in
that the inclusion of group mean differences for floor dust lead at Round 4 significantly
improved the goodness of fit. Floor dust lead differs significantly among groups. However,
Model 5 suggests that differences hi neighborhood soil lead and household dust lead are
adequate for predicting mean post-abatement blood lead concentrations, so the Model 5
figure shows a large and statistically significant effect of household dust lead, taking into
account group differences hi dust lead. Neighborhood soil lead differences do not add
significant predictiveness at postabatement Round 4, once dust lead is taken into account.
This is strikingly different than the Round 1 preabatement model for Cincinnati, which shows
large and statistically significant relationships from soil lead to dust lead, and large,
distinguishable, and statistically significant effects of neighborhood soil lead and household
dust lead on blood lead. The dashed line connecting soil lead to floor dust lead at Round 4
shows that a relationship exists, but cannot be estimated well because of the confounding of
dust intercepts with neighborhood soil lead. Window lead is a statistically significant
predictor of floor dust lead hi Round 1, as well as a larger and more significant predictor of
dust lead at Round 4.
Figure 5-50 shows a more disaggregated version of Figure 5-49. The parameters are
derived from Model 36 in Table 5-40. Since treatment group intercepts are used for floor
dust lead and blood lead hi Rounds 1 and 4, direct soil lead effects are not estimatable. The
relation of floor dust lead to window lead is large and statistically significant, both hi
Round 1 and Round 4. The relationship of floor dust lead to blood lead is large, statistically
significant, and quantitatively rather similar in Rounds 1 and 4, even though dust lead
concentrations were typically much lower hi Cincinnati than in Boston. The change in blood
lead from Round 1 to Round 4 shows little relationship to abatement group, decreasing
sharply in control group CIN NT (G) and increasing hi control group CIN NT (M),
decreasing in dust abatement groups CIN I-SE (D) and CIN I-SE (F), but decreasing less hi
the soil abatement group SEI (P). The environmental pathways are significant, as one might
have expected from studies carried out hi Cincinnati a decade earlier (Bornschein et al.,
1985). Restricting the study to fully rehabilitated residences probably reduced lead paint
effects, but other external changes in lead exposure (manifested hi part through window dust
5-141
-------
Round 1
Figure 5-50. Pathway diagram showing the output from longitudinal structural equation
Model J6 for the Cincinnati study. Format is the same as Figure 5-48.
Pathways shown by dashed lines cannot be estimated using only
neighborhood-average soil lead concentration. Height of the bars in the
boxes is proportional to the intercept of the regression model, with
separate intercepts for each treatment group shown in the Round 1 and 3
floor dust lead and blood lead boxes.
5-142
-------
lead concentrations) seemed to be more important factors than soil and exterior dust
abatement or Round 4 floor dust lead. Additional analyses to evaluate the role of other
external changes would be useful.
5.7.2 Summary of Results by Study
This integrated assessment of the USLADP includes a reevaluation of the results of the
analyses carried out by the original investigators and of the conclusions reached by the
investigators based on their analyses. While the numerical results of their analyses have been
largely confirmed here, other interpretations of the results are also consistent with these
numerical findings and, in some cases, may be more plausible than the conclusions published
by the investigators. The results of the original investigations have also been extended here
by carry ing out additional analyses, using a consistent set of powerful analytical techniques
not available when the original reports were published.
5.7.2.1 Boston Study
The Boston study shows very clear evidence of an effect of soil lead abatement in
reducing blood lead in children currently residing in lead-contaminated housing. The effect
was detected in the whole group of children that received soil abatement, amounting to about
1.9 j»g/dL or 17 percent on average, but was much larger in the subgroup of children ages
1.5 to 3.5 years, amounting to about 2.5 /*g/dL or 20% of then: mean starting blood lead
concentration. Since these children had high lead burdens to start with, the blood lead
reduction was actually about half of the potentially removable blood lead. This is based on
our estimate that the contribution of pre-abatement blood lead to post-abatement blood lead
was 59% of the pre-abatement blood lead concentration, so that only about 41% of the pre-
abatement blood lead was potentially removable. The contribution of pre-abatement body
burden occurs by resorption of pre-abatement lead stored hi the skeleton back into the blood
after the soil lead abatement had reduced environmental lead exposure.
The Boston study also included a treatment group that received only ulterior dust
abatement, whereas the soil abatement group received both soil and dust abatement. The
reference group or control group in the Boston study received neither soil abatement nor dust
abatement, but all three groups received ulterior paint stabilization, a potentially non-trivial
5-143
-------
intervention. Therefore, even the control group in the Boston study was not entirely a "no
treatment" group, and these children may have benefitted as well from the ulterior paint
stabilization. The dust lead abatement group showed a large transient reduction in blood lead
at in the early postabatement stage of the study, similar to the blood lead reduction achieved
in the soil abatement group, but by the end of Phase 1 of the study, the blood lead reduction
in the dust abatement group was only 1.3 ptg/dL. The blood lead reduction in the soil
abatement group persisted throughout the study, including into Phase 2.
In Phase 2 of the study the roles of the treatment groups were reversed. The original
dust abatement and control groups were given soil and interior dust abatement. The first of
these two groups, which received two interventions (dust abatement hi Phase 1 and soil
abatement in Phase 2), showed a further striking reduction of 3.8 /xg/dL or about 40%
compared to the group that received only Phase 1 soil abatement, and about 2 jug/dL or 20%
reduction compared to the group that received no abatement hi Phase 1 and also received soil
abatement at Phase 2. These effects were all statistically significant (even "highly
significant" by conventional standards of statistical confidence).
The sensitivity of these results was tested by several different methods of statistical
analysis. Outcomes based on repeated measures analysis of variance were not sensitive to
methodology, being quantitatively similar in other methods, such as longitudinal structural
equations models.
To test the response of subsets of children, the data were stratified by age, race,
gender, and initial blood lead. There were some differences hi blood lead response between
boys and girls. In Phase 1, there were larger and more significant responses to soil
abatement among boys than girls, whereas hi Phase 2 there was a somewhat larger effect
among girls hi the treatment group that received both dust abatement hi Phase 1 and soil
abatement in Phase 2. There was also a suggestion hi repeated measures analyses of
covariance of some differences hi either blood lead responses or environmental lead pathways
in residences with boys versus girls, but the sample size was too small to allow much
exploration of these interesting hypotheses. There were also large effects of soil lead
abatement and dust lead abatement identified hi Afro-American children, possibly related to
differential responses to dust lead, but the sample size did not allowed more detailed
exploration of these hypotheses. While there were some large reductions hi blood lead hi
5-144
-------
some younger children v/ho started the study at ages 9 to 17 months, there were not enough
of these younger children hi the Boston study to show statistically significant effects hi this
age group, even though the estimated size of the effects was sometimes similar to that seen
hi the largest age group, 18 to 41 months. Soil abatement effects were somewhat smaller hi
children older than age 42 months, although this group also had a much smaller sample size
than the middle group.
Pre-abatement blood lead concentrations were truncated to the range 7 to 24 j^g/dL in
the Boston study. The upper limit was imposed by a Massachusetts requirement that children
with blood lead levels of 25 /ig/dL or more be referred to authorities for treatment or
intervention; thus, such children could not be retained hi the study lest they be given medical
treatment or environmental interventions not assigned in the study design. The lower
truncation limit was imposed because of the concern of the Boston investigators that changes
hi blood lead hi children with blood lead less than 7 /ig/dL might be too difficult to detect.
After discussion with EPA staff, external reviewers, and staff of the other study teams, the
decision to truncate the range of blood lead values was accepted as appropriate.
Nonetheless, EPA has evaluated possible effects of blood lead truncation hi this report. The
Boston data were reanalyzed here using a number of additional truncation subsets. Each
further truncation of the data reduced the sample size, which generally reduced the statistical
significance of the estimated effects. In most cases, the magnitude of the effects remained
the same or similar folio whig each truncation. For children with initial blood lead levels of
at least 10 /*g/dL, the magnitude of the effect and its statistical confidence, increased. In
general, it does not appear likely that the findings of the Boston study would have been very
different if a truncation different from 7-24 /ig/dL had been used.
There was, however, another indication that the blood lead truncation may have had
some effect, even though it did not alter the conclusions from Boston as a longitudinal
intervention study. Cross-sectional structural equation models were used here to assess the
initial relationship of preabatement blood lead, dust lead, and soil lead hi order to investigate
environmental pathways before abatement. While there was a strong relationship between
floor dust lead, window dust lead, and soil lead (as has been found hi many analogous
studies), the relationship of environmental lead to blood lead was relatively weak hi Round 1
of the Boston Study. However, the relationship between blood lead and environmental lead
5-145
-------
found after abatement, using repeated measures analyses of covariance and longitudinal
structural equation models, was quite strong, even in the control group. Our hypothesis is
that the study design, which selected residences with high soil lead but with no children
whose blood lead exceeded 25 pig/dL, may have weakened the initial relationship between
blood lead and environmental lead. It is possible that a group of children who were lead-
burdened and potentially very responsive to abatement were excluded by this unavoidable
requirement. If this is true, the Boston study might have shown even larger effects if such
children had been included.
Thus, while the findings of the Boston study about the effects of soil and dust
abatement certainly appear to be valid (and may have underestimated the effects of
abatement), these findings should not be used to draw inferences about the entire population
of Boston children, since the study was not designed as a representative cross-sectional
population study. Neither should the Boston study be used to infer that abatement was either
equally effective or not effective at soil lead concentrations less than 1000 /*g/g, since no
such residences were included. While it would be reasonable to infer that remediation of
yards with soil lead less than 1000 /*g/g may have a positive but quantitatively smaller
benefit for children residing there, the Boston study neither proves nor disproves such an
inference.
Finally, the Boston study design is the only example of a randomized experimental
design among the three studies. Neighborhood-level differences that were presumably
controlled by randomly assigning treatments across neighborhoods were not similarly
controlled hi Baltimore or Cincinnati. The advantages of the randomized experiment may
have facilitated detection of effects in the Boston study, even hi the face of limitations
imposed by blood lead truncation.
5.7.2.2 Cincinnati Study
Unlike the Boston study, in which there were indications of changes hi environmental
lead exposure or environmental dust lead pathways hi the soil lead and dust lead abatement
groups, there were no substantial indications of any such effects hi the Cincinnati study.
While the Cincinnati study showed clear differences among neighborhoods, the differences
were not aligned with soil or dust abatement, nor were they attributable completely to
5-146
-------
differences in soil lead or floor dust lead. Window dust lead was, however, an important
contributor to floor dust lead, which was a statistically significant predictor of blood lead in
the best-fitting Cincinnati models. There were also strong relationships between floor dust
lead and other interface media, such as lead at the interior entry of the residence unit. It is
likely that lead from another external source was a contributing factor hi lead exposure for
children in some neighborhoods. Unfortunately, the "control" neighborhood of CIN NT(M)
appears to be one of these, and some children in the Phase 1 soil abatement neighborhood of
CIN SEI(P) may also have been affected. An essential requirement of an intervention study
is that the effects of important factors that could affect the outcome of the study be controlled
by design, by randomization, or by stratification and covariate adjustment. It is not clear
that the Cincinnati study met this requirement.
Among the largest and most significant differences in the study is the difference hi
blood lead response between the two no-treatment groups. This difference suggests that
significant factors other than soil abatement were affecting blood lead concentrations in
Cincinnati. These factors have not yet been identified by analysis or via other information.
Soil abatement hi CIN SEI(P) hi Phase 1 appeared to have a positive effect compared to one
of the groups, CIN NT(M), but a negative effect compared to the other treatment group,
CIN NT(G). These inclusive findings emphasize the difficulty in identifying and maintaining
appropriately matched control groups hi neighborhood-level environmental intervention
studies.
5.7.2.3 Baltimore Study
The Baltimore study showed virtually no effect from soil lead abatement. While blood
lead in some children hi the soil abatement group decreased substantially, there were also
decreases hi some children in the control group hi the other neighborhood. Likewise, some
of the children hi the soil abatement group and hi the control group showed large increases hi
blood lead after the soil abatement occurred. Several factors appear to be associated with
these findings:
1. The Baltimore study was the only study that did not carry out ulterior dust
abatement or ulterior paint stabilization, and many homes had large concentrations
of lead in ulterior dust and paint;
5-147
-------
2. Soil lead concentrations in the yards around many of the residences were relatively
low when measured on a yard-average basis, even though for nearly all homes there
was at least one location in the yard with soil lead greater than 500
3. All of the remediated housing was located in one neighborhood, and most of the
non-remediated housing in another neighborhood, which may not have allowed
adequate control for between-neighborhood differences;
4. No data were available on changes hi environmental lead concentrations after
abatement in the control groups hi either neighborhood.
There is little indication that the soil lead abatement substantially or persistently reduced
childhood lead exposure in Baltimore. In view of the large quantity of lead hi ulterior dust
and paint hi most residences, it is likely that this unremediated reservoir of exposure
continued to affect blood lead hi children after soil abatement was carried out. While soil
abatment and exterior paint stabilization may eventually cause reductions hi the component of
interior dust lead concentration attributable to exterior sources, it did not appear to do so
during the time frame of this study to an extent that was detectable. This is not too suprising
in view of the probable ongoing recontamination of the dust from interior sources such as
paint and from unremediated exterior sources such as resuspended surface soil from nearby
residences. The environmental data collected hi the study were not adequate to identify such
processes, however.
The design of the Baltimore study cannot preclude differences or changes hi
neighborhood-level lead exposure that may also have been important, such as was observed
in Cincinnati. While there was a control group of houses hi Area 1 (BAL P2) that were not
remediated, all but two had no soil samples with lead concentrations above 500 /*g/g and
therefore were not comparable to the remediated houses which all had at least one soil
sample above 500 jtg/g. Additional analyses of pre-abatement data to identify differences
between the neighborhoods may be useful.
5.7.3 Summary of Results
The data presented hi this section lead to the following conclusions:
(1) Soil abatement in each study effectively reduced the concentration of lead
hi the soil hi the areas where soil abatement was performed.
5-148
-------
(2) In the Boston and Cincinnati studies, the effectiveness of soil abatement
was persistent through the end of the study. There were no foliowup
measurements of soil in Baltimore to demonstrate persistency.
(3) Reductions of dust lead due to exterior dust abatement, performed only hi
Cincinnati, v/ere not persistent, indicating a source of lead other than soil
at the neighborhood level.
(4) Hand lead measurements often reflected general trends in blood lead
measurements and may be a reasonable estimate of recent exposure.
Hand lead, as measured in these studies, can be a useful complement to
blood lead measurements.
(5) Paint stabilization as performed on all homes with lead-based paint in
Boston (ulterior) and Baltimore (exterior), was intended to reduce the
potential confounding effects from contamination of soil and dust, but hi
retrospect, paint stabilization itself represents one form of intervention hi
this study.
(6) The Boston study may have also affected blood lead concentrations hi the
soil lead and dust lead abatement groups, either by modifying exposure
(as suggested by changes hi blood lead versus dust lead regression
coefficients) or by changing soil-to-dust pathways (as suggested by
structural equation models). These changes are additional possible effects
of abatement beyond the persistent reduction in soil lead and dust lead
concentrations.
(7) There was little evidence of changes hi dust lead exposure or in dust lead
pathways hi the for the soil abatement neighborhood hi the Cincinnati
study, based on longitudinal ANCOVA or structural equation models.
(8) Changes hi blood lead hi the Cincinnati study were associated with
changes in dust lead, but soil abatement was not effective hi reducing dust
lead compared to changes hi some other groups in this study.
(9) There was a marginal indication that blood lead reduction was greater hi
the Baltimore soil abatement group than hi the small control group hi the
same neighborhood that was not abated since soil lead concentrations were
low, but no indication of benefit compared to a control group in another
neighborhood.
(10) Based on the Baltimore and Cincinnati studies, there appear to be some
relatively large differences hi neighborhood-level changes hi lead
exposure hi these urban areas that may constitute a major source of
variability hi response to soil or dust abatement.
5-149
-------
(11) Assessments of the Boston data suggest that some differences may exist
between boys and girls in their response to soil abatement, which may be
related to age or behavior.
(12) Children in the age group 18-41 months showed the greatest reduction in
blood lead from soil abatement in the Boston study, about 2.5 ug/dl in
Phase 1. The effects were larger in Phase 2. While there seemed to be a
large effect of soil abatement for younger children, it was not statistically
significant in the Boston study due to the small number of children less
than 18 months of age at the tune of abatement.
(13) Blood lead reduction of about 1.9 ug/dl associated with soil and ulterior
dust abatement occurred during Phase 1 of the Boston study, and persisted
into Phase 2 with no further abatement.
(14) Soil abatement during Phase 2 of the Boston study was associated with a
reduction in blood lead 2.0 ug/dl in the group that received only paint
stabilization in Phase 1, and with a reduction of 3.8 ug/dl in the group
that received interior paint stabilization and dust abatement in Phase 1,
compared with the group that received no Phase 2 abatement.
(15) While the initial truncation of blood lead range in the Boston study may
have attenuated the initial relationship of blood lead to soil lead,
truncation had little effect on the final results.
5.7.4 Limitations of the Statistical Methods
The statistical methods used here were reasonable and appropriate, and could be used
by other investigators with access to standard statistical software packages. However, the
methods have certain limitations that should be understood. The repeated measures analyses
assume only that the response variables are correlated with each other, with no implication of
temporal causality. The goodness of fit of the models was significantly improved by use of
covariate analyses.
A problem arises if the response variable must be transformed (e.g., by a logarithmic
transformation for blood lead or for hand lead) in order to reduce skewness and to stabilize
variances across treatment groups. The implied model for the original untransformed
variable is then multiplicative in treatment effects and random variation. This is probably
acceptable for the analysis of variance, but is likely to produce a physically or biologically
meaningless specification for the covariate model when the covariates are indicators of
5-150
-------
distinct and additive sources of lead, such as soil lead and interior lead-based paint. The
logarithmic model does not reproduce the additive nature of the separate exposure pathways.
Extension of repeated measures analyses to covariates such as environmental lead levels
that change with time can be done using a single technique, structural equation modeling.
These methods provide more powerful interpretive tools. The availability of environmental
data to characterize time-varying lead exposures in the Boston and Cincinnati studies suggests
that more powerful statistical methods, such as structural equation models, may be more
appropriate.
There were substantial differences in the design of the three studies that precluded
completely identical analyses of the data. It was technically possible to create a combined
data set, given that all three studies included data on blood lead and hand lead before and
after abatement, carefully coordinated measures of family demographic characteristics, and
both soil and dust lead at the child's residence. Also, some parameters are the same across
studies, such as the persistence parameter for blood lead used in structural equation models.
However, the substantial differences in study design, such as the characterization of the
"control" groups, pre-abatement paint stabilization, age distribution at the time of abatement,
ethnic and racial characteristics of the populations, and pre-abatement soil lead exposure
meant that mathematically similar measures of effect in each study would have very different
interpretations, and would not be clearly generalizable to other study designs, much less to
soil lead abatement in other communities. Therefore, no "combined" analyses of pooled data
from all three studies was attempted here.
5.7.5 Comparison Across the Three Studies
The effectiveness of soil lead abatement in reducing blood lead varied greatly among
the three cities. The variability in abatement effects is probably due to substantial differences
in lead sources and pathways among the neighborhoods in these studies. These differences
for each study are discussed below.
The Baltimore study had two neighborhoods, Lower Park Heights and Walbrook
Junction. The area to which abatement was assigned (Park Heights) had enrolled some
families whose residences did not have soil lead levels that were high enough to justify
abatement. The nonabatement houses in Park Heights were used as an additional control
5-151
-------
group. Unlike the other two studies, the soil abatement in Baltimore was not accompanied
by interior dust abatement. There was essentially no significant effect of soil abatement in
the abated houses, compared to the control group. It is likely that interior paint contributed
to child lead exposure, either directly by ingestion of paint chips, or indirectly by the hand-
to-mouth exposure pathway, as follows:
ulterior paint =* interior dust =* hands =* blood.
Cross-sectional and longitudinal structural equation analyses could be used to explore this
hypothesis. However, because there were no repeated measurements of household dust lead,
it was not possible to assess changes in exposure over time except by use of hand lead data.
Concerning the Baltimore study, we conclude that:
It is likely that soil lead abatement had little effect on the primary factors
contributing to elevated pediatric blood lead levels in these two neighborhoods;
those factors appear to be interior lead-based paint and interior dust lead.
The Boston study was conducted with blood and hand leads measured at one
preabatement round and at about 8 months after abatement. Soil and dust lead measurements
were available for pre- and postabatement at about the same time. These data allowed a very
complete analysis of blood lead responses to changes in dust and soil lead over tune.
Relative to the no treatment group, the results showed clearly that there was a persistent
reduction (1 to 1.9 jig/dL) in average blood lead levels for the soil lead abatement children
and that, on average, the postabatement blood leads were lowest in premises that had the
lowest postabatement soil lead and dust lead loadings. Interior and exterior lead paint were
not significant predictors of blood lead for Boston children. Concerning the Boston study,
we conclude:
When soil and dust lead levels show a persistent decrease as a result of
effective abatement, blood lead levels also show a persistent decline.
Because the Cincinnati study had collected blood lead and environmental samples hi six
Cincinnati neighborhoods, analyses comparable to those reported for the Baltimore and
Boston studies can be made. After some analyses using models similar to those for
Baltimore and Boston, it became evident that the neighborhoods within each of then-
treatment group were not comparable in every way. Although there was a strong dependence
5-152
-------
of blood lead on environmental lead, particularly on hand lead and on current floor or entry
dust lead there was no clear pattern of change or response of interior dust lead levels after
abatement.
We are inclined to accept the conclusion of the Cincinnati investigators that blood and
dust lead levels were affected differently at different times and places by other events not
under their control. However, the dose-dependence exhibited hi the models suggests that
reducing ulterior dust lead levels did reduce blood lead levels, at least for a while. The
problem is that the abatements did not always persistently reduce dust lead levels.
We therefore conclude that:
There were additional sources of environmental lead exposure that had
different effects on the neighborhoods during the course of the Cincinnati study
and were not related to the abatement methods used in the study. It mil be
necessary to use other analysis methods, such as structural equations
modeling, in order to determine the extent to which changes in Cincinnati child
blood lead levels may have occurred in response to changes in lead exposure.
5-153
-------
-------
6. INTEGRATED SUMMARY AND CONCLUSIONS
6.1 PROJECT OVERVIEW
This project focuses on the exposure environment of the individual child, looking at
three indicators of exposure: blood lead, hand lead, and house dust lead. From the
perspective of the child's environment, changes in the soil concentration are expected to
bring about changes hi the house dust concentration, the hand dust loading, and the blood
lead concentration.
In the past 25 years, concern for children with lead poisoning has steadily increased
with mounting evidence for the subtle but serious metabolic and developmental effects of lead
exposure levels previously thought to be safe. Childhood lead poisoning was formerly
considered an acute medical problem usually traced to swallowed chips of peeling lead-based
paint. Scientific evidence has systematically revealed deleterious effects of lead from several
sources at lower levels of exposure. Agencies such as the U.S. Environmental Protection
Agency and the Centers for Disease Control and Prevention (CDC) have repeatedly lowered
the level of concern for children's lead burden that recommends environmental or clinical
intervention from a blood lead level of 30 j«g/dL established in 1978 by CDC to 25 /Ag/dL in
1985, just prior to the start of the project, then to the present level of 10 jig/dL, which was
defined in October 1991 by CDC as a blood lead level that should trigger community-wide
prevention activities if observed in many children.
The purpose of Urban Soil Lead Abatement Demonstration Project (USLADP) was to
determine to what extent intervention hi the form of soil abatement hi residential
neighborhoods would be effective as a means to reduce childhood lead exposure. Each of
the three studies hi the project is a longitudinal study of the impact of intervention on the
lead exposure of children. The studies focused on evaluation of the exposure environment of
the children living mainly in inner city neighborhoods. Measurements of lead hi key external
environmental media (e.g., soil, exterior and ulterior dust, and paint) were obtained prior to
soil abatement, along with more direct indices of personal exposure hi terms of hand wipes
and blood lead levels. Abatement of soil lead generally involved removal of contaminated
soil and replacement with clean soil. Postabatement lead levels hi the above media and
6-1
-------
children's blood lead were remeasured at varying intervals to determine the effect of soil
abatement, alone or hi combination with paint stabilization or dust abatement, on blood lead
concentrations. There are few other longitudinal studies of this type, and none of this scope
or duration. Because the three studies were conducted using mutually agreed upon protocols,
with few exceptions, a common ground exists for understanding an array of information
available from the three individual studies that broadens the base of information beyond the
limits of a single study or location.
Although the three studies were conducted independently, an effort was made to
coordinate the critical scientific aspects of each study in order to provide comparable data at
their completion. This effort included seventeen workshops where the study designs,
sampling procedures, analytical protocols, and QA/QC requirements of each study were
discussed with a goal toward reaching a common agreement. In most cases, a consensus was
reached on the resolution of specific issues, but the individual studies were not bound to
conform to that consensus or to adhere to it throughout the study. This procedure produced
similar studies with some differences in study design and experimental procedures.
The individual results for each of the three cities were originally presented at an EPA-
sponsored symposium in August 1992. These presentations included the data analysis and
conclusions for each of the three individual city studies. Following this open discussion with
the scientific community, the three research teams submitted their respective reports to the
designated EPA regional offices (Boston, Region I; Baltimore, Region III; and Cincinnati,
Region V). These reports and their associated data sets were then provided to EPA's Office
of Research and Development (ORD) and Office of Solid Waste and Emergency Response
(OSWER) for further analysis and preparation of this Integrated Report.
The EPA review of the study designs, chemical analytical procedures and data quality
measures has found no major flaws that would cast doubt on the findings of the individual
reports. The data sets submitted to EPA were systemically scrutinized for errors and
inconsistencies, and were reviewed and revised by the principal investigators for each of the
three cities prior to the completion of the analyses reported here. These corrections were
minor and would not have altered the conclusions of the individual city reports.
This EPA Integrated Report has reached its present form after an extensive review
process. First, the reports of the individual studies were peer reviewed by non-EPA experts,
6-2
-------
revised, and presented to EPA in their final form, along with the data sets that were used as
the basis for the individual reports. These data sets were then reanalyzed by EPA using
rigorous statistical techniques to extract information not easily accessible from any individual
study. Earlier drafts of this report, based on those analyses, have undergone several rounds
of internal and external review. This has included release of the report hi draft form for
public comment and external review at two separate expert workshops. Further statistical
analyses (based hi part on peer review comment recommendations) have since been carried
out, and the report incorporated changes reflecting the new analyses and earlier comments
from the external experts. However, due to time constraints and other factors, it has not
been possible to carry out the entire range of analyses that may have been desirable to more
fully address important and interesting issues related to the interpretation of findings from the
subject studies.
Electronic copies of the underlying three cities data sets will be made available to
members of the scientific community for continued review and analysis along with the
release of the final version of this report. This continuing reanalysis means that new
perspectives on the USLADP data may emerge. Although it is unlikely that major findings
have been overlooked during the above-noted extensive review phases, it is not at all
unreasonable to expect that still further information will be retrieved and reported by further
evaluation to be made possible by this open policy for data release.
6.2 SUMMARY OF FINDINGS
6.2.1 Comparison of EPA Integrated Report Results with Individual
Study Results
This integrated assessment looks at the three individual studies collectively to determine
if a broad overview can be taken of the project results when each study is placed hi its
correct perspective.
6.2.1.1 Boston Study
The key findings of this integrated assessment with regard to the Boston study are as
follows:
6-3
-------
1. The median preabatement concentration of lead in soil was relatively high in
Boston, averaging about 2,400 jig/g with few samples below 1,000
2. Abatement of the soil effectively reduced the median concentration of lead in the
soil to about 150 fig/g (an average decrease of about 2,300
3. Soil was clearly a part of the exposure pathway to the child, contributing
significantly to house dust lead.
4. Other sources of lead, such as interior lead-based paint were minimized by
stabilization.
5. The reductions of lead in both soil and house dust persisted for at least two years.
6. Blood lead levels were reduced by approximately 1.9 ing/dL at 10 mo after soil lead
abatement.
7. Additional reductions hi blood lead of about 2.0 /*g/dL (relative to non-abated) were
observed at 22 mo postabatement for children hi houses where the soil lead was
abated and the ulterior house dust lead was consequently reduced and remained low.
The Boston study used analysis of variance methods based on blood lead differences,
and analysis of covariance methods with the longitudinal aspect included by use of the
preabatement blood lead concentration (Round 1) as a covariate. The results of their "crude"
analysis (Table 15-10 in the Boston study report) are virtually identical to the effect size
estimates we calculated for the group as a whole using repeated measures ANOVA and also
using a longitudinal structural equations model. Table 6-1 provides a comparison of the
results from the Boston individual city report and from this report. The effect size estimates
are somewhat smaller in their "base" model, which the longitudinal analysis of covariance
model adjusted only for pre-abatement blood lead. In view of the differences hi methods and
approaches, the overall conclusions are very similar.
The Boston investigators also studied the sensitivity of the effect size estimates to a
large number of other covariates, including environmental factors, family demographic
factors, behavioral factors, and biological covariates. None of these changed the estimated
effect of bos SPI vs BOS P-S (soil abatement vs control) from their base model, 1.49 /*g/dL,
6-4
-------
TABLE 6-1. COMPARISON OF PHASE 1 EFFECT SIZE ESTIMATES
BETWEEN THE BOSTON STUDY REPORT AND THIS REPORT
Group Study
Abate Versus Control
BOS SPI BOS P-S
BOS SPI BOS PI-S
BOS PI-S BOS P-S
Boston Report1
Crude Base
Model Model
1.92 1.49
1.53 1.28
0.39 0.21
This Report1
RM LSEM
ANOVA Model 17
1.87 1.86
1.54 1.56
0.33 0.30
its are /tg/dL reduction of Pb in blood.
by more than 0.22 /ng/dL. The factors were entered one at a time. The largest decrease was
seen with inclusion of race as a factor (which reduced the effect to 1.27 /*g/dL) and with
inclusion of pre-abatement lead paint (which reduced the estimated effect to 1.34 jig/dL).
Five factors decreased the effect size, which nevertheless remained statistically significant:
water lead concentration, time away from home, time away from study area, playing or
sitting on inside floor, and ferritin level. The other 15 factors tested increased the estimated
effect size, particularly age (to 1.61 jtg/dL) and hand washing before meals (to 1.63 jug/dL),
as well as: gender, socioeconomic status, mouthing variables, chipping paint, yard play,
outdoor eating, hand washing after outdoor activity, pets that go outdoors, imported canned
food, lead-related occupations, lead-related hobbies, cigarette smoking, and owner
occupancy. Many of these factors are important hi identifying individual exposure
components and lead risk factors, and are worthy of additional scientific investigation.
However, none of these factors appear to have interacted so strongly with soil and dust
abatement as to have qualitatively affected the conclusions of the study, except for relatively
small effects related to age, race, and lead paint level. Much of the lead paint effect is
mediated, both statistically and physically, by lead concentrations or loadings hi house dust.
It is likely that the use of household dust as a covariate hi the models of this report
effectively subsumed the lead paint effect, and that the dust abatement that was carried out in
the Boston study along with soil abatement may have affected some fraction of the blood lead
response that might have been otherwise attributed to lead-based paint. Even so, the overall
6-5
-------
treatment group effect in the model that included lead paint was only slightly less significant
(P = 0.05) than the base model (P = 0.02). On the other hand, including chipping paint in
the model increased the effect to 1.53 /tg/dL (P = 0.02 for the group model, P = 0.01 for
the BOS SPI vs BOS P-S effect). Additional studies involving the paint contribution to the
total lead exposure pathways, and assessment of the possible effects and interaction between
paint condition and paint lead loading on lead exposure, are needed to understand the
relatively small modifications of effect size attributable to lead paint.
Age and race effects are larger than the paint effects and were evaluated in this report.
Larger effects were identified for children of ages 18 to 41 months, and for children of
Afro-American ancestry, than for the sample as a whole. The Afro-American children also
seemed to show larger responses to dust abatement than did the sample as a whole.
In summary, the abatement of soil in the Boston study resulted in a measureable,
statistically significant decline hi blood lead concentrations in children, and this decline
continued for at least two years. It appears that the following conditions were present, and
perhaps necessary for this effect: (a) a notably elevated starting soil lead concentration (e.g.,
in excess of 1,000 to 2,000 /ng/g); (b) a marked reduction of more than 1,000 /*g/g in soil
lead consequent to soil abatement accompanied by (c) a parallel marked and persisting
decrease in house dust lead.
These conclusions are consistent with those reported by the Boston research team. This
integrated assessment found no basis for modifying their conclusions, although we choose not
to express these findings as a broadly generalizeable linear relationship between soil and
blood, such as change hi micrograms of lead per deciliter of blood per change hi micrograms
of lead per gram of soil, because we believe that such a linear expression of abatement
effects is highly site specific for the soil-to-blood relationship. We found evidence that the
dust-to-blood relationship is more significant than the soil-to-blood relationship and therefore
the abatement effect also depends on soil-to-dust transfer, which may be very site-specific.
6.2.1.2 Baltimore Study
With regard to the Baltimore analyses conducted for this integrated assessment, the
participants hi the abatement neighborhood that did not receive abatement were treated as a
separate control group, rather than combined with the nonabatement neighborhood (as the
6-6
-------
Baltimore research team did). The reason for this was to establish a control group not
influenced by differences between neighborhoods. This alternative approach used in this
integrated assessment had little impact on the statistical significance of soil abatement effects
as reported by the Baltimore research team.
The key findings of this integrated assessment for Baltimore are:
1. The preabatement concentrations of lead hi soil were notably lower (i.e., averaging
around 500 to 700 |»g/g, with few over 1,000/ig/g) than in Boston.
2. The actual reduction of lead hi soil by abatement was small (a change of about
400 jwg/g), compared to the Boston study (a change of about 2,300 /*g/g).
3. Measurements of blood lead were made for only ten months following abatement;
and no significant decreases in blood lead consequent to soil abatement were
observed compared to non-abatement control group children.
4. Except for exterior lead-based paint, there was no control of other sources of lead,
such as the stabilization of interior lead-based paint (as done in Boston) or
abatement of house dust (as done in Boston and Cincinnati).
5. Follow-up measurements of soil (except immediately postabatement) were not made
to establish the persistency of soil abatement, and its possible effects on house dust.
The Baltimore report used a generalized linear regression model (GLIM). In its
simplest form, the regression model can be expressed as a linear model using log-
transformed variables. The Baltimore blood lead model 1 is a simple ANOVA model,
Log(BCij) =
+ e
with only two treatment groups, Area 1 and Area 2. However, Area 1 includes some non-
abated residences as well the residences that received soil abatement, whereas Area 2
includes only non-abated residences. Therefore, the results hi the Baltimore report cannot be
directly compared with the results reported here, where we have separated the abated and
non-abated residences into two groups and used the non-abated residences hi Area 1 as a
second control group. Model 2 hi the Baltimore report is a simple ANCOVA model,
log(BCy) = G^ + b2j Agey + b3j
log(Handij) + ey.
+ b4j Seasony + b6j My log(Handy) + b7j
6-7
-------
In this notation, Age is a semi-categorical variable, Season is included only for preabatement
rounds 1 and 2 that covered many months, and My is a dummy variable for low or high
mouthing behavior. While temporal comparisons are possible, no temporal correlation model
is assumed, and the Baltimore report notes that the lack of temporal modeling is a deficiency
in the analyses.
The Baltimore analyses were carried out for two distinct subgroups of children. The
first set of analyses used only those children who were present hi all six rounds. The second
set of analyses used all children who were present in each round. Analyses for this report
used children who were present in Rounds 3, 4, and 6. The set of children who were
present m all rounds is included in the EPA set, but does not include other children in the
EPA set such as those children who were recruited at Round 3, especially very young
children. The second set of children hi the Baltimore study is much closer to the EPA child
set in Rounds 4 and 6, but includes in Round 3 some additional children who dropped out
after Round 3. Therefore, the EPA effects size estimates are based on different groups of
children than in the Baltimore report.
Effect sizes were calculated in Table 6-2 as simple differences of treatment group
effects reported in Tables 7-7 and 7-8 of the Baltimore report. The effects were small and
probably not statistically significant, although the lack of correlation structure hi the
Baltimore models makes any estimates of standard errors rather questionable. The
differences in blood lead are negative between the treatment group (BAL SP) and the control
group (BAL PI and BAL P2). There is little reason to believe that major treatment group
differences would have been identified by other analyses of these data.
Other findings in the Baltimore study are of interest. There were some indications of
significant differences associated with hand land, with a modifying effect due to child
mouthing behaviors. There was also a strong effect of socioeconomic status on blood lead
and dust lead, and an age effect with maximum blood leads at ages 1 to 3 years (12 to
36 months), a general finding hi these studies.
Thus, hi Baltimore, where the differences between pre- and postabaternent soil lead
concentrations were much less than hi Boston, and where the soil abatement criteria left
some properties only partially abated, and where no ulterior paint stabilization or dust
6-8
-------
TABLE 6-2. EFFECT SIZE ESTIMATES FROM THE BALTIMORE REPORT
COMPARING BLOOD LEAD REDUCTION IN BAL SP VERSUS CONTROLS
Rounds
Rounds 3 and 4
Rounds 3 and 6
Child Group
All 6 Rounds
Each Round
All 6 Rounds
Each Round
Baltimore
ANOVA
-0.55
-0.07
-0.92
-1.55
Model1'2
ANCOVA
0.12
-0.10
-0.71
-1.17
This Report1-3
BAL SP BAL SP
vsBALPl vsBALP2
0.07 1.774
-0.54 0.67
are ;tg/dL reduction of Pb in blood.
Baltimore controls are BAL PI and BAL P2
3Children present in Rounds 3,4, and 6
4P=0.16; others, P>0.2.
abatement was performed, no detectable effects of soil lead abatement on blood lead levels
were found.
These conclusions are consistent with those reported by the Baltimore research group,
and are not inconsistent with those above for the Boston study. At soil concentrations much
lower than the Boston study, the Baltimore group would have likely been able to see only a
very modest change in blood lead concentrations (perhaps less than 0.2 jitg/dL), assuming
similarity between the study groups hi Boston and Baltimore and the same linear relationship
between change in soil concentration and change hi blood lead. Furthermore, the ulterior
paint stabilization and house dust abatement performed hi Boston perhaps enhanced and
reinforced the impact of soil abatement on childhood blood lead, whereas hi Baltimore, any
possible small impact of soil abatement would have likely been swamped by the large
reservoir of lead hi the interior paint and the large unabated amounts of lead in ulterior house
dust.
6.2.1.3 Cincinnati Study
As for the Cincinnati study, because of differences hi the neighborhoods, we found that
combining neighborhoods into treatment groups often obscures important effects, and chose
to analyze each of the six Cincinnati neighborhoods as separate treatment groups. One
neighborhood, CIN I-SE(B) had an insufficient number of participants and was dropped from
6-9
-------
some analyses. The CIN I-SE(B) group started with nine families, but by Round 5 there was
only one participating family in the study. The two control neighborhoods, CIN NT(G) and
CIN NT(M), were also found to be substantially different, and that the three remaining
treatment groups, CIN SEI(B), CIN I-SE(D), and CIN I-SE(F), were more comparable, both
demographically and in geographic proximity, to CIN NT(M) than to CIN NT(G).
The Cmcinnati study used several different regression (ANCOVA) models, and cross-
sectional structural equation models. The report also included results of a simple correlation
analysis that did not allow for multiple covariate adjustments, and is not further described.
The response variables in the regression models included the difference in blood lead
between Round 1 and Round 4, hand lead differences, and differences in interior floor dust
loading and in exterior dust loading. The final regression model for the change in blood lead
involved only blood lead concentration (which we denote Blood), hand lead loading (which
we denote Hand), age of the child at the Round 4 blood lead measurement (which we denote
BloodR4), and socioeconomic status (denoted SES). In our notation, their model is:
BloodR4 - BloodRl = 8.52 + 0.038 (HandR4 - HandRl) - 0.00079 AgeR4*HandR4
- 0.17 SES-0.43 BloodRl.
This model has one point of similarity to our Cincinnati longitudinal SEM models. By
transposing the BloodRl on the left side of the equation, we have a linear relation that is
expressed algebraically as BloodR4 = 8.52 + ... other terms + 0.57 BloodRl, which is
close to the value of the blood lead persistence parameter A14 obtained for most of the
Cincinnati LSEM models, such as A14 = 0.58 in Model J6 used in the effects size
comparisons. Otherwise, blood lead is not predicted by neighborhood, nor by abatement
group, nor by environmental lead concentrations or loadings, but by another time-variable
and child-specific variable, hand wipe lead loading, which tends to increase with the child's
age. The regression model for hand lead change also excludes treatment group or
environmental variables, except indirectly through Round 1 hand lead.
The report also presents a structural equation model for blood lead and hand lead
differences, and for changes in interior and exterior dust lead. Their equations for blood and
hand lead are, in our notation:
6-10
-------
BloodR4 - BloodRl = 10.28 - 0.18 SES - 0.064 AgeR5 - 0.46 BloodRl
HandR4 - HandRl = 5.78 + 0.002 HandR5 - 0.62 HandRl.
The two dust lead equations are totally unconnected to blood lead or hand lead.
The report also shows cross-sectional structural equation models for Round 1, Round 3,
and Round 4 respectively. The Round 1 SEM model shows large and statistically significant
age effects, and effects of mouthing behavior. Areas and neighborhoods show no significant
differences. The model uses no environmental covariates, but reports a significant regression
of log(BloodRl) on log(HandRl). The simultaneous equation for log(HandRl) depends
strongly on age and not at all on treatment group or neighborhood. Neither equation uses
any of the environmental covariates, but both include a significant fixed effects factor for
"families", which is analogous to the random effects term Hh(g) in our repeated measures
ANOVA and ANCOVA models. However, their findings of no significant neighborhood
differences or environmental factors differs somewhat from some of the findings in our
cross-sectional and longitudinal SEM models. Differences in model format and structure
make direct comparisons very difficult.
The Cincinnati investigators concluded that the Phase 1 changes hi blood lead
concentrations and in hand lead loadings were not significantly different among the three
abatement groups, using either multiple regression models or structural equation models.
They did not compare across different neighborhoods within treatment groups, which was an
additional source of variability in the study. We cannot therefore directly compare our effect
sizes or treatment differences across neighborhoods with their aggregated results. Since their
models are not directly comparable to our models without additional substantive analyses of
the role of hand wipe lead, we cannot directly compare effect sizes using longitudinal SEM.
The Cincinnati report giving a cross-sectional SEM for Round 4 (their Table 4-63)
presents a comprehensive and detailed SEM which is hi substantial qualitative agreement with
the longitudinal SEM we presented for Cincinnati Round 4 blood lead and dust lead. The
use of hand lead hi their model precludes direct comparisons with the longitudinal SEM
shown here hi Table 5-39. The use of log(HandR4) as a covariate that is only partially
adjusted by window and floor dust lead loadings, age, and SES permits the finding of large,
6-11
-------
statistically signficant, but negative relationships between log(BloodR4) and dust lead
loadings on the floor, interior entry, and exterior. Additional analyses of this model would
be useful. The model uses neighborhood or area as an adjustment covariate for hand-to-
blood, dust-to-blood, dust-to-hand, paint-to-dust, and exterior-to-floor pathways, with some
significant differences. While the application of this model does not allow comparison of
effect sizes relative to Round 1, there is a qualitative similarity in our fhidings with those of
the Cincinnati investigators.
On this basis, we concluded that, hi most cases, the effect of soil abatement could not
be clearly determined, and offer the following explanation for this conclusion:
1. Most of the soil parcels hi each neighborhood were not adjacent to the living units,
and this soil was therefore not the primary source of lead hi house dust. Evidence
for this statement includes the observation that street dust lead concentrations are
much higher than soil concentrations, indicating there is a large source of lead
contributing to street dust hi addition to soil lead.
2. The preabatement median soil lead concentrations hi the three treatment groups
were about 300 jtg/g hi CIN SEI(P), 700 jig/g in CIN I-SE(F), and 800 ptg/g in
CIN I-SE(D), and the postabatement soil concentrations were less than 100 pg/g,
so that the reduction of lead hi soil was small, as hi Baltimore.
Evidence for the impact of dust abatement or dust and soil abatement consists of a
statistically significant difference between changes hi blood lead between Rounds 1 and 4,
approximately one year apart. Some Cincinnati neighborhoods showed decreased blood lead
concentrations hi response to dust abatement or dust and soil abatement. The two
neighborhoods that received only ulterior dust abatement hi the first year, CIN I-SE(D) and
CIN I-SE(F), showed a small decrease hi blood lead concentrations, compared to large
increases hi the nearest control group, CIN NT(M). The treatment group that received soil,
exterior dust and ulterior dust abatement, CIN SEI(P), showed a smaller effect than did the
CIN I-SE(D) and CIN I-SE(F) neighborhoods. After consultation with the Cincinnati
research team, we suspect that there was recontamination of street dust hi CIN SEI(P) during
the study, probably caused by demolition of nearby buildings hi the neighborhood.
The consistent theme across the outcomes for all three studies is that soil abatement
must be both effective and persistent hi markedly reducing soil lead concentrations
accompanied by a corresponding reduction hi house dust lead hi order to result hi any
detectable reduction of blood lead. The location of the soil relative to the exposure
6-12
-------
environment of the child is important. In this project, the movement of lead from soil or
street dust into the home seems to be a key factor in determining blood lead concentrations.
Although these USLADP results provide substantial evidence for the link between soil or
street dust and house dust lead, there is insufficient information by which to clearly quantify
this relationship in terms of the lowest level of soil or street dust lead reduction that will
yield a measurable decrease of lead in blood.
6.2.2 Synthesis of Findings Across the Three Studies
While the USLADP was not intended to compare different methods for soil abatement,
the differences in design and methodology among the three studies helped to identify
conditions for which soil abatement may be an effective intervention, and conditions under
which soil abatement is less likely to be effective. Abatement or intervention can be
effective if it can achieve one or both of the following goals:
1. Abatement or intervention produces an effective and persistent reduction in the
concentrations of lead in soil and in household dust.
2. Abatement or intervention changes childhood lead exposure by reducing the intake
of lead-contaminated media, or effectively breaks the transport pathway from the
lead-contaminated source to the child's activity areas.
These are not mutually exclusive goals, but there are important distinctions among them.
The first goal, reducing lead concentrations, can be achieved without changing exposure or
transport. For example, removing bare lead-contaminated soil from a yard and replacing it
with bare soil that is not contaminated will not immediately change the child's exposure to
interior dust lead nor the transport of surface soil from the yard into the house. However,
the child's intake of lead directly from any soil ingestion will immediately be reduced, and
one would expect that over some period of tune, there will be a reduction of the child's
intake of lead from household dust because the soil component of household dust lead has
been eliminated. All three studies achieved the elimination of lead in yard soil. It is
important to note the requirement that the soil not be recontaminated by unremediated
sources such as exterior paint and by transport of lead from unremediated areas. Even hi the
Boston study, a few yards became substantially recontaminated. However, most of the
sampled locations hi the Boston and Cincinnati did not suffer significantly recontaminated
6-13
-------
soil after abatement. The Baltimore sites were not followed up over a similar period of time.
Both Boston and Cincinnati residences received ulterior dust abatement. The Boston
residences showed slight evidence of recontamination, whereas most the residences in the
areas that received ulterior dust abatement (with or without soil abatment) during Phase 1 of
the study showed significant recontamination. The floor dust lead concentrations showed a
significant association with window lead and mat lead, suggesting exterior sources of
recontamination. Long-term changes in dust lead were not followed up in the Baltimore
study. Significant blood lead reduction was detected only in the Boston study, where
persistent reduction of dust lead occurred in most residences that received soil lead and
interior dust abatement. The effect was even greater in Phase 2 in the group PI-S that
received both Phase 1 dust abatement and both soil and dust abatement in Phase 2.
The second goal, reduction of exposure, requires reducing the amount of potentially
lead-contaminated media consumed by the child. Repeated measures analysis of covariance
of the Boston study suggests that this may have occurred, based on some statistically
significant changes in the regression coefficients between blood lead and dust lead after
abatement. Longitudinal structural equation models for Boston also suggest some changes in
soil-to-dust or dust-to-blood pathways. Similar analyses of the Cincinnati data find little
evidence for changes in regression coefficients. The regression coefficients are generally
believed to indicate components of the exposure pathway, either intake of lead-contaminated
media by the child or transmission of lead contamination from one medium to another more
accessible to the child. Soil abatement can reduce exposure by covering soil with sod or
other barriers that reduce the child's access to surface soil particles. The reduction in
exposure is distinct from reducing the lead concentration in the soil to which the child is
exposed. Likewise, frequent and effective washing or vacuuming of household dust can
reduce the amount of dust (dust loading) that is accessible to the child, however much lead is
in the dust. Changes in behavior, such as more frequent hand washing or greater parental
attention, can also reduce contact with dust and soil. Since all of these studies may have
initiated behavioral changes from the moment of recruitment simply by informing parents and
caretakers of potential lead hazards, such changes cannot be detected with this study design.
The second goal can also be achieved by any process that reduces transport of the
contaminant from the source to the areas in which the child may come into contact with it.
6-14
-------
Covering bare soil with sod, concrete, or other barriers will clearly prevent contamination of
house dust and outside play areas, just as encapsulation of paint will prevent paint chips from
contaminating dust, so long as the barrier remains intact. Removing the source of
contamination was shown to be effective in Boston, but hi addition to this, there is also some
possibility that the post-abatement pathway regression coefficient from soil to dust may have
been changed. However, there may also have been a serious attenuation of the apparent
pathway hi the Round 1 data set, possibly attributable to the blood lead truncation of the
study. Additional studies on the effects of soil abatement on environmental lead pathway
kinetics would be useful. In general, any method that attempts to estimate post-intervention
or post-abatement blood lead concentrations (for example, EPA's IEUBK Model or "slope
factor" models) should take into account not only the changes hi environmental lead
concentrations that may occur as the results of abatement or intervention, but also the
changes in the pathways to childhood exposure that may occur following abatement or
intervention.
Finally, one should recognize that any environmental lead abatement or intervention
may be limited hi its ability to reduce blood lead concentrations in currently lead-burdened
children. It appears that, in the first year after abatement, at most 40 to 50 percent of the
child's previous blood lead burden may be removable by soil abatement or any other
combination of abatements and interventions apart from medical treatment by chelation.
Thus, there may be a greater effect of lead abatement hi preventing lead exposure for future
versus current residents, but this possibility cannot be readily assessed, if at all, on the basis
of the existing "Three-Cities Lead Study" data sets evaluated hi the present report.
6.2.3 Application of Findings to Conceptual Framework of Soil and Dust
Lead Exposure Pathways
This integrated assessment attempts to answer the following question: If residential soil
is abated will blood lead concentrations decline? To confirm or reject this soil lead/blood
lead hypothesis, this report builds a framework of logical arguments described below. Each
step of the pathway from soil to blood must be scrutinized closely and related data examined
hi detail. This means mat if dust lead derived from soil is not ingested, either directly or
6-15
-------
after passing through other sources, then blood lead concentrations cannot respond to changes
in soil lead concentrations.
1. There is a substantial amount of lead in soil.
Lead was measured in soil hi the range of less than 50 jwg/g to more than
18,000 jtg/g. If a parcel of 100 m2 had an average of 500 /ig Pb/g soil, then the
upper 2 cm of soil on this parcel (about 4,000,000 g) would contain 2 billion ^g or
two kilograms of lead. Before abatement, there was an estimated 25,000
kilograms of soil lead on the participating properties of this project.
A 2-cm soil core was deemed better than a 15-cm core commonly used in previous
studies. When there is a decreasing gradient between the top and bottom of the
15-cm core, the effect is to dilute the concentration, giving a distorted picture of
what is available at the surface. In this project, some measurements were made of
the soil concentration in the bottom 2-cm of the 15-cm core in order to determine
the depth of excavation. The Boston study reported there was not a large gradient
between the top and bottom of the 15-cm core, as had been expected.
Finally, there is little information on the types of surfaces that a child plays on. ,
If these surfaces are mostly soil, as opposed to asphalt or concrete, then the soil
measurement may be a good estimate of exposure. However, exterior dust is
probably a better estimate of exposure from hard play surfaces (item 5 below).
Exterior dust represents lead from several sources, including soil, and may also be
a better estimate of the lead transferred to household dust.
2. Lead in soil must be connected by environmental pathways to other compartments
of the child's environment, such as exterior dust.
Limited evidence for this statement was shown in the Cincinnati study. In the
Cincinnati study, the relationship between soil and exterior dust was found to be
very weak, giving rise to the next statement.
3. There are sources of lead other than soil that contribute to exterior dust.
Because the changes in lead in soil do not account for all of the changes hi exterior
dust, it is reasonable to conclude from the Cincinnati study that there are other
sources for lead in exterior dust. In Cincinnati, the soil parcels were generally not
on the individual properties of the participating families, as was the case in Boston
and Baltimore. There are no measurements of exterior dust in the Boston or
Baltimore studies.
4. Lead in exterior dust can also move into other components of the child's
environment, such as ulterior dust.
In the Cincinnati study, when exterior dust lead concentrations changed, ulterior
dust lead concentrations also changed. This was especially obvious when the
6-16
-------
exterior dust sample closest to the residence was compared to the interior floor
dust sample taken just inside the entryway door.
A living unit with 130 m2 of floor space (1,400 ft2) and 1,000 /*g Pb/m2
(a relatively high value from tables hi Section 3.3) would have 130,000 pig of lead,
or less than 0.01% of the lead available from soil in paragraph 1 above (see
Figure 6-1). Additional lead would be hi rugs, upholstered furniture, and window
areas.
5. There are sources of lead other than exterior dust that contribute to interior dust.
Taken individually, none of the studies decisively demonstrated this effect. The
most obvious source of lead inside the home is lead-based paint, which was
common in the Boston and Baltimore studies, but less important hi the Cincinnati
study. Because neither Boston nor Baltimore measured exterior dust,
measurements of ulterior dust hi these studies cannot easily be broken down into
contributions from lead-based paint and from exterior dust. However, structural
equation analyses on the Boston study showed a strong influence of both ulterior
and exterior lead-based paint on ulterior dust.
6. Lead hi soil can move directly onto the child's hand.
Conceptually, the transfer of lead from soil to the child's hand is difficult to
measure. A child playing outside usually gets soil on his/her hands, but it is not
certain whether this soil is adequately represented by a composite of 2 cm soil
cores.
7. Lead hi exterior dust can move directly onto the child's hand.
There is no portion of these studies that directly measures this effect. Baltimore
reported that the lead loading on hands increased during the summer months, by
inference due to the increased playtime outside. During the interviews with the
family, questions were asked hi all three studies about the activity patterns of the
children, including the amount of time spent outside. In the Cincinnati study, the
child was observed during the interview period and the handwipes were taken at
the end of the interview.
8. Lead hi ulterior dust can move directly onto the child's hand.
In most cases, when ulterior dust changed, hand dust changed. Because hand dust
lead is only a measure of the amount of lead on the hand, not the concentration nor
the amount of dust, it is difficult to make a quantitative estimate of this pathway.
It is not likely that the amount of dust on the hand is strictly a function of the
amount of dust on the playing surface, as there is probably an equilibrium effect
where some dust falls off after time, depending on such factors as moisture content
of dust and soil, and conditions on the hand surface. There is no aspect of these
studies that could measure this interesting problem.
6-17
-------
2 kg Pb in Soil
(500 ppm)
0.13gPb
in dust
(HOOng/m*
\
1.6 kg Pb in paint
(2 mg/cm2)
0.5 kg Pb in paint
(0.9 mg/cm2)
Figure 6-1. Total amounts of lead in various compartments of a child's environment,
using the assumptions for concentration (soil, top 2 cm) or lead loading
(dust and paint) in parentheses. Although house dust is only a small
fraction of the total lead in the child's environment, it is the most accessible
component. The concentrations and loadings are illustrative, not typical.
9. Lead in interior dust can also move into other components of the child's
environment, such as food.
This pathway was not investigated by any of the three studies. Measurements of
lead in food before and after kitchen preparation would be required. Conceptually,
this lead and other routes such as the direct mouthing activities on toys, furniture,
and window sills is included in the measurement of interior dust when the
assumption is made that a child ingests about 100 mg dust/day by all routes and
through all activity patterns.
10. There are sources of lead other than dust that contribute to the child's lead
exposure.
In this project, lead was measured in drinking water once or twice during each
study. Low ambient levels (ca. 0.1 /*g/m3) of lead in air (typical of U.S.
metropolitan areas in 1990) were assumed, as were national averages of lead in
food. Ethnic food preferences and individual use of cosmetics or other lead
containing products were not investigated.
6-18
-------
6.3 INTEGRATED PROJECT CONCLUSIONS
The main conclusions of this Integrated Report report are two-fold:
When soil is a significant source of lead in the child's environment, under certain
conditions, the abatement of that soil -will result in a reduction in exposure that
cause a reduction in childhood blood lead concentrations.
(1)
(2) Although these conditions for a reduction in blood are not fully understood, it is
likely that jive factors are important in determining the magnitude of any possible
reduction: (1) the past history of exposure of the child to lead, as reflected in the
preabatement blood lead; (2) the initial soil lead concentration the magnitude of
the reduction in soil lead concentrations; (3) the initial interior house dust lead
loading and the magnitude of the reduction in house dust lead loading; (4) the
magnitude of other sources of lead exposure, relative to soil; and (5) the strength
of the exposure pathway between soil and the child relative to other lead
exposure pathways in the child's environment.
The basis for the first conclusion is: in Boston, where the soil lead concentrations were
high and the contribution from lead-based paint was reduced by paint stabilization, there was
a measurable reduction of blood lead concentrations. This reduction continued to increase
for two years following abatement in Boston.
Conversely, in Baltimore and Cincinnati, where soil was not a significant source of lead
relative to other sources, there was no measurable reduction of blood lead except in cases
where those sources were also removed or abated. In Baltimore, these sources may have
been interior lead-based paint that was not stabilized, or house dust that was not abated.
In Cincinnati, the principal source of lead seemed to be neighborhood dust that may have
been contaminated with lead-based paint.
The basis for the second conclusion is: in those cases where all important elements of
the exposure pathway were available for assessment, the structural equation model analyses
showed that preabatement blood lead concentration was a major predictor of postabatement
blood lead, suggesting mat the remobilization of bone lead is a major component of the
measured blood lead.
All other factors being equal, the measurable reduction hi blood lead was observed only
at higher concentrations of soil lead. In the absence of information about other sources of
lead, no clear statement can be made about the possibility of smaller reductions in blood lead
at lower soil lead concentrations.
6-19
-------
In spite of the recent successes in reducing exposure to lead by removing lead from
gasoline and canned food, lead exposure remains a complex issue. This integrated
assessment attempts to assess exposure to lead in soil and house dust. Lead in soil and
lead-based paint are closely linked in the child's environment. If there is exterior lead-based
paint, then soil lead is likely to be elevated with a consequent elevation hi house dust lead.
If there is interior lead-based paint, then efforts to reduce the impact of soil lead on house
dust will be only partially effective. The maximum reduction in lead exposure will not be
achieved unless both paint and soil abatement are implemented.
There is evidence from all three studies that lead moves through the child's
environment. This means that lead hi soil contributes to lead hi street or playground dust,
lead in exterior paint contributes to lead hi soil, and lead hi street dust contributes to lead hi
house dust. A more detailed analysis of the data may show the relative contribution from
two or more sources, but the present analyses imply that this transfer takes place.
The analysis of the data from the three studies showed evidence that blood lead
responds to changes hi house dust lead. There is also evidence for the continued impact of
other, independent sources following abatement of one source. This means that abatement of
soil or exterior paint does not necessarily reduce the contribution of lead from other sources
such as ulterior lead-based paint.
The conclusions of this report suggest that soil abatement can have a measurable effect
on reducing exposure to lead if there is a substantial amount of lead in soil and if this soil
lead is the primary source of lead hi house dust. In such cases, both soil abatement and
ulterior dust removal should be performed to be fully effective. In addition, if soil
abatement is carried out, then paint abatement should also be considered, where appropriate,
to lessen the probability of recontamination of soil and/or house dust. Likewise, soil
abatement should be considered hi conjunction with paint abatement when it is likely that soil
will otherwise continue to contaminate house dust after a paint abatement is completed.
From one perspective, decisions about soil abatement need to be made on an individual
home basis. This report shows that, on an individual house basis, soil abatement may
potentially reduce the movement of lead into the home and its incorporation into house dust.
The magnitude of this reduction will depend on the concentration of lead hi the soil, the
amount of soil-derived dust that moves into the home, the frequency and methods of cleaning
6-20
-------
in the home and the cleanability of the home. The number and ages of children and the
presence of indoor/outdoor pets are factors known to increase rate of dust movement,
whereas frequent cleaning with an effective vacuum cleaner, use of entry dust mats, and
removing shoes at the door serve to reduce the impact of soil lead on house dust.
From another perspective, soil abatement at the neighborhood level poses problems not
pertinent to individual homes. Playground, vacant lot, and other plots of soil may pose an
immediate problem if they are accessible to children and there is a direct pathway for dust
generated by this soil to enter the home. Likewise, sources of lead other than soil may
contribute more to exterior dust than soil itself. The evidence in this report suggests that the
key to reducing lead exposure at the neighborhood level is to abate significant sources of lead
contributing to exterior dust, in addition to the soil and paint abatement that would be
performed on an individual property.
6-21
-------
-------
7. REFERENCES
Angle, C. R.; Marcus, A.; Cheng, I. -H.; Mclntire, M.S. (1984) Omaha childhood blood lead and environmental
lead: a linear total exposure model. Environ. Res. 35: 160-170.
Angle, C. R.; Mclntire, M. S. (1977) Children, the barometer of environmental lead. Adv. Pediatr. 27:3-31.
Annest, J. L.; Pirkle, J. L.; Makuc, D.; Neese, J. W.; Bayse, D. D.; Kovar, M.G. (1983) Chronological trend
hi blood lead levels between 1976 and 1980. N. Engl. J. Med. 308:1373-1377.
Aschengrau, A.; Beiser, A.; Bellinger, D.; Copenhafer, D.; Weitzman, M. (1994) The Impact of Soil Lead
Abatement on Urban Children's Blood Lead Levels: Phase II Results from the Boston Lead-in-Soil
Demonstration Project. Environ. Res. 67:125-148.
Barnett, V.; Lewis, T. (1984) Outliers hi Statistical Data. John Wiley and Sons, NY.
Barry, P. S. I. (1981) Concentrations of lead in the tissues of children. Br. J. Ind. Med. 38:61-71.
Bornschein, R. L.; Clark, C. S.; Pan, U. W.; Succop, P. A. et al. (1990) Midvale Community Lead Study.
Department of Environ. Health, University Cincinnati Medical Center. July 1990.
Bornschein, R. L.; Clark, C. S.; Grote, J.; Peace, B.; Roda, S.; Succop, P.A. (1988) Soil lead—Blood lead
relationship hi a former lead mining town. In: Environmental Geochemistry and Health, Monograph
Series 4, Lead hi Soil: Issues and Guidelines. (Eds) B. E. Davies and B. G. Wixson. Science Review
Limited, Northwood, England, pp. 149-160.
Bornschein, R. L.; Succop, P.; Dietrich, R. N.; Clark, C. S.; Que Hee, S.; Hammond, P. B. (1985) The
influence of social and environmental factors on dust lead, hand lead, and blood lead levels hi young
children. Environ. Res. 38: 108-118.
Buncher, C. R; Succop, P. A.; Dietrich, K. N. (1991) Structural equation modeling in environmental risk
assessment. Environ. Health Persp. 90: 209-213.
Clark, S.; Bornschein, R.; Succop, P.; Peace, B.; Ryan, J.; Kochanowski, A. (1988) The Cincinnati Soil-lead
abatement Demonstration Project. In: Environmental Geochemistry and Health, Monograph Series 4,
Lead in Soil: Issues and Guidelines. (Eds) B. E. Davies and B. G. Wixson. Science Review Limited,
Northwood, England, pp. 287-300.
David, O. J.; Wintrob, H. L.; Arcoleo, C. G. (1982) Blood lead stability. Arch. Environ. Health 37: 147-150.
Fuller, W. A. (1987) Measurement error models. New York: John Wiley and Sons.
Gallacher, J. E. J.; Elwood, P. C.; Phillips, K. M.; Davies, B. E.; Jones, D. T. (1984) Relation between pica
and blood lead hi areas of differing lead exposure. Arch. Dis. Child. 59:40-44.
Gallacher, J. E. J.; Harris, L.; Elwood, P. C.; (1983) Lead toxicity from water. Nature (London) 305:280.
Grant, L. D.; Elias, R. W.; Goyer, R.; Nicholson, W.; Olem, H. (1990) Indirect health effects associated with
acidic deposition. National Acid Precipitation Assessment Program. SOS/T Report 23. U.S.
Government Printing Office, Washington, DC.
7-1
-------
Gulson, B. L.; Mahaffey, K. R.; Mizon, K. J.; Korsch, M. J.; Cameron, M. A.; Vimpani, G. (1995)
Contribution of tissue lead to blood lead in adult female subjects based on stable lead isotope methods.
J. Lab. Clin. Med. 125(6) 703-712.
Klepper, S.; Kamlet, M. S.; Frank, R. G. (1993) Regressor diagnostics for the errors-in-variables model -
an application to the health effects of pollution. J. Environ. Econ. Management 24:190-211.
Learner, E. E. (1978) Specification searches: Ad hoc inference with non-experimental data. New York: John
Wiley and Sons.
Marcus, A. H. (1991) Relationship between soil lead, dust lead, and blood lead over time: A reanalysis of the
Boston lead data. Report from Battelle Columbus Division, Arlington Office, to USEPA Office of Toxic
Substances. Contract No. 68-02-4246.
Marcus, A. H. (1992) Use of site-specific data in models for lead risk assessment and risk management.
In: An Update of Exposure and Effects of Lead, B. Beck (Ed), Fund. Appl. Toxicol. 18: 10-16.
Marcus, A. H.; Elias, R. W. (1994) Estimating the contribution of lead-based paint to soil lead, dust lead, and
childhood blood lead, Lead in Paint, Soil, and Dust: Health Risks, Exposure Studies, Control Measures,
Measurement Methods, and Quality Assurance, ASTM STP 1226, Michael E. Beard and S.D. Allen Iske,
Eds., American Society for Testing and Materials, Philadelphia, 1994.
Menton, R. G.; Burgoon, D. A.; Marcus, A. H. (1994) Pathways of lead contamination for the Brigham and
Women's Hospital Longitudinal Lead Study, Lead in Paint, Soil and Dust: Health Risks, Exposure
Studies, Control Measures, Measurement Methods, and Quality Assurance, ASTM STP 1226, Michael E.
Beard and S.D. Allen Iske, Eds., American Society for Testing and Materials, Philadelphia, 1994.
Mushak, P. (1993) New directions in the toxicokinetics of human lead exposure. Neurotoxicology 14(2-3)29-42.
Rabinowitz, M. B. (1987) Stable isotope mass spectrometry in childhood lead poisoning. Biological Trace
Element Research. 12: 223-229.
Roberts, J. W.; Camaan, D. E.; Spittler, T. M. (1991) Reducing lead exposure from remodeling and soil track-
in in older homes. Air and Waste Management Association Paper 91-134.2, 84th Annual Meeting and
Exhibition, Vancouver, British Columbia, June 16-21, 1991.
Roels, H. A.; Buchet, J. -P.; Lauwerys, R. R.; Braux, P.; Claeys-Thoreau, F. Lafontaine, A.; Verduyn, G.
(1980) Exposure to lead by oral and the pulmonary routes of children living in the vicinity of a primary
lead smelter. Environ. Res. 22:81-94.
Roels, H. A.; Buchet, J. -P.; Lauwerys, R. R.; Hubermont, G.; Braux, P.; Claeys-Thoreau, F. Lafontaine, A.;
Van Overschelde, J. (1976) Impact of air pollution by lead on the heme biosynthetic pathway in school-
age children. Arch. Environ. Health 31:310-316.
Rothman, K. J. (1990) No adjustments are needed for multiple comparisons. Epidemiology 1:43-46.
SAS Institute, Inc. (1992) SAS* Technical Report P-229, SAS/STAT* Software: Changes and enhancements,
release 6.07; Gary, NC: SAS Institute, Inc.
SAS Institute, Inc. (1993) SAS/ETS® User's Guide, Version 6, Second Edition, Gary, NC: SAS Institute, Inc.
Stanek, E. J.; Calabreese, E. J. (1995) Daily estimates of soil ingestion in children. Environ. Health Perspect.
103:276-285.
7-2
-------
Succop, P. A.; O'Flaherty, E.; Bornschein, R. L.; et al. (1987) A kinetic model for estimating changes in the
concentration of lead in the blood of young children. In: International Conference: Heavy Metals hi the
Environment, (Eds) Lindberg S. E., HutchinsonT. C. New Orleans, September 1987. (EP Consultants
Ltd., Edinburgh, pp. 289-291).
U.S. Environmental Protection Agency (1986) Air quality criteria for lead. Research Triangle Park, NC: Office
of Health and Environmental Assessment, Environmental Criteria and Assessment Office; EPA report no
EPA-600/8-83/028aF-dF. 4v. Available from: NTIS, Springfield, VA; PB87-142378.
U.S. Environmental Protection Agency (1989) Exposure Factors Handbook. U.S. EPA Office of Health and
Environmental Assessment, Washington, DC. EPA/600/8-89/043.
U.S. Environmental Protection Agency. (1991b) National Air Quality and Emissions Trends Report 1989.
Report prepared by U.S. EPA OAQPS, RTF, NC. EPA report no. EPA 450/4-91-003.
U.S. Environmental Protection Agency. (1994) Guidance Manual for the Integrated Exposure Uptake Biokinetic
Model for Lead hi Children. EPA report no. EPA/540/R-93/081. Available from the National Technical
Information Service as PB93-963510.
U.S. Environmental Protection Agency. (1995) Season rhythyms of blood-lead levels: Boston, 1979-1983. EPA
report no. EPA 747-R-94-003. Available from the National Lead Information Center, telephone number
1-800-424-5323.
Yankel, A.J.; von Lindern, I.H.; Walter, S.D. (1977) The Silver Valley lead study: the relationship of childhood
lead poisoning and environmental exposure. J. Air Poll. Control Assoc. 27: 763-767.
7-3
-------
-------
APPENDIX A:
GROUP MEAN PARAMETERS FOR EACH STUDY BY
SAMPLE TYPE, TREATMENT GROUP, AND ROUND
The data in Table A-l were derived using the PROC UNIVARIATE feature of SAS 6.10
(SAS, 1994). The treatment groups are as described hi Chapter 5, using data identical to
that plotted hi Figures 5-8 through 5-32. Data for blood lead concentration and hand lead
are calculated with one value for each child; for floor and window dust, one arithmetic mean
value for each living unit; and for soil, one arithmetic mean value for each property or soil
parcel. The group assignments and numbers of individuals are different from the individual
study reports and different also from the summaries of these reports hi Chapter 4. In
particular, the data are different from Tables 4-2 through 4-4.
A-l
-------
3
00
A-2
-------
0
1 i
Is
U'
PH
3
CO
a
Srt <"•?
IS
r— 1
a
s
VO
-
1
Treatment
Group
I
8
o
00
00
00
•o
CO
s.
OS
CO
«0
s
VO
CO
R
-
IH
E
CQ
VO
oo
oo
CO
OS
VO
VO
V>
00
i-H
Os
s
3
o
l-H
t>
oo
«
-
s
PH
CO
o
CQ
s
i-H
1-H
en
i-H
VO
VO
oo
s
VO
o
CM
t-
m
s
Tl
CO
CM
8
s
«
5
p-
r-
p-
co
^
p-
00
Os
P-
co
CO
CO
CM
ro
VO
•n
o
in
O
o
CO
1— <
o
vH
CM
vd
l-H
s
*
OS
VO
CO
VO
ss
oo
VO
P?
oo
VO
OS
CO
•n
T-H
s?
co
CO
-
CO
$
*
s
s
0
&
X
CO
oo
p~
T-H
oo
CM
OS
CO
CM
n
oo
CM
OO
o
OS
CO
CO
00
l-H
CO
o
l-H
o\
P-;
OS
CM
CO
SI
Tl
trt
o
CO
VO
oo
00
oo
VO
oo
VO
CO
CO
CO
CO
l-H
Os
CM
-
00
Os
CO
00
l-H
8
-
CM
oo
»0
*
CO
1
*
1-H
CO
o
CQ
18
I
1
Os
co
s
*
l-H
CO
CO
CO
p-
00
VD
CM
VO
2
a
00
CO
CM
!
CO
Os
00
CO
CM
S3
CO
CO
%
&
00
1-H
o
VO
1-H
l-H
CO
co
r-
£
OS
OS
p~
l-H
OS
00
in
8
s
CO
o
VO
1-H
p-
oo
CM
oo
CM
'
8
r-H
t-H
r— (
O
en
O
t^»
«
5
S
VO
CO
o
CO
Os
OS
CO
l-H
I
I
o
CO
9
T-H
OS
OO
»
VO
VO
•n
CO
oo
CO
l-H
Os
CO
CM
l-H
Os
co
00
$
p-
oo
!*
00
8
VO
o
Os
s
CM
CO
co
XI
OS
CO
CM
P-
s
g
VO'
l-H
CM
00
CM
l-H
P-
i-H
i-H
CO
l-H
S3
-
9
CO
»
1-H
T-H
O
VO
oo
*
t~~
CO
CO
VO
CM
CO
CO
l-H
CO
CL,
CO
o
CQ
CM
00
«
en
r-H
O
p-
vo
OS
ss
CO
VO
0
T-H
g
CO
CM
P-
oo
s
OS
o
oo
OS
Os
8
oo
a
s
CO
%
VO
CM
CO
CO
s
-------
o
•c *
1
S
jl
a>
g
vo
55
1
tf
a
«J *-*
g O
H
X,
S?
K
"I
eo
1
1
VO
•n
S
r-4
T— t
en
en
n
t-
i
T— t
3
«S
en
^_,
CO
&<
CO
O
m
1
en
S
T-H
g
15
oo
c-
1
ON
eS
en
e^
1
(N
r— 4
OO
o
i—4
o
vo
1—4
T-H
o
s
in
ON
i-H
ON
f-
en
en
ON
oo
C-4
T—4
VO
S;
g
m
oo
T-4
ON
VO
0\
T— C
^
A-4
-------
a
CM
a
CM
VD
CM
vS
a
s
ON
VO
ON
CO
o>
o\
CM
CM
CM
8
cs
a
&
CM
VO
10
VO
CM
00
o\
00
CM
CM
CM
ON
CM
CM
00
8
a
ON
vo
CM
8
8
VO
ON
VO
t '
a
a
s
in
CM
ON
ON
CM
ON
ON
ON
CM
CM
CO
CM
CM
CM
CO
CM
up
00
CQ
ple
•a
A-5
-------
g
0
•a
§ S
SP
g
^
oo
en
i^
!£ CM
1-
T— «
1
VO
j-5
o
(§
§^
1 o
S
I
a>
O«
ra
CO
ON
en
oo
S
VO
8
en
g
o
oo
n-
•n
T_<
o
o
«
o
§
O
JO
A<
"S
O
PQ
a
3
O
m
o
VO
en
S
o
T-l
g
VO
VO
*
CM
2"
•a
W3
3>
R
VO
8
8
o
g
I
VO
in
en
oo
oc
o
^
s
0
g
•n
•*
in
^
•
c?
^
8
if)
8
S
0
ON
OO
,-«
|_<
S
S
oc
CO
CN
0
00
0
VO
m
oo
CM
ON
VO
13*
O
r-H
T-4
0
ON
OO
ON
CO
^
s
0
oo
p
•n
-------
g
t/3
a
So;
o
+3
H
u
PH
a
]§
t— 1
_,
d
PH
VO
T— (
1
a
5 OH
» §
O
g
u
f
<>3
1
00
g
§
oo
o
0
T-H
-
0]
&
f>
PH
s
CO
1
b
n
CO
CM
o
s
ON
CM
^
a
«
/bfl
~5h
3.
CO
CO
CO
s
*
-
ON
CM
VO
OO
n
i
§
SO
8
OO
oo
p-
1
I
T-H
oo
o
VO
s
VO
VD
T— (
er
rn
S5
O
VO
S
1
!
«
N
CO
vo
00
CM
a
g
n
i
§
VO
P-
ON
CO
T— t
OO
•o
T-H
£
CO
s
oo
»
n
g
i
ri
^H
VO
T-H
oo
in
§
n
vo
0
ON
T-H
oo
VO
o
00
o
CO
%
CM
CO
Si
,~,
p-
VO
n
CO
s
s
00
0
00
T— I
T-H
FT
NB. S
H
Z
O
8
CM
T—H
XI
2
r-t
0
0
CM
•o
h
§
X}
s
VO
T-H
CO
0
CO
o
JN
x>
*.
r_t
N
ON
-
§
*
CO
VO
CM
ON
ON
CO
S
8
m
3N
ON
ON
CM
i
8
•n
CM
CO
ON
VO
S
CO
•o
3N
T-H
oo
CO
s
T-H
CM
r-
ON
1
3N
CO
0
CM
•n
^
^H
^
H
Z
U
CO
D
1
VO
i
ON
CO
0
r***
.n
CM
•n
VO
§
8
00
CO
CO
ON
ft
ON
CO
n
JN
ON
n
!M
S
50
(7,
-
CO
VO
VO
3N
b
CO
O
S
oo
«
D
i
00
CO
VO
CO
CM
s
p.
VO
1
oo
8
C3
oo
VO
T-H
CM
CO
CO
VO
oo
p-
•n
n
O
VO
VO
VO
CO
PJ
8;
T-H
^
r-M
&7
****'
S
CO
Z
CJ
u
U
P^
«
Q
J-H
«
E
i
s
VO
n
S
CO
ON
CO
S
co
CM
^.^
W
W)
5?
Si
VO
n
N
CO
T— i
T-H
co
CO
VO
CO
ON
x>
x>
VO
a
m
-
a.
A-7
-------
m
wg
u
•s
II
•c ^
PL,
OO
&
Ig
ss
t— <
a
1
VO
r-H
2
•g
« 0,
i |
S o
£-
ft}
Cl^
^?
f-1
42
a<
m
00
0
o
0
0
o
o
*
>n
o
a
0
0
JQ
PU
9
Q
t*
o
o
E
•3
ca
.9
O
.
,
'
'
,
'
vo
*&ft
•B^'
^0
*s
t
1
t
1
•
,
'
r-
vo
o\
OO
0
o
CO
OO
T— 1
o
T-H
T-H
e
H-t
U
o
>n
VO
vo
vo
0\
8
CO
cs
ON
«
VO
n-
OO
CO
VO
CO
VO
o
OO
§
o
CO
OO
CO
1
g
•0
•0
1
t-H
en
-
0
o
0
0
o
0
~
«
1
1
1
1
1
1
«
1
1
1
I
1
1
'
r-
Y— 1
CO
T-H
VO
CO
CO
T)
^.
CO
CO
CO
VO
CO
CO
T— 4
§
«
l-i<
z
u
vo
VO
S3
CM
•o
c-
co
CO
3
cs
CN
00
i
s
rj-
rt-
o
a
1—4
CO
CO
o
OO
VO
00
£
%
OO
CO
CM
s
-
0
o
0
o
0
0
T— C
«n
1
•
1
1
1
,
'
VO
1
,
1
1
1
1
1
-
£
OO
CO
CO
CO
CO
OO
VO
CO
s
CO
-
g
w
oo
K-H
g
u
ON
00
T-H
r-H
T-H
CO
CO
CO
3
s
«
VO
cs
>n
r-
o
rf
VO
SI
CO
cs
cs
cs
CO
s
o\
T— C
g
SI
CO
CO
f-
CN
cs
-
0
o
0
0
0
0
CS
«
1
1
1
1
1
1
1
VO
1
,
1
1
1
1
1
l>
o\
VO
0
VO
T-H
OO
CN
«
VO
s
CO
o\
Ti
CN
OO
§
c^
T-<
9
CO
CO
CO
CO
1
s
a
0
rH
t-H
-
-
o
o
o
o
o
o
CO
-
I
1
1
1
1
I
vo
i
O
®
I.
A-8
-------
8
H
a
s
VO
VO
I
00
r-
co
s
i
CO
CO
oo
oo
Wi
CO
CO
CO
cs
CO
3
CO
VO
VO
VO
CO
ON
VO
ON
co
I
ON
CO
CO
VO
ON
OS
CO
t-
00
co
CO
co
3
co
VO
o
oo
CO
r>
CO
VO
CO
CO
CO
CO
1—I
co
a
ON
O\
oo
VO
s
OO
vo
co
-------
M
;*
CO
U
Z
a&-
i
O
o
•a
jl
CM
00
8
is
o>
CM
vo
55
o
(§
o cu
6 ^
CO 2
fi °
.
t
§
co
ON
cr
K
g
CO
O
t— <
§
§
B
CO
CN
,_,
§
I]
co
£
3
1
w
Q
§
E
u
VC
CS
CO
o
r— (
51
CO
i— <
CS
fl
CS
N Lv
1>
O
£
E
s
rH
£
oo
vo
a
CO
I
VO
CO
ON
V:
1
s
cs
a
3
^
t
•
'
t
'
,
I
«n
'
.
'
•
'
I
'
VO
1
.
'
t
'
,
•
^
§
i-
s
CO
CO
t-H
VO
$
X
1— 1
6T
^""^
•^
J
t>
c<-
s
ON
CO
CO
i— *
S
0;
OO
es
CS
1
c
CO
s
o
00
oo
a
CO
1
VO
CO
VO
ON
8
o
5
Tf
1
.
i
I
t
J
I
»0
'
1
1
1
'
(
1
VO
t
1
1
1
1
(
1
^
oc
cs
cr
CO
ON
CO
r-
o
VO
oo
ON
^
£
^^
-H
J
VI
§
ON
CO
i-H
»— I
ON
00
vo
ON
CS
cr
00
CO
CO
CO
3O
OO
ON
s
VO
CO
8
£5
00
* — 1
VO
»—<
ON
T— (
CO
s
J^
s
1
1
1
1
1
(
1
«
'
1
1
'
1
.
1
VO
1
1
'
1
1
1
1
oc
I/I
S
oo
VO
i-H
CS
CO
N
CO
s
Jj
00
J
•a
L,
1
Q
E
cr
ON
T-H
CO
s
oo
t-
s
ON
n
cs
^~,
'a
bb
A
1
ON
CS
r— t
O
r-H
a
N
N
O
CO
1
CO
o\
CO
-------
CO
i
1
•k
CM
oo
52
CO
-------
.a
•& a
PW
3
»
Is
a
PH
VO
T— t
z
•g
§04
11
£
<»
£
Ti
co
'
•
'
i
i
,
i
f-
•^
a
4-4
Q
>H
§
14
incinnati
u
§
T— 4
CM
1—4
ON
OO
00
R
ON
-
|
^4
z
z
u
f
o~b
T—4
T— 4
OO
1—4
OO
o\
in
«
R
ON
CM
S
T—4
VO
1—t
1—4
VO
i— 1
CO
CO
CO
t**
CO
VO
1
CO
T-H
•n
OO
CM
T—4
OO
X3
T—4
-
CM
8
>n
OO
i-H
CM
"n
ft
»-H
«
1
1
'
t
i
I
i
VO
i
i
i
i
i
i
i
t>
S
CM
VO
CO
VO
CM
CM
in
T-H
ON
CM
£
m
T-H
g
2
g
U
a
6
rO
Oc
I
1
a
CM
o
CO
T—4
CO
r-
ON
O
CO
T-H
OO
•n
vo
•o
a
CM
"Si
=L
O
CM
i-H
ON
i-H
T— 1
OO
VO
in
ON
SI
•<*•
!— 1
T-H
Tf
•n
VO
OO
CO
CO
CM
g
VO
CM
r-
T— 4
-
s
^
g
0
1
i-H
en
en
T-H
1
00
s
VO
T—4
in
OO
CM
ON
VO
T-H
1
CO
ON
S
OO
VO
T—4
T-H
OO
5?
CM
OO
CO
•n
co
co
CO
CO
1
en
T— 1
OO
en
T-H
OO
en
Tfr
OS
T-H
T—4
OS
T-H
T— t
OS
T-H
T-H
T-H
OS
T-H
T-H
T— (
Os
T-H
T-H
T-H
OS
T-H
T-H
T-H
m
t
1
•
1
'
,
'
VO
1
t
'
-
'
,
1
t-
CO
CM
•<*
i-H
CO
in
CM
m
CO
00
T-H
VO
OO
OO
s
T-H
§
n
ft
U
1
T-H
T-H
CO
ON
O
in
CM
ON
VO
ON
VO
1—4
T-H
|
00
CM
ON
T—4
•n
T-H
o\
CM
CM
CM
>n
CM
CM
CM
OO
T-H
CO
i
T-H
en
a
CO
o
OO
f-
CM
CM
T-H
VO
t-
CM
s
i-H
CO
VO
1—t
CM
a
Tt-
CO
t— 1
CO
VO
1—4
CO
T-H
1— 1
00
00
CO
in
T-H
T-H
T-H
CM
in
i
t
•
•
'
1
I
vo
'
1
1
1
t
1
1
-
A-12
-------
3
8*
o
oo
00
VO
•n
as
VO
os
9
3
N
S
CO
00
oo
vS
CM
5?
oo
CO
CO
CO
CO
VO
CO
oo
o\
OS
VO
OS
oo
>n
co
co
•n
00
§5
£
oo
CO
>n
CO
CM
a
CM
OO
CM
00
0)
CM
w
r>
§
U
fi
Z
D
§,
z
1X3
%
a
co
O
•a
1
1st
CS
'5
A-13
-------
g
:*
S§
1
h
Arithmetic
Mean
!
s
Jl
»—»
O'
p<
VO
T—*
-
I
X
g §*
a Q
§ o
H"
-------
PQ
en
CG
Arithmetic"!
Mean |
|
B
Is
§
3
g
vo
T-H
Z
•o
o
ff!
„
§ p.
S r*^
^
^
>>
rSn
rt
t/3
1
T-H
g
cs
5;
1
5
OS
X— V
&
Jx",
§
-a
O
i-J
«
o
fe
b
•c
e
°E
a
0
00
JN
D
•5
JJ
8
so
OO
^J~^
p
"3
OO
o
IO
so
i
i
1
t-
ro
so
h
OO
JS
1
n
i
-
a
S3
3
o\
T-H
T-H
s
T>H
Os
cs
CO
T-H
•
1
1
,
•
,
1
SO
'
1
1
1
1
,
t
1
o
5
1
CO
o
3
Os
CO
R
f*^
,_,
yj
§
•s
a
A
eu
1
^
q
£
o
1
OO
o
OO
0
IO
00
T— )
CO
,-^,
cd
1
o
1
x>
CO
CO
ON
OO
a
CO
>n
o
o
in
n
o
OO
n
ON
Os
o
8
a
rs
J
SO
O
SO
•o
a
•
t
•
,
•
,
•
VO
i
-
i
i
i
i
5
f
Os
IN
s
>n
o
§
|
CO
0
m
w
I-H
1
n
n
•*
CO
SO
CO
§
n
cs
CO
ON
CM
8
Os
IO
OO
sS
T— I
J
8
SO
CO
OO
CO
o
JS
1
SO
CO
OO
%
SO
CO
o
CO
n
CO
n
co
CO
IO
co
IO
o
CO
IO
CO
10
•o
-
>n
'
•
1
1
•
t
1
VO
I
•
1
1
•
1
-
1
ON
ON
Os
T-H
•n
n
R
SO
3
SO
8
CO
n
XI
so
:o
n
^
Q
w
HH
B
n
s
CO
0
2
S
•o
a
ro
1
00
IO
i
CO
o
1
•
-
1
t
1
1
VO
1
I
1
1
1
i
t
^
00
CO
r>
0
OO
0
•<*
Js
o
OS
SO
cs
CO
OO
00
CN
T-H
^
8
I
HH
Z
U
O
<3~
%
W
1
A-15
-------
'€
5
i*
II
S§
§1
«s
S
8 8
»-< ~* 00
is
co
o
°°
TT
If)
00
"">
vo
O\
*—*
CO
oo
tsi
cs
VO
CM
co
o\
VO
vo
o »H
CO O\
oo
vo
51;
O
vo
CS
CS
00
cs
CN
*
i
Oil vo
vo
ON
a
a
o\
o^
CO
a
s«-
CN
co
w
in
vo
10
VO
CM
IT)
VO
CS
CO
"O
vo
§
g
.O.
e
'CO
^
•i
CO
I
s
M
-
o
A-16
-------
Os
•e
OS
OO
OS
a
in
in
Os
co
CO
1
CO
CO
00
1—4
OS
a
IN
s
i
in
oo
c-
VO
in
VO
in
oo
o
in
in
in
OS
VO
PQ
00
Os
OO
o
o
VO
VO
OS
OS
00'
00
m
>n
oo
CO
Os
00
CO
VO
CO
oo
04
VO
CO
Os
VO
1—4
8
en
5
5
VO
Os
VO
VO
g
00
VO
in
o
CO
VO
s
3
o)
VO
•n
8
CO
s
ex
CO
VO
CO
X
VO
ON
CO
CO
VO
in
VO
Os
VO
CO
3
CO
VO
CO
CO
o
CO
g
IN
04
IN
OS
Os
VO
IN
04
04
04
oo
CO
IN
Os
00
Os
in
00
CO
Os
Os
CO
in
co
CO
CO
Os
04
s
00
04
Os
04
o\
Os
Os
\
o
CM
VO
04
VO
04
CO
in
VO
04
>n
VO
m
z
o
u
o
a
i
0
C/3
00
I
A-17
-------
g
u
•c *
n
1
S
CO
ON
CM
%
0
U
4-*
K
>>
3
incinnati
u
CO
CO
IT
r~-
g
S
oo
VO
*— 1
t-
CO
JB
v3
c*-
1
i
CO
CO
CO
CO
§
X
ro
T— i
*
oo
S5
c-
m
co
CM
in
ON
CO
0
ON
a
in
i
CO
en
E
CO
ON
uo
a
vo
r--
g
OC
T— 1
I—I
i
CM
OO
s
ON
t-
oc
CO
CO
CO
c
oc
ON
O
CO
CO
CO
1— I
T-H
3N
-
fc7
w
GO
3
"O
a
•4-1
Q
s
ON
i
M-
CO
CO
i
CM
s
oo
in
CO
T— !
CO
CM
0
^
$.
c
1
f-
CM
s
CM
TH
o
CO
oo
oc
CM
ON
o>
i
^
n-
ON
a
s
-
1
§
§
1
oo
i-H
i—4
CO
9
s
«
s
§
i
«
VO
g
CM
CM
vo
I/-
c<
1
in
CO
oc
S
CO
0
t— (
o
CM
CO
•n
t>
ON
1
CO
co
oo
CO
CM"
oo
t— (
op
o
i-H
s
%
J
ON
cr
cr
S
ON
vo
CO
CM
«
CM
OO
IT
S
in
CO
CM
0?
vo
•n
CO
CO
I
1
m
CO
CM
*-H
i-H
ON
S
CO
-
oc
Tt-
CS
CO
rf
T-H
t— t
CM
oo
1— 1
CM
CO
Tt'
-
in
ON
cr
S
T— 1
CO
vo
*— 1
ON
CO
T-H
in
in
CM
vo
1
1
S
CM
O
vo
CM
S
CM
8
1— 1
•>
in
1
S
oo
in
oo
oo
VD
in
a
-
Q
%
~
J
g
|
T-H
ON
•n
ON
CO
T-H
CO
a
CM
e
g
s
CO
T-H
ON
VO
CM
VO
s
CO
ON
OC
CM
es
CO
•<*
vo
vo
s
T— <
ON
ON
r-
t--
ro
a
*
t--
•r
CO
1
S
T-H
i-H
cn
a
CM
in
1
CM
S
ON
CO
CO
T-H
-
CO
00
T-H
CM
vo
s
S..
—
§
o
g
u
CO
A-18
-------
PQ
t»
Ig
o
S g
11
•Sj
g
OH
OO
s
s~
« CM
5~
a
|
VO
*
•g
o
£
3
1 1
£0
!
1
00
t
8
1
CO
§
*— (
oo
f-
ii
-*-l
£^
^
1
II Cincinnati
o\
r>
1
n
vo
CO
s
x>
M
T-H
2T
S
s
.£
0.
3
S
O
VO
8
00
CO
a
CM
h
§
Os
CM
5
oo
VO
oo
a
CO
1
1
§
CO
ON
3
o
T-H
?5
•*
o
s
8
M
§
i
UN
ro
CO
CM
•n
to
CM
§
S3
8
Js
*
VO
O
CO
CO
o
Js
CM
ffl
Os
7s
ON
OO
^
VO
s
CM
0
g
rH
S
s
,_,
s
H
B
CM
1
S
1
^.
s
2
S)
CM
CO
n
s
VO
CM
CM
Os
n
?5
CO
o
o
oo
o
in
CO
VO
CO
CM
in
VO
i— t
&
^.
1
o
§
3
CO
N
ro
ON
CM
•n
1
CO
ON
O
i
Jg
n
r>
CO
vo
00
r\
CM
oo
00
CO
CO
1
CM
0
O
CO
g
co
r-
H
CM
O
o
n
3
s
ON
T-H
g
H
I
VO
5
Js
•o
n
vo
r-l
•n
ON
CM
O
ON
a
ON
ON
CM
•O
•s^
vS
VO
c~
CO
ON
ON
O
VO,
O
o
CO
3s
8
T-H
^.
ON
1
o
S
o
VO
§
in
•n
i
Os
O
ON
CM
i
S
VO
ro
CM
VO
1
in
JN
0
•n
vo
t-H
VO
1
CO
CM
m
ON
r~
Js
o
T-H
ON
o
Js
£
*
oo
s
,-H
g-
s
O
1
Q
b
CO
ON
vo
oo
S
I
m
oo
0
CO
s
ON
»-H
CO
CM
^_^
h
W)
^
S
-
•<*•
00
VO
51
CO
co
CO
1
i
1
CM
m
CM
§
00
CO
a
*
«9
A-19
-------
g
o
Arithmeti
Mean
3
a
]§
rH
CX
cu
VO
rH
*
I
Oi
S a,
£0
R
H
U
&
CO
I
§
1
rH
i
vo
S
in
*
•s
n
%
Q
&
W
incinnati
U
g
o\
Y—t
01
T-H
^
CS
en
rH
a
vo
1?
"Bh
^
VO
en
1
1
«n
•<*•
en
CM
f-
^
§
oo
§
en
o\
"
o
ON
O
rH
g
§
g
u
1
S
rH
S
oo
rH
rH
ON
CM
r-
rH
rH
en
rH
rH
«
S
rH
OO
en
1
rH
OO
VO
rH
OO
S
rH
8
8
en
^
1
1
1
O
o
CN
T—l
n
0
en
oo
rH
r-
S
rH
S
i
P
vo
rH
CO
rH
rH
CO
rH
S
N
g
U
en
rH
m
rH
in
ON
CM
en
oo
s»
o
en
oo
CM
Stf
i
O
g*:
§s
o
"CO
A-20
-------
m
5
en
a
§S
i
en
&
8
8
a
es
vo
en
oo
es
s
00
en
in
en
vo
es
ON
en
ON
oo
en
oo
N
ON
a
>n
es
vo
vo
oo
eN
es
es)
vo
S3
VO
oo
es
es
vo
oo
CM
f-
*—<
en
00
S5
a
oo
en
en
ON
VO
%
3
O
VO
«n
VO
en
fc
en
55
vo
VO
in
oo
vo
i
•n
00
g
VO
es
es
VO
SI
en
es
>n
ON
oo
in
VO
en
in
VO
CM
es
en
vo
g
I
V)
u
§
Q
s
00
b
A-21
-------
Si
3
oo
t-
CM
1
1
cs
to
ts
CM
00
CO
cs
oo
oo
1
oo
in
oo
S
in
O\
I
CO
vo
CO
00
in
00
oo
CO
in
vo
oo
.8
5
5*
en
o
S
CO
CO
VO
§
oo
o
c-
oo
VO
CO
oo
CO
S3
CO
00
o
CN
in
vo
in
8
IN
S;
(S
OO
CO
VO
n
O\
CO
CO
CO
CO
CO
in
CO
B -
K rvi
O Sj
Cd C
O h
©9
•o
o
g
w
CO
vo
CM
CO
VO
CO
•n
vo
CO
in
B
O
W.
'CO
4!
_4D
&l
co
u
6
fi
i
Q
co
M
bb
U
o
u
u
A-22
-------
o
s
Me
a
s
VO
ON
ON
8
•n
i
ON
oo
CO
o
no
ON
ON
OO
CO
ON
CO
OO
oo
I/I
1
s
s
cn
s
s
O
n
3N
CM
00
s
>n
00
CO
oo
ON
i
8
CO
co
•n
00
•o
30
ro
00
ON
CO
cs
00
CS
CO
VO
(N
VO
cs
fM
VO
m
o
\O
O
(N
00
f-
oo
oo
VO
f-
ON
VO
(SI
>n
VO
1.
CS
CO
VO
VO
CO
VO
ON
VO
cs
a
co
s
w
(14
0
w
fi
CO
bo
a
oo
•£b
A-23
-------
cs
a
V)
Q
s
CO
0\
VO
VO
o
a
cs
in
in
VO
VO
CN!
vo
cs
VO
in
in
vo
vo
in
VO
CO
in
§•
0
0
0
U
co
K
A-24
-------
M
H
U
SP:
i
O
BS
•a
1
J/3
1«
li
CO
1-H
t~
SN
CM
CN
ON
a
a
a
8
vo
•0
"S
5
CO
00
•fi
b
ID
ON
O
VO
o
o
vo
ON
ON
CO
oo
-------
,0
o\
CM
Oi
CM
8
CM
CM
CM
CM
vo
vo
8
•n
o\
8
00
o\
8
*\*
••g
8
8
0\
o\
00
8
00
00
8
VO
0\
VO
8
8
VO
VO
VO
VO
VO
VO
VO
VO
VO
>n
8
VO
VO
s
vo
«n
vo
CM
VO
VO
vo
00
CM
«n
o\
VO
CM
vo
VO
VO
VO
VO
&
o
co
PQ
"So
O
A-26
-------
]i
'$*
£
PH
SS
CO
a
Ig
s g;
y
55
1
-2
*H
J£
s
ON
S
CM
CM
O
oo
00
VO
ON
«n
o
in
VO
§
H
I
1
13
O
o
2
I Cincinnati
m
CO
VO
in
os
8
g
in
f-
^
•*
CO
9
i
S
i— t
vo
§
o
^
^
8
>n
t~
in
CO
0
oo
CM
s
t— 1
in
VO
^
OS
^
f^'
in
in
CO
vo
Tf
o
**
CO
T-,
f-
ON
Os
CM
OO
ON
vo
i— <
CO
in
00
in
"*
oo
CO
vo
1— 1
in
V— t
ON
ON
vo
oo
ON
VO
VO
0
m
CM
2
f
I
B
00
VO
ON
CO
ON
i-<
9
VO
VO
^sj1
CO
VO
SJJ
CO
T— (
5
CM
s
ON
CO
CO
in
T— 1
ON
i— i
CO
o
00
f-
t-.
cli
0
OS
VO
CO
s
CM
,J
ON
^
'sJ"
«n
m
CO
<*
CO
s
in
r-
0
VO
ON
00
ON
CO
f-
c->
^
in
CO
in
ON
VO
_
n
CM
PH
PQ
A-27
-------
!*
M
i*
CO _
Jjj fcp
O B
^ O
(3 2
tf Q
O rsr
J*l 3
*f
g
_o
•S 4!
s~
f3
a,
•*
oo
CO
a
]§
»— (
o*
g
P*
V£>
*— I
2;
•CJ
(§
4-1
8 PI
c$ ^~<
g o
H
$
o
o
oo
vd
o
•
S
CM
S
¥—1
CO
s
Tt-
i-H
0
CO
s
7*
cs
s
t*-
1— (
1— )
s
VO
s
**
o\
o\
n
S
o\
o\
CM
>n
oo
CO
•*
S
CM
VO
VO
oo
8
•*
8
CM
o
t-H
l>
s
CO
%
CM
$
CO
1—4
vri
o
o\
s
vri
o
CO
•*
g
CM
8
CM
o\
t?~
•*
o
•<*
t~-
8
O\'
o
>n
i— i
o
o
CO
1— i
0
r-
8
>n
§
>n
oo
t--
CM
8
o\
8
>n
8
o
8
VO
O
O
t-~
ON
CO
&
co
f-
cs
^
Vfj
o
ON
S
00
in
vo
vb
m
o
CO
§
CM
00
CO
3
iri
S
r^
o
in
vd
ON
in
•*
S
"
0
oo
•*
-
VO
oo
in
g
ON
8
ON
O
O
CO
t— i
O
O
ON
0
ON
r-
>n
00
oo
in
8
ON
O
O
f-
8
CO
r-l
°
o
t — i
o
o
ON
OO
VO
c—
oo
CM
•n
co
oo
s
VO
3
-------
S
fc»
r^
r
p
<7
|z
(.
•
f-t
^
•^
•S
£
c
H
^S
s
^ff
' t»
o
•s
^^
y
OH
S
CO
cx
gj ^
J
u
CM
VO
Z
O
1 0,
II
H
«
K
"S,
1/3
^
O
T-H
0
•^c
0
o
CO
VO
pj
cs
r-
CO
(J
^-H
!S
^
0
c^
o
in
VO
ON
in
^
CO
S
ON
in
00
C-)
in
oo
o
S
in
in
S
^
^
*
in
CO
ON
O
vo
Tf
2
>-H
OO
So
in
o
in
S
in
£
ON
O
O
in
o
ON
r-H
0
OO
g
in
°
^j.
ON
t~
VO
^
1-1
O
in
en
O
in
CO
O
v-H
o3
VO
in
oo
VO
f-
-
5
ti
p
00
in
CO
T-l
•n
CO
i-H
ON
cs
oo
0
t-
2
fs*
!>•
«
(N
t-
in
ON
M
OO
o'
S
in
ON
M-
oo
CO
o
m
r-
in
ON
O
n\
ON
in
f-'
^
vo
§
VO
oo
Tt
^1
OO
in
o
o
in
o
o
in
*— 4
00
9
in
ON
in
t-
m
o
oo
c-
•n
o
o
o
o
(N
ts
in
in
cQ
in
oo
VO
A-29
-------
-------
APPENDIX B:
THE P-VALUES FOR THE TABLES IN CHAPTER 5
B-l
-------
TABLE B-l. P-VALUES FOR TABLE 5-1. PREABATEMENT CROSS-SECTIONAL
STRUCTURAL EQUATION MODELS FOR BOSTON STUDY
FLOOR DUST LEAD CONCENTRATION
SEM EQUATION
COEFFICIENTS
INTERCEPT
S
L
O
P
E
Floor -> Blood
Soil -* Blood
Window -» Blood
INTERCEPT
s
L
o
P
E
Soil -* Floor
Window -* Floor
INTERCEPT
S
L
O
P
E
Floor ->• Blood
Soil -*• Blood
Window -» Blood
dust Pb cone
Window -* Blood
dust Pb load
INTERCEPT
S
L
O
P
E
Soil -» Dust
Window -> Floor
dust Pb cone
Window -» Floor
dust Pb load
Model 1
Ggr 0.0001
B
*
Fgr 0.5199
Fgr
Cgr 0.0037
Dgr 0.5030
Dgr 0.0001
Ggr 0.0001
B
Fgr 0.3774
F
gr
gr
Cgr 0.2354
Dgr 0.4051
Dgr 0.0001
Dgr
Model 2 Model 3
0.0001 0.0001
0.4447 0.3555
0.4423
0.0041 0.0035
0.4149 0.5026
0.0001 0.0001
FLOOR DUST
0.0001 0.0001
0.3412 0.3459
0.0338 0.0303
0.6178 0.6436
0.0008 0.0011
Model 4 Model 5
0.0001
0.5553
0.2589
0.0034
0.4957
0.0001
LEAD LOADING
0.0001
0.4130
0.2332
0.2435
0.3678
0.0001
0.0001
0.0027
0.1231
0.0031
0.4537
0.0001
0.0001
0.4926
0.3658
0.0303
0.6295
0.0011
Model 6
0.0001
0.1150
0.0010
0.0107
0.0001
0.1840
0.0044
0.0221
B-2
-------
TABLE B-2. P-VALUES FOR TABLE 5-2. PREABATEMENT CROSS-SECTIONAL
STRUCTURAL EQUATION MODELS FOR BOSTON STUDY:
BLOOD LEAD TRUNCATED (9-22 pg/dL)
FLOOR DUST LEAD CONCENTRATION
SEM EQUATION
COEFFICIENTS
INTERCEPT
S
L
0
P
E
Floor -» Blood
Soil -> Blood
Window -» Blood
INTERCEPT
S
L
O
E
Soil -*• Floor
Window -» Floor
Model 1
Ggr 0.0001
Bgr
Fgr 0.1786
Fgr
Cgr 0.0217
Dgr 0.3386
Lgr 0.0001
Model 2
0.0001
0.1849
0.1457
0.0241
0.0360
0.0001
Model 3
0.0001
0.1961
0.1697
0.0235
0.3118
0.0001
FLOOR DUST
INTERCEPT
S
L
O
P
F
Floor -* Blood
Soil -* Blood
Window -* Blood
dust Pb cone
Window -> Blood
dust Pb load
INTERCEPT
S
L
0
P
E
Soil -> Dust
Window -* Floor
dust Pb cone
Window -* Floor
dust Pb load
Model 7
Ggr 0.0001
Bgr
Fgr 0.1832
pgr
Fgr
Cgr 0.3562
Dgr 0.3302
Lgr 0.0001
Lgr
Model 8
0.0001
0.2025
0.0304
0.7126
0.0050
Model 9
0.0001
0.2954
0.1807
0.0288
0.7664
0.0043
Model 4
0.0001
0.9114
0.3622
0.8260
0.0245
0.3143
0.0001
Model 5
0.0001
0.0178
0.0806
0.0125
0.3973
0.0001
Model 6
0.0001
0.1041
0.0026
0.0270
LEAD LOADING
Model 10
0.0001
0.1929
0.1271
0.3757
0.3395
0.0001
Model 11
0.0001
0.1763
0.3898
0.0282
0.7870
0.0041
Model 12
0.0001
0.0699
0.0071
0.0451
B-3
-------
TABLE B-3. P-VALUES FOR TABLE 5-3. PREABATEMENT CROSS-SECTIONAL
STRUCTURAL EQUATION MODELS FOR CINCINNATI STUDY:
DUST TYPE MODELS
DUST LEAD CONCENTRATION ALL AGES
SEM EQUATION
COEFFICIENTS
INTERCEPT Gg
Slope: Dust -»
Slope: Soil -*•
INTERCEPT
Slope: Soil -*
• Blood
Blood
Dust
B
F
Cg
D
With No Soil -* Blood Slope
Floor
0.0001
0.3647
—
0.0001
0.0001
With Soil -»
Entry . Window Floor
.0008
.3081
—
.0001
.0006
.0051
.3947
—
.0001
.0075
0.0031
0.9984
0.7063
0.0001
0.0001
Entry
.1603
.0001
.0706
.0001
.0009
Blood Slope
Window
.9967
.9997
.9995
.0001
.0075
Soil
0.0001
—
0.3015
0.0001
0.0001
DUST LEAD LOADING ALL AGES
INTERCEPT
Slope: Dust -»
Slope: Soil -*
INTERCEPT
Slope: Soil •*
INTERCEPT
Slope: Dust -»
Slope: Soil •*
INTERCEPT
Slope: Soil -*•
INTERCEPT
Slope: Dust -»
Slope: Soil -*
INTERCEPT
Slope: Soil -*•
• Blood
Blood
Dust
• Blood
Blood
Dust
• Blood
Blood
Dust
Gg
B
F
Cg
D
Gg
B
F
Cg
D
Gg
B
F
Cg
D
.0001
.5829
—
.2788
.0051
.7219
.0046
—
.0020
.0870
.0352
.2544
—
.3536
.0885
.0001
.4486
—
.2644
.0409
DUST LEAD
.4987
.2169
—
.0684
.0518
DUST
.0001
.0001
—
.2802
.7528
.0001
.7328
—
.0657
.7441
.0001
.1529
.3231
.2964
.0049
CONCENTRATION for
.7807
.2881
—
.0025
.0691
LEAD
.0001
.0001
—
.3109
.8682
.0001
.0001
.0549
.0113
.0602
.0001
.7449
.7617
.2536
.0506
AGE 42+
.0002
.0001
.0554
.0581
.2108
.0001
.7918
.3478
.0759
.6443
MONTHS
.0002
.0001
.0304
.0023
.0905
.0001
—
.3421
.2943
.0049
.0001
—
.0503
.0100
.0686
LOAD for AGE 42+ MONTHS
.0003
.0001
.0355
.4705
.0779
.1287
.0883
.7934
.3427
.6516
.0003
.0001
.0305
.3327
.9976
.0001
—
.0425
.3995
.0903
B-4
-------
TABLE B-4. P-VALUES FOR TABLE 5-4. PREABATEMENT CROSS-SECTIONAL
STRUCTURAL EQUATION MODELS FOR CINCINNATI STUDY: FLOOR DUST
SEM EQUATION
COEFFICIENTS
INTERCEPT Ggr
S
L
O
P
E
Floor-* Blood
Soil -> Blood
Window -» Blood
INTERCEPT Ggr
S
L
O
P
E
Soil -> Floor
Window -* Floor
Model 1
0.0001
0.4202
0.0347
0.0037
0.0009
0.0002
FLOOR DUST LEAD CONCENTRATION
Model 2 Models Model 4 Models Model 6 Model 7
0.0001 0.0001 0.0001 0.0001 0.0001 0.0001
0.0405 0.0486 0.9455 0.4629 0.3647
0.8660 0.7143 0.3015
0.7622 0.9645
0.0036 0.0037 0.0037 0.0037 0.0001 0.0001
0.0006 0.0009 0.0009 0.0009 0.0001 0.0001
0.0002 0.0002 0.0002 0.0002
Model 8
0.0031
0.9984
0.7063
0.0001
0.0001
FLOOR DUST LEAD LOADING
INTERCEPT Ggr
S
L
0
P
E
Floor -» Blood
Soil -» Blood
Window -» Blood
INTERCEPT Ggt
S
L
O
P
E
Soil -> Dust
Window -» Floor
Model 9
0.0001
0.4339
0.2831
0.5756
0.0026
0.0001
Model 10 Model 11 Model 12
0.0001 0.0001
0.4052 0.5829
0.4177 0.3421
0.3439
0.5611 0.2943 0.2788
0.0025 0.0049 0.0051
0.0001
Model 13
0.0001
0.1529
0.3231
0.2964
0.0049
B-5
-------
TABLE B-5. P-VALUES FOR TABLE 5-5. REPEATED MEASURES ANALYSIS OF
VARIANCE FOR BOSTON STUDY: EFFECT OF AGE REDUCTION
IN BLOOD LEAD (Er) BETWEEN ROUNDS 1 AND 3
Age Group
Study Group
Abate
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
Control
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
All Ages 0-17 Months
(N=150) (N=19)
18-41 Months 42+ Months
(N=100) (N=31)
Response (Er)
0.0042
0.6157
0.0159
0.0064
0.6414
0.0212
0.7312
0.3844
0.2019
Log
0.7543
0.4093
0.2307
0.0020
0.1280
0.0769
Response (Er)
0.0084
0.2917
0.0809
0.4271
0.6163
0.2320
0.2465
0.9614
0.3186
TABLE B-6. P-VALUES FOR TABLE 5-6. REPEATED MEASURES ANALYSIS OF
VARIANCE FOR BOSTON STUDY: EFFECT OF AGE REDUCTION
IN BLOOD LEAD (Er) BETWEEN ROUNDS 3 AND 4
Study Group
Abate
BOS P-S
BOS PI-S
BOS PI-S
BOS P-S
BOS PI-S
BOS PI-S
Control
BOS SPI
BOS SPI
BOS P-S
BOS SPI
BOS SPI
BOS P-S
Age Group
All Ages 0-17 Months 18-41 Months 42+ Months
(N=147) (N=18) (N=98) (N=31)
0.1222
0.0006
0.0788
0.3772
0.0217
0.1931
Response (Er)
0.7834
0.4907
0.6295
Log Response (Er)
0.6609
0.6127
0.8366
0.5365
0.0138
0.0761
0.8137
0.0736
0.1280
0.1172
0.1007
0.9819
0.2065
0.3125
0.7795
B-6
-------
TABLE B-7. P-VALUES FOR TABLE 5-7. REPEATED MEASURES ANALYSIS OF
VARIANCE FOR BOSTON STUDY: EFFECT OF RACE OR SEX
Group
Study Group
Rounds 1-3:
All
(N=150)
Black
(N=75)
Nonblack
(N=32)
Male
(N=80)
Female
(N=70)
Rounds 3-4: (N=147) (N=74)
(N=32)
(N=78)
(N=69)
Abate
Control
Reduction in Blood Lead (Er) Between Rounds 1 and 3
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
BOS P-S BOS SPI
BOS PI-S BOS SPI
BOS P-S BOS P-S
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
BOS P-S BOS SPI
BOS PI-S BOS SPI
BOS P-S BOS P-S
0.0042
0.6157
0.0159
0.0502
0.4116
0.2670
0.2477
0.9031
0.1136
0.0162
0.9105
0.0205
0.1306
0.5665
0.3213
Reduction in Blood Lead (Er) Between Rounds 3 and 4
0.1222
0.0006
0.0788
Reduction
0.0136
0.0044
0.5798
in Log Blood
0.5241 0.2881
0.3577 0.0498
0.1990 0.4091
Lead (E) Between Rounds 1
0.2198
0.0039
0.0929
and 3
0.0064 0.0644 0.6097 0.0099 0.1796
0.6414 0.2843 0.4840 0.9756 0.5870
0.0212 0.4432 0.1178 0.0099 0.3987
Reduction in Log Blood Lead (Er) Between Rounds 3 and 4
0.3772 0.0136 0.9262 0.2881 0.2198
0.0217 0.0044 0.6607 0.0498 0.0039
0.1931 0.5798 0.8046 0.4091 0.0929
B-7
-------
TABLE B-8. P-VALUES FOR TABLE 5-8. REPEATED MEASURES ANALYSIS OF
VARIANCE FOR BOSTON STUDY: EFFECT OF TRUNCATION REDUCTION
IN BLOOD LEAD (Er) BETWEEN ROUNDS 1 AND 3
Truncation Category
7-24 /tg/dL
Study Group Age 18-41: (N=100)
Age 42-52: (N=31)
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
Abate
SPI
PI-S
SPI
SPI
PI-S
SPI
SPI
PI-S
SPI
SPI
PI
SPI
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
BOS
Control
P-S
P-S
PI-S
P-S
P-S
PI-S
P-S
P-S
PI-S
P-S
P-S
PI-S
Change in
0.0020
0.1280
0.0769
Change hi
0.4271
0.6163
0.2320
Change hi Log
0.0084
0.2917
0.0809
Change hi Log
0.2465
0.9614
0.3186
10-19 /tg/dL 7-19 /tg/dL
(N=67) (N=92)
(N=16)
Blood Lead for Age
0.0129
0.3616
0.1021
Blood Lead for Age
0.7966
0.4695
0.3448
Blood Lead for Age
0.0336
0.5450
0.1189
Blood Lead for Age
0.7846
0.3759
0.2644
(N=29)
Group 18-41
0.0039
0.2665
0.0590
Group 42-52
0.2969
0.8885
0.2733
Group 18-41
0.0115
0.3828
0.0810
Group 42-52
0.1917
0.8064
0.3302
10-24 /tg/dL
(N=75)
(N=18)
Months
0.0070
0.1847
0.1250
Months
1.00000
0.2901
0.
Months
0.
0.
0.
Months
0.
0.
0.
2901
0270
4433
1143
9389
2618
2367
TABLE B-9. P-VALUES FOR TABLE 5-9. REPEATED MEASURES ANALYSIS OF
VARIANCE FOR BOSTON STUDY: EFFECT OF TRUNCATION
REDUCTION IN BLOOD LEAD (Er) BETWEEN ROUNDS 3 AND 4
Study Group
Abate
BOS
BOS
BOS
BOS
BOS
BOS
P-S
PI-S
PI-S
P-S
PI-S
P-S
Control
BOS
BOS
BOS
BOS
BOS
BOS
SPI
SPI
P-S
SPI
SPI
P-S
7-24 /tg/dL
(N=31)
Truncation
10-19 /tg/dL
(N=16)
Change in Blood Lead
0.5365
0.0138
0.0765
Change
0.8137
0.0736
0.2526
0.0330
0.3025
hi Log Blood Lead
0.2430
0.0377
0.3431
Category
7-19 /tg/dL
(N=29)
for Age Group 18-41
0.7232
0.0788
0.1618
for Age Group 18-41
0.9443
0.1479
0.1689
10-24 /tg/dL
(N=18)
months
0.1696
0.0040
0.1368
Months
0.1739
0.0093
0.2295
B-8
-------
TABLE B-10. P-VALUES FOR TABLE 5-10. REPEATED MEASURES ANALYSIS
OF VARIANCE FOR CINCINNATI STUDY: EFFECT OF AGE IN BETWEEN
ROUNDS 1 AND 4
Study Group
Abate
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN I-SE (D)
CIN I-SE (D)
CIN I-SE (F)
CIN I-SE (F)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN I-SE (D)
CIN I-SE (D)
CIN I-SE (F)
CIN I-SE (F)
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (G)
CIN NT (M)
CIN NT (G)
CIN NT (M)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (G)
CIN NT (M)
CIN NT (G)
CIN NT (M)
Age Group
All Ages 9-17 Months 18-41 Months
(N=223) (N=69) (N=80)
Reduction in Blood Lead (Er) Between Rounds 1
0.0729 0.0014
0.0365 0.0199
0.6023 0.0534
0.0492 0.5731
0.3640 0.7196
Reduction hi Log Blood
Between Rounds 1
0.0032 0.0001
0.0016 0.0046
0.2959 0.0098
0.0595 0.6488
0.2139 0.5151
0.0722
0.5882
0.1038
0.9563
0.8852
Lead (Er)
and 4
0.0669
0.1986
0.1618
0.9666
0.7619
42+ Months
(N=70)
and 4
0.1783
0.1992
0.0173
0.0028
0.0818
0.5310
0.0384
0.0219
0.0078
0.0676
B-9
-------
TABLE B-ll. P-VALUES FOR TABLE 5-11. REPEATED MEASURES ANALYSIS
OF VARIANCE FOR CINCINNATI STUDY: EFFECT OF AGE IN BETWEEN
ROUNDS 4 AND 7
Age Group
Study Group
All Ages
(N=223)
9-17 Months
(N=69)
18-41 Months
(N=80)
42+ Months
(N=70)
Abate
Control
Reduction hi Blood Lead (Er) Between Rounds 4 and 7
GIN NT (G)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN NT (M)
GIN NT(G)
GIN NT (M)
GIN I-SE (D)
GIN I-SE (F)
0.1440
0.4298
0.3201
0.9920
0.5164
0.3869
0.6781
0.5018
0.7354
0.5826
0.7044
0.8856
0.7442
0.6530
0.8982
Reduction hi Log Blood Lead (E,) Between Rounds 4 and 7
GIN NT (G)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN NT (M)
GIN NT(G)
GIN NT (M)
GIN I-SE (D)
GIN I-SE (F)
0.0622
0.1257
0.3429
0.7484
0.6793
0.6735
0.4815
0.8921
0.6539
0.7254
0.1227
0.1866
0.6276
0.3885
0.5682
B-10
-------
TABLE B-12. P-VALUES FOR TABLE 5-12. REPEATED MEASURES ANALYSIS
OF VARIANCE FOR CINCINNATI STUDY: EFFECT OF TRUNCATION BETWEEN
ROUNDS 1 AND 4
Study Group
Abate Versus Control
Abate
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CINI-SE(F)
All
Truncation
7-24 /ig/dL
Reduction in Blood Lead
(N=69)
0.0014
0.0199
0.0534
0.5731
0.7196
(N=33)
0.4899
0.6991
0.4300
Reduction in Blood Lead
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
(N=80)
0.0722
0.5882
0.1038
0.9563
0.8852
(N=67)
0.1018
0.6432
0.1366
0.5682
0.6966
Reduction in Blood Lead
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
(N=70)
0.1783
0.1992
0.0173
0.0028
0.0818
(N=47)
0.4788
0.2337
0.0594
0.0107
0.3625
Category
10-19 jtg/dL 10-24 jtg/dL
(Er) Age 9-17 Months
(N=15)
0.8627
0.8599
0.6409
(Er) Age 18-41 Months
(N=38) (N=43)
0.0527 0.0521
0.2596 0.3909
0.1417 0.1048
0.4924 0.3518
0.7628 0.8940
(Er) Age 42+ Months
(N=31) (N=36)
0.5605 0.6069
0.5561 0.6233
0.2629 0.3350
0.2353 0.1276
0.9944 0.9834
B-ll
-------
TABLE B-13. P-VALTJES
OF VARIANCE FOR
FOR TABLE 5-13. REPEATED MEASURES ANALYSIS
CINCINNATI STUDY: EFFECT OF TRUNCATION
BETWEEN ROUNDS 1 AND 4
Truncation Category
Study Group
Abate Versus Control
Abate
GIN NT (G)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN NT (G)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN NT (G)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
GIN SEI (P)
Control
GIN NT (M)
GIN NT(G)
GIN NT (M)
GIN I-SE (D)
GIN I-SE (F)
GIN NT (M)
GIN NT(G)
GIN NT (M)
GIN I-SE (D)
GIN I-SE (F)
GIN NT (M)
GIN NT(G)
GIN NT (M)
GIN I-SE (D)
GIN I-SE (F)
All
Reduction in Log
(N=69)
0.0001
0.0046
0.0098
0.6488
0.5151
Reduction in Log
(N=80)
0.0669
0.1986
0.1618
0.9666
0.7619
Reduction in Log
(N=70)
0.5310
0.0384
0.0219
0.0078
0.0676
7-24 /ig/L
Blood Lead (Er) Age
(N=33)
0.6858
0.9496
0.2463
Blood Lead (Er) Age
(N=67)
0.1757
0.4559
0.2726
0.7355
0.6738
Blood Lead (EJ Age
(N=47)
0.1589
0.0941
0.0147
0.3864
10-19 /*g/L
9-17 Months
(N=15)
0.9889
0.7852
0.7416
18-41 Months
(N=38)
42+ Months
(N=31)
B-12
-------
TABLE B-14. P-VALUES FOR TABLE 5-14. REPEATED MEASURES ANALYSIS
OF VARIANCE FOR CINCINNATI STUDY: EFFECT OF TRUNCATION
BETWEEN ROUNDS 4 AND 7
Truncation Category
Study Group
Abate Versus Control
Abate
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
All 7-24 /tg/dL
Reduction
(N=69)
0.3869
0.6781
0.5018
0.7354
0.5826
Reduction
(N=80)
0.7044
0.8856
0.7442
0.6530
0.8982
Reduction
(N=70)
in Blood Lead
(N=33)
0.9241
0.4648
0.3079
in Blood Lead
(N=67)
0.8282
0.8166
0.7475
0.5430
0.8460
in Blood Lead
(N=47)
0.2750
0.2785
0.9935
0.5976
0.8317
10-19 Mg/dL
(Er) Age 9-17
(N=15)
0.8588
0.7252
0.9711
(Er) Age 18-41
(N=38)
0.6802
0.8214
0.7602
0.6156
(E,) Age 42+
(N=31)
0.3188
0.3822
0.9229
0.6980
0.8346
10-24 pg/L
Months
Months
(N=43)
0.9919
0.7274
0.8558
0.5004
0.8702
Months
(N=36)
0.2820
0.3515
0.9220
0.6659
0.8277
B-13
-------
TABLE B-15. P-VALUES FOR TABLE 5-15. REPEATED MEASURES ANALYSIS
OF VARIANCE FOR CINCINNATI STUDY: EFFECT OF TRUNCATION
BETWEEN ROUNDS 4 AND 7
Truncation Category
Study Group
Abate Versus Control
Abate
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT (G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
Control
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT .(M)
CIN I-SE (D)
CIN I-SE (F)
CIN NT (M)
CIN NT(G)
CIN NT (M)
CIN I-SE (D)
CIN I-SE (F)
All
Reduction
(N=69)
Reduction
(N=80)
0.6735
0.4815
0.8921
0.6539
0.7254
Reduction
(N=70)
0.1227
0.1866
0.6276
0.3885
0.5682
7-24 Otg/dL) 10-19 (/ig/dL) 10-24 pg/L
in Log Blood Lead
(N=33)
0.5801
0.7779
0.8283
in Log Blood Lead
(N=67)
0.8043
0.8345
0.8715
0.5018
0.6618
in Log Blood Lead
(N=47)
0.5405
0.4497
0.8869
0.5114
0.7
(Er) Age 9-17 Months
(N=15)
(Er) Age 18-41 Months
(N=38) (N=43)
(Er) Age 42+ Months
(N=31) (N=36)
B-14
-------
TABLE B-16. P-VALUES FOR TABLE 5-16. REPEATED MEASURES ANALYSIS
OF VARIANCE FOR BALTIMORE STUDY: EFFECT OF AGE
Age Group
Study Group
All Ages
(N=463)
< 18 Months
(N=16)
18-41 Months
(N=88)
42+ Months
(N=161)
Abate
Control
Reduction in Blood Lead (Er) Between Rounds 3 and 4
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALP1 0.8893 0.2790 0.8040 0.9097
BALP2 0.1625 0.3593 0.1168
Reduction hi Blood Lead (Er) Between Rounds 3 and 6
BAL PI 0.3676 0.9313 0.6495 0.7862
BA1LP2 0.6456 0.1654 0.6910
Reduction in Log Blood Lead (Er) Between Rounds 3 and 4
BAL PI 0.8563 0.9761
BAL P2 0.9882 0.3854
Reduction hi Log Blood Lead (Er) Between Rounds 3 and 6
BAL PI 0.8420
BAL P2 0.9683
B-15
-------
TABLE B-17. P-VALUES FOR TABLE 5-17. REPEATED MEASURES ANALYSIS
OF VARIANCE FOR BALTIMORE STUDY: EFFECT OF TRUNCATION
BETWEEN ROUNDS 3 AND 4
TRUNCATION CATEGORY
ALL 10-19 10-24 7-19 7^24
<18:(N=16) <18:(N=2) <18: (N=4) <18: (N=15) <18: (N=7)
18-42: (N=88) 18-42: (N=32) 18-42: (N=42) 18-42: (N=54) 18-42: (N=64)
STUDY GROUP >42: (N=161) >42: (N=47) >42: (N=53) >42: (N=110) >42: (N=120)
ABATE
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
BALSP
CONTROL
BALP1
BALP2
BALP1
BALP2
BALP1
BALP2
BALP1
BALP2
BALP1
BALP2
BALP1
BALP2
REDUCTION IN BLOOD LEAD (E,) FOR AGE < 18 MONTHS
REDUCTION IN BLOOD LEAD (E,) FOR AGE 18-41 MONTHS
0.8040 0.3455 0.1255 0.6136
0.3593
REDUCTION IN BLOOD LEAD (Er) FOR AGE 42+ MONTHS
0.9097 0.7233 0.8511 0.7934 0.8338
0.1168 0.2096 0.3194
REDUCTION IN LOG BLOOD LEAD (E,) FOR AGE < 18 MONTHS
REDUCTION IN LOG BLOOD LEAD (E,) FOR AGE 18-41 MONTHS
REDUCTION IN LOG BLOOD LEAD (Er) FOR AGE 42+ MONTHS
0.9761 0.5998 0.6122 0.4523 0.7746
0.3854 0.4184 0.4700
B-16
-------
TABLE B-18. P-VALUES FOR TABLE 5-18. REPEATED MEASURES ANALYSIS
OF COVARIANCE FOR BOSTON STUDY: EFFECT OF AGE AND LOG DUST
LEAD CONCENTRATION REDUCTION IN LOG BLOOD LEAD (E,.) BETWEEN
ROUNDS 1 AND 3
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
Age Group
All Ages 9-17 Months 18-41 Months
(N=142) (N=17) . (N=97)
0.8014
0.1341
0.2106
Covariate: Log Dust Pb
0.9442
0.1334
0.1672
0.8276
0.3036
0.4492
Concentration
0.8363
0.2589
0.4131
0.3125
0.1323
0.5690
0.4275
0.1440
0.4895
42+ Months
(N=28)
0.4279
0.5034
0.8213
0.4013
0.4962
0.8216
TABLE B-19. P-VALUES FOR TABLE 5-19. REPEATED MEASURES ANALYSIS
OF COVARIANCE FOR BOSTON STUDY: EFFECT OF AGE AND
LOG DUST LEAD AND SOIL LEAD CONCENTRATION REDUCTION
IN LOG BLOOD LEAD (Er) BETWEEN ROUNDS 1 AND 3
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
Age Group
All Ages 9-17 Months 18-41 Months
(N=142) (N=17) (N=97)
0.1845
0.5148
0.5601
42+ Months
(N=28)
0.7040
0.5661
0.6962
Covariate: Log Dust Lead Concentration
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S,
BOS PI-S
0.4880
0.1557
0.4426
0.4556
0.6810
0.9828
Covariate: Log Soil Lead Concentration
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
0.5469
0.6789
0.2891
0.8314
0.6742
0.5028
B-17
-------
TABLE B-20. P-VALUES FOR TABLE 5-20. REPEATED MEASURES ANALYSIS
OF COVARIANCE FOR: BOSTON STUDY EFFECT OF AGE AND LOG DUST
LEAD CONCENTRATION REDUCTION IN LOG BLOOD LEAD (E,) BETWEEN
ROUNDS 3 AND 4
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
Age Group
All Ages 9-17 Months 18-41 Months 42+ Months
(N=142) (N=17) (N=97) (N=28)
0.6993 0.4752 0.1995
0.0367 0.0291
0.0275 0.0026
0.5128
0.8716
0.5167
Covariate: Log Dust Lead Concentration
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
0.7205 0.4608
0.0245
0.0206
0.4667
0.9146
0.5090
TABLE B-21. P-VALUES FOR TABLE 5-21. REPEATED MEASURES ANALYSIS
OF COVARIANCE FOR BOSTON STUDY: EFFECT OF AGE AND LOG DUST
LEAD LOADING REDUCTION IN BLOOD LEAD (Er) BETWEEN ROUNDS 3 AND 4
Age Group
Study Group
Abate Versus Control
All Ages
(N=128)
9-17 Months
(N=15)
18-41 Months
(N=89)
Covariate: Log Dust Lead Loading
42+ Months
(N=24)
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
0.8045
0.2436
0.1774
0.5540
0.0558
0.1251
0.5013
0.2273
0.1596
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
0.9244
0.0788
0.0700
0.5885
0.0167
0.0370
0.7092
0.1821
0.1568
B-18
-------
TABLE B-22. P-VALUES FOR TABLE 5-22. REPEATED MEASURES ANALYSIS
OF COVARIANCE FOR BOSTON STUDY: EFFECT OF AGE AND LOG DUST
LEAD CONCENTRATION ON BLACKS REDUCTION IN LOG
BLOOD LEAD (Er) BETWEEN ROUNDS 1 AND 3
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
Age Group
All Ages 9-17 Months 18-41 Months 42+ Months
(N=71) (N=ll) (N=44) (N=16)
0.5834
0.0185
0.0694
0.2446
0.0070
0.0663
0.5699
0.4295
0.6882
Covariate: Log Dust Lead Concentration
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
0.6832
0.0188
0.0583
0.3190
0.0078
0.0583
0.5484
0.4414
0.7043
TABLE B-23. P-VALUES FOR TABLE 5-23. REPEATED MEASURES ANALYSIS
OF COVARIANCE FOR BOSTON STUDY; EFFECT OF AGE AND
LOG DUST LEAD LOAD ON BLACKS REDUCTION
IN LOG BLOOD LEAD (Er) BETWEEN ROUNDS 1 AND 3
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
All Ages
(N=71)
0.7408
0.2493
0.4199
Covariate: Log
0.8594
0.3266
0.3000
Age Group
9-17 Months 18-41 Months
(N=ll) (N=44)
0.2904
0.0482
0.2902
Dust Lead Load
0.6841
0.1386
0.2902
42+ Months
(N=16)
0.6548
0.7336
0.6941
0.6595
0.6151
0.7171
B-19
-------
TABLE B-24. P-VALUES FOR TABLE 5-24. REPEATED MEASURES ANALYSIS
OF COVARIANCE FOR BOSTON STUDY: EFFECT OF AGE AND LOG DUST
LEAD CONCENTRATION, SOIL LEAD CONCENTRATION ON BLACKS
REDUCTION IN BLOOD LEAD (E_) BETWEEN ROUNDS 1 AND 3
Study Group
Abate Versus Control
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
All Ages
(N=71)
0.6991
0.3351
0.5369
Covariate: Log Dust
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
0.6574
0.0220
0.0603
Covariate: Log Soil
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
0.8834
0.4605
0.5550
Age Group
9-17 Months 18-41 Months
(N=ll) (N=44)
0.3049
0.5147
0.8734
Lead Concentration
0.2806
0.0068
0.0451
Lead Concentration
0.9902
0.1370
0.1267
42+ Months
(N=16)
TABLE B-25. P-VALUES FOR TABLE 5-25. REPEATED MEASURES ANALYSIS
OF COVARIANCE EFFECT OF AGE AND LOG DUST LEAD LOADING ON
BLACKS IN BOSTON STUDY REDUCTION IN LOG BLOOD LEAD (Er)
BETWEEN ROUNDS 3 AND 4
Study Group
Abate Versus Control
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
BOS SPI BOS P-S
BOS PI-S BOS P-S
BOS SPI BOS PI-S
' All Ages
(N=128)
0.6347
0.3690
0.2207
Covariate: Log
0.8309 •
0.1594
0.1431
Age Group
9-17 Months 18-41 Months
(N=15) (N=89)
0.4728
0.0479
0.1276
Dust Lead Loading
0.4090
0.0152
0.0707
42+ Months
(N=24)
0.4505
0.1304
0.1004
0.6792
0.1042
0.0983
B-20
-------
TABLE B-26. P-VALUES FOR TABLE 5-26, REPEATED MEASURES ANALYSIS
OF COVARIANC EEFFECT OF AGE AND LOG DUST LEAD CONCENTRATION
ON BLACKS IN BOSTON STUDY REDUCTION IN LOG BLOOD LEAD (Er)
BETWEEN ROUNDS 3 AND 4
Study Group
Abate Versus Control
BOS SPI
BOS PI-S
BOS SPI
BOS SPI
BOS PI-S
BOS SPI
BOS P-S
BOS P-S
BOS PI-S
BOS P-S
BOS P-S
BOS PI-S
All Ages
(N=64)
0.8096
0.1268
0.1316
Covariate: Log Dust
0.8977
0.0859
0.1251
Age Group
9-17 Months 18-41 Months
(N=8) (N=40)
0.9335
0.0483
0.1710
Lead Concentration
0.9555
0.0349
0.1695
42+ Months
(N=16)
0.9820
0.4692
0.5004
0.9438
0.4962
0.4878
TABLE B-27. P-VALUES FOR TABLE 5-27. REPEATED MEASURES ANALYSIS
OF COVARIANCE IN CINCINNATI STUDY REDUCTION IN BLOOD LEAD (Er)
BETWEEN ROUNDS 1 AND 4
Log
Floor Dust
Study Group Concentration
Log
Log Log Floor
Entry Dust Window Dust Dust Pb
Concentration Concentration Loading
Abate Versus Control
CIN NT(G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT(M)
CIN NT(G)
CIN NT(M)
CIN I-SE(D)
CIN I-SE(F)
.8510
.7127
.7360
.4674
.7149
.3061
.6939
.3876
.8180
,5748
Intercept Effect
.0949
.9510
.1110
.3970
.9169
.1403
.6924
.1861
.5635
.2227
Log
Entry
Dust Pb
Loading
.0494
.0815
.3416
.0610
.2581
Log
Window
Dust Pb
Loading
.8739
.1331
.4395
.0144
.6430
Covariate Effect
CIN NT(G)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN SEI (P)
CIN NT(M)
CIN NT(G)
CIN NT(M)
CIN I-SE(D)
CIN I-SE(F)
.9754
:5155
.8118
.3838
.6102
.4978
.9633
.4789
.9945
.6906
.0496
.7537
.0803
.4317
.9691
.2925
.8128
.2221
.3684
.2797
.2720
.3314
.6066
.1133
.3975
.5571
.3686
.3659
.0251
.6734
B-21
-------
TABLE B-28. P-VALUES FOR TABLE 5-30. LONGITUDINAL STRUCTURAL
EQUATION MODELS REGRESSION COEFFICIENTS IN BOSTON STUDY
FREE BLOOD LEAD PERSISTENCE FACTOR
REGRESSION COEFFICIENT
Predictor
Variable
ALL GROUPS
BOS SPI
^^ BOSPI-S
BOS P-S
ALL GROUPS
SoilPb Bosspl
ROUnd3 BOSPI-S
BOS P-S
ALL GROUPS
DustPb Bosspl
Cone.
Rounds BOSPI-S
BOS P-S
Blood Lead Round 1
Intercept
Soil Pb Round 1
Dust Pb Round 1
MODEL
1
0.4767
0.3463
0.7575
0.0710
0.0001
0.1934
0.0191
MODEL MODEL
2 10
RESPONSE VARIABLE: BLOOD
0.7504 0.8488
0.9943
0.9693
0.4404
0.7376
0.8584
0.9944
0.7467
0.8365
0.1794 0.4960
RESPONSE VARIABLE: BLOOD
0.0001 0.0001
0.1831 0.6006
0.0400 0.8170
RESPONSE VARIABLE: FLOOR DUST LEAD
Intercept
Soil Pb Round 1
Window Dust Pb Round 1
0.0001
0.1110
0.0001
0.0001 0.0001
0.1921 0.6307
0.0001 0.0001
MODEL
11
LEAD ROUND 3
0.8468
0.9943
0.9942
0.7466
0.8341 ,
0.4942
LEAD ROUND 1
0.0001
0.6042
0.8209
CONCENTRATION
0.0022
0.6234
0.0001
MODEL
17
0.8530
0.5376
0.4464
0.9746
0.0055
0.3964
0.0001
0.1672
0.0183
ROUND 1
0.0002
0.9375
0.0001
RESPONSE VARIABLE: FLOOR DUST LEAD
CONCENTRATION ROUND 3
ALL GROUPS
Intercept BQS spl
BOS PI-S
BOS P-S
Soil Pb ALL GROUPS
Cone. gos SPI
RoUnd3 BOSPI-S
BOS P-S
Window Dust Pb Cone.
Rounds
0.0001
0.0001
0.0077
0.0001 0.0001
0.0001 0.0001
0.0135 0.0076
0.0001
0.6734
0.8706
0.8352
0.4355
0.2045
0.0001
0.0003
0.0697
B-22
-------
TABLE B-29. P-VALUES FOR TABLE 5-32. LONGITUDINAL STRUCTURAL
EQUATION MODELS REGRESSION COEFFICIENTS IN BOSTON STUDY
USING FIXED BLOOD LEAD PERSISTENCE FACTOR
Predictor
Variable
MODEL
1
REGRESSION COEFFICIENT
MODEL MODEL MODEL MODEL
2 10 11 17
MODEL
30
RESPONSE VARIABLE: BLOOD LEAD ROUND 3
ALL GROUPS
BOS SPI
mterCept BOSPI-S
BOS P-S
Soil ALL GROUPS
Pb BOS SPI
Round 3 BOS PI-S
BOS P-S
Floor ALL GROUPS
Dust Pb BOS SPI
Cone. BOS PI-S
RoungS BOS P-S
Blood Pb Round 1
0.0269
0.1503
0.0568
0.0318 0.1524 0.1021
0.8045
0.0327
0.0269
0.6593 0.3903 0.5743
0.6401
0.1294
0.0757
0.1917 0.0468
0.1311 0.2348
0.0507 0.0221
0.0291 0.0312
0.9768
0.0119
0.0200
0.3162
0.0437
RESPONSE VARIABLE: BLOOD LEAD ROUND 1
Intercept
Soil Pb Round 1
Dust Pb Round 1
0.0001
0.3330
0.7917
0.0001 0.0001 0.0001 0.0001
0.3255 0.4958 0.8490 0.4735
0.7907 0.5487 0.5383 0.6413
0.0001
, 0.7502
0.6100
RESPONSE VARIABLE: FLOOR DUST LEAD
CONCENTRATION ROUND 1
Intercept
Soil Pb Round 1
Window Dust Pb Round 1
0.0001
0.1372
0.0001
0.0001 0.0001 0.0001 0.0001
0.1323 0.2802 0.1824 0.3020
0.0001 0.0001 0.0001 0.0001
0.0001
0.1927
0.0001
RESPONSE VARIABLE: FLOOR DUST LEAD
CONCENTRATION ROUND 3
ALL GROUPS
BOS SPI
MerCept BOSPI-S
BOS P-S
Soil Pb ALL GROUPS
Cone. BOS Spj
ROUnd3 BOSPI-S
BOS P-S
Window Dust Pb
Cone. Round 3
0.0001
0.0001
0.0079
0.0001 0.0001 0.0001
0.0001
0.8698
0.3021
0.0001 0.0001 0.0001
0.3886
0.2298
0.3883
0.0082 0.0062 0.0157 0.0107
0.0001
0.5948
0.5005
0.3506
0.3651
0.2386
0.0166
B-23
-------
TABLE B-30. P-VALUES FOR TABLE 5-34. LONGITUDINAL STRUCTURAL
EQUATION MODELS REGRESSION COEFFICIENTS IN BOSTON STUDY
USING FIXED BLOOD LEAD PERSISTENCE FACTOR FOR MALES
REGRESSION COEFFICIENT
Predictor
Variable
MODEL
1
MODEL MODEL MODEL MODEL
2 10 11 17
MODEL
30
RESPONSE VARIABLE: BLOOD LEAD ROUND 3
ALL GROUPS
BOS SPI
BOS PI-S
BOS P-S
ALL GROUPS
Soil Pb BQS spl
Round 3
BOS PI-S
BOS P-S
ALL GROUPS
Floor Dust BOSSPI
Pb Cone.
Rounds BOSPI-S
BOS P-S
Blood Pb Round 1
0.0002
0.0001
0.4309
0.0086 0.0513 0.0088
0.1630
0.0120
0.1089
0.3791 0.9339 0.3859
0£f\f\(\
.o30"
0.0001
0.0307
0.4235 0.1455
0.1533 0.6422
0.3364 0.3385
0.1506 0.3398
0.0717
0.0031
0.0285
0.9301
0.2619
RESPONSE VARIABLE: BLOOD LEAD ROUND 1
Intercept
Soil Pb Round 1
Dust Pb Round 1
0.0001
0.0034
0.8835
0.0001 0.0001 0.0001 0.0001
0.0042 0.0127 0.4435 0.0061
0.6506 0.7607 0.3940 0.6529
0.0001
0.3653
0.3225
RESPONSE VARIABLE: FLOOR DUST LEAD CONCENTRATION
Intercept
Soil Pb Round 1
Window Dust Pb Round 1
ALL GROUPS
BOS SPI
***** BOSPI-S
BOS P-S
Soil Pb ALL GROUPS
Cone. opko CIST
Rounds
BOS PI-S
BOS P-S
Window Dust Pb Cone.
Rounds
0.0001
0.2664
0.0001
0.0001
0.0001
0.3830
ROUND 1
0.0001 0.0001 0.0609 0.0001
0.2448 0.3570 0.412 0.3282
0.0001 0.0001 0.0001 0.0001
RESPONSE VARIABLE: FLOOR DUST LEAD
CONCENTRATION ROUND 3
0.0001 ' 0.0001 0.0001
0.0001
0.362
0.7345
0.0001 0.0001 0^0001
0.2240
0.8994
0.0243
0.5289 0.6532 0.0267 0.3282
0.0479
0.0440
0.0001
0.0001
0.0573
0.7002
0.2593
0.7696
0.0207
0.0415
B-24
-------
TABLE B-31. P-VALUES FOR TABLE 5-36. LONGITUDINAL STRUCTURAL
EQUATION MODELS REGRESSION COEFFICIENTS IN BOSTON STUDY
USING FIXED BLOOD LEAD PERSISTENCE FACTOR FOR FEMALES
REGRESSION COEFFICIENT
Predictor
Variable
ALL GROUPS
BOS SPI
***** BOS PI-S
BOS P-S
ALL GROUPS
Soil Pb BOS spl
RoUnd3 BOS PI-S
BOS P-S
ALL GROUPS
Floor Dust BQS spl
lb C°n°' BOS PI-S
ROUnd3 BOS P-S
Blood Pb Round 1
Intercept
Soil Pb Round 1
Dust Pb Round 1
MODEL MODEL
1 2
RESPONSE
0.1339 0.0463
0.1015
0.5711
0.3810
0.1018
0.0422 0.2234
RESPONSE
0.0001 0.0001
0.9857 0.9177
0.1127 0.1609
MODEL MODEL
10 11
MODEL
17
MODEL
30
VARIABLE: BLOOD LEAD ROUND 3
0.0713 0.8430
0.8050 0.8166
0.4793 0.0337
0.0549 0.0003
0.0427 0.0031
0.1088
0.0283
0.0016
0.4828
0.9175
VARIABLE: BLOOD LEAD ROUND 1
0.0001 0.0001
0.8133 0.3281
0.3534 0.0609
RESPONSE VARIABLE: FLOOR DUST LEAD
Intercept
Soil Pb Round 1
Window Dust Pb
Round 1
0.0001 0.0001
0.0586 0.1121
0.0034 0.0013
ROUND 1
0.0002 0.0003
0.1854 0.1529
0.0001 0.0006
RESPONSE VARIABLE: FLOOR DUST LEAD
ALL GROUPS
BOS SPI
Intercept BQS ^
BOS P-S
SoilPb ALL GROUPS
Cone.
Round 3 BOS SPI
BOS PI-S
BOS P-S
Window Dust Pb
Cone. Round 3
0.0001 0.0001
0.0086 0.0156
0.7834 0.7313
ROUND 3
0.0001
0.2191
0.6655
0.6745
0.0398
0.1871
0.9655
0.9279
0.6324 0.2873
0.0001
0.8552
0.2571
0.7959
0.4893
0.4179
0.7927
0.0029
0.0001
0.3401
0.0443
CONCENTRATION
0.0003
0.2059
0.0001
0.0003
0.1702
0.0009
CONCENTRATION
0.0001
0.0192
0.5848
0.2695
0.6518
0.6479
0.1515
0.9565
0.9338
0.3209
B-25
-------
TABLE B-32. P-VALUES FOR TABLE 5-38. LONGITUDINAL STRUCTURAL
EQUATION MODELS REGRESSION COEFFICIENTS IN BOSTON STUDY
USING FIXED BLOOD LEAD PERSISTENCE FACTOR FOR AGES 18-41 MONTHS
REGRESSION COEFFICIENT
Predictor
Variable
MODEL
1
MODEL
2
MODEL
10
MODEL
11
RESPONSE VARIABLE: BLOOD
Intercept
SoilPb
Round 3
Floor
DustPb
Cone.
Rounds
Blood Pb
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
Round 1
0.0074
0.2395
0.2386
0.0290
0.4125
0.2681
0.5112
0.2375
0.0935
0.2787
0.2091
0.0451
0.0452
0.6496
0.0460
0.1066
0.0033
0.0087
RESPONSE VARIABLE: BLOOD
Intercept
Soil Pb Round 1
Dust Pb Round 1
Intercept
Soil Pb Round 1
Window Dust Pb
Round 1
Intercept
ALL
GROUPS
BOS SPI
BOS PI-S
BOS P-S
0.0001
0.9807
0.7040
RESPONSE
0.0001
0.2118
0.0001
RESPONSE
0.0001
0.0001
0.8312
0.9488
VARIABLE:
0.0001
0.1550
0.0001
VARIABLE:
0.0001
0.0001
0.7984
0.7331
FLOOR
0.0001
0.4371
0.0001
FLOOR
0.0001
0.0001
0.6676
0.4729
DUST LEAD
0.0001
0.8325
0.0001
DUST LEAD
0.0001
0.6513
0.8737
MODEL
17
MODEL
30 UNITS
LEAD ROUND 3
0.6781
0.0079
, 0.0093
0.2337
0.0710
0.7693
0.0209
0.0166
0.0797
0.0442
LEAD ROUND 1
0.0001
0.7772
0.9239
0.0001
0.8262
0.7176
CONCENTRATION ROUND 1
0.0001
0.4744
0.0001
0.0001
0.7428
0.0001
CONCENTRATION ROUND 3
0.0001
0.0001
0.7169
0.9172
B-26
-------
TABLE B-32. P-VALUES FOR TABLE 5-38 (cont'd). LONGITUDINAL
STRUCTURAL EQUATION MODELS FOR: REGRESSION COEFFICIENTS
IN BOSTON STUDY USING FIXED BLOOD LEAD PERSISTENCE FACTOR
FOR AGES 18-41 MONTHS
REGRESSION COEFFICIENT
Predictor
Variable
ALL
Soil Pb GROUPS
Cone.
Round 3 BOS SPI
BOS PI-S
BOS P-S
Window Dust Pb
Cone. Round 3
MODEL MODEL MODEL MODEL
1 2 10 11
0.0001 0.0001 0.0001
0.0065
0.8239
0.6024
0.0036 0.0033 0.0106 0.0011
MODEL MODEL
17 30 UNITS
0.0001
0.0136
0.8635
0.6551
0.0127 0.0030
B-27
-------
TABLE B-33. P-VALUES FOR TABLE 5-40. LONGITUDINAL STRUCTURAL
EQUATION MODELS REGRESSION COEFFICIENTS IN CINCINNATI STUDY
USING FIXED BLOOD LEAD PERSISTENCE FACTOR
Predictor
Variable
REGRESSION COEFFICIENT
MODEL MODEL MODEL MODEL
1 5 6 J5
MODEL
J6 UNITS
RESPONSE VARIABLE: BLOOD LEAD ROUND 4
ALL GROUPS
GIN I-SE(D)
Intercept GIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Floor Dust Pb Round 4
Soil Pb Round 4
Blood Pb Round 1 (Fixed)
0.0050 0.0827 0.4888
0.8904
0.1178
0.0328
0.0469
0.9332
0.0038 0.0020 0.0001 0.0012
0.9681 0.6385 0.5356
0.2241
0.4976
0.3122
0.0304
0.0919
0.0011
RESPONSE VARIABLE: BLOOD LEAD ROUND 1
Intercept ALL GROUPS
CIN I-SE(D)
CIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Floor Dust Pb Round 1
Soil Pb Round 1
Intercept ALL GROUPS
CIN I-SE(D)
CIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Window Dust Pb
Round 1
Soil Pb Round 1
0.0001 0.0001 0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.481 0.0801 0.0230 0.0004
0.8027 0.0460 0.0677
RESPONSE VARIABLE: FLOOR DUST LEAD
ROUND 1
0.0001 0.0001 0.0001
0.0001
0.0659
0.0119
0.0001
0.0011
0.0002 0.0006 0.0003 0.0001
0.0025 0.0089 0.0040
0.0001
0.0001
0.0001
0.0001
0.0001
0.0023
CONCENTRATION
0.0002
0.1975
0.0223
0.0001
0.0029
0.0001
B-28
-------
TABLE B-33. P-VALUES FOR TABLE 5-40 (cont'd). LONGITUDINAL
STRUCTURAL EQUATION MODELS REGRESSION COEFFICIENTS
IN CINCINNATI STUDY USING FIXED BLOOD LEAD PERSISTENCE FACTOR
REGRESSION COEFFICIENT
Predictor
Variable
MODEL MODEL
1 5
MODEL MODEL MODEL
6 J5 J6 UNITS
RESPONSE VARIABLE: FLOOR DUST LEAD CONCENTRATION
ROUND 1
Intercept ALL GROUPS
CIN I-SE(D)
CIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Window Dust Pb
Round 1
Soil Pb Round 1
RESPONSE VARIABLE: FLOOR DUST LEAD CONCENTRATION
ROUND 4
Intercept ALL GROUPS
CIN I-SE(D)
CIN I-SE(F)
CIN NT(G)
CIN NT(M)
CIN SEI(P)
Window Dust Pb Round 1
Soil Pb Round 1
0.0001
0.0001
0.0001
0.0002
0.0818
0.0001
0.0001 0.0001
0.0107
0.0001 0.0001 0.0001
0.0002 0.0001 0.0009
0.0001 0.0001 0.0001
0.4535 0.4119 0.4972
0.0001 0.0001 0.0001
0.0001 0.0001 0.0001
ftU.S. GOVERNMENT PRINTING OFFICE: 1996^752-260/49044
B-29
-------
-------
-------
m
TJ
o
CO
CO
.CO
$• CD
< CO
">
CO
CD
------- |